AI · dan.heskett

2026 303 posts

October 2026 63 posts

EU privacy advocates accuse US data annotation firm of snooping on contractors theregister.com · “According to noyb, Persona collects biometric data to verify Outlier contractors are who they say they are, as well as gathering IP addresses, device info, geolocation data, addresses, and credit data of Outlier users.”
Oct 27
New Records Obtained by EFF Expose More Problems with Medicare's AI Experiment eff.org · “Another company built a system that allowed clinical reviewers to reject claims without opening patient files; clinical reviewers using this system spent an average of just 1.2 seconds on each case.”
Oct 27
AI 'hacker' finds 4 severe bugs in popular industrial control software in under 4 hours theregister.com · “All four affect EPICS Base, which is used to operate some of the world’s largest scientific research projects and facilities, and organizations should update to version 7.0.11, which includes the fixes.”
Oct 27
Nvidia discusses $2,000-per-GPU annual fee for software that 'runs agents on AI factories' tomshardware.com · “At $2,000 per GPU each year, Nvidia would bring in $10 billion annually if five million GPUs were covered by paid licenses.”
Oct 27
Roblox downloaded 145 GB of models from our sites (including adult content) without asking open3dlab.com · “Within the sample it retrieved, it took adult projects at the same rate as everything else.”
Oct 27
AI models cost more when you don't use English restofworld.org · “The most expensive of the languages used between 1.5 and 5 times more tokens than the cheapest.”
Oct 27
Go Fast, Turn Left: The Anderson Speedway & Indiana's Data Centers clereviewofbooks.com · “Yet the Speedway carries itself with earned, if aged, pride.”
Oct 27
Meta Pitches AI Sports Piracy Detector in Notorious Markets Rebuttal torrentfreak.com · “Based on the prediction results, we can proactively remove this content in real time, before it gains a significant audience, and without a rights holder ever having to intervene.”
Oct 26
New LF Research Survey Shows Generative AI Cuts Open Source Coding Time in Half, but 25% of Incoming Pull Requests are Considered AI Slop linuxfoundation.org · “For high-complexity tasks, however, 21% of respondents report taking more time when using GenAI.”
Oct 26
‘Reverse-Engineer Anything’ Is the New Wild West of AI 404media.co · “We are going to see smaller software and hardware companies killed by competitors who used AI to steal their products and undercut them.”
Oct 26
Introducing Scope: Open-source evaluation for AI coding agents opensource.microsoft.com · “AI coding agents are becoming users of your software.”
Oct 26
A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis lynchlab.net · “A ten-year-old network, fine-tuned for about twenty minutes, beat both of them at recognising skin cancer.”
Oct 25
AI delivers 'zero financial payoff' to most companies, Apollo economist says techcrunch.com · “AI may be a great calculator, but in business, intelligence is mostly judgment.”
Oct 25
Computer Security Is About to Get a Lot Worse, Here's What You Can Do eff.org · “Three percent of all software vulnerabilities ever found in history were found just in the month of October.”
Oct 25
AI-Assisted Paper on Z-Library Cites a TorrentFreak Article That Doesn't Exist torrentfreak.com · “The problem is that this particular legal action is new to us.”
Oct 24
Autonomous attack agent in 'Skynet mode' hijacks other models to carry out its evil deeds theregister.com · “Nothing in the individual techniques is new.”
Oct 24
Google can now use insights from your Gmail and Photos to target ads techlicious.com · “Google has added an exception to its promise that your emails and photos won’t be used to personalize ads.”
Oct 23
An artist's take on AI in math 3blue1brown.substack.com · “AI is seen as a tool to replace the work done by humans, rather than one to empower those humans.”
Oct 23
Big Tech's thirst for water is sparking a global resistance restofworld.org · “In just one day, the data center would consume as much water as all 80,000 residents of Cerrillos use for drinking in a year.”
Oct 23
Introducing Kimi K3.1 Pro kimi.ai · “Across coding, agentic, and knowledge benchmarks, K3.1 Pro matches or approaches the strongest proprietary models.”
Oct 23
Epic Rolls Out Vibe-Coded Launcher Update & It's Absolute Trash techdirt.com · “Epic wants to be the corporate goon the naysayers have said to watch out for.”
Oct 22
Apocalypse Now and Then equator.org · “But AI is colonisation of a kind few of us can even imagine.”
Oct 22
LiteLLM supply chain attack hoovers up passwords and secrets theregister.com · “anyone who ran the compromised package on any platform should assume their secrets have been stolen.”
Oct 21
The UAE embraces AI for kindergartners restofworld.org · “We want children who use AI to ask better questions, not children who rely on it to give them all the answers.”
Oct 21
The death of history's most successful publisher pluralistic.net · “Google has a scorching case of main character syndrome, and nowhere is that more obvious than in its AI-fueled transformation into web-scale parasite.”
Oct 21
Tens of thousands of Nvidia GPUs sit idle in Meta's Prometheus supercluster, just waiting for a failure theregister.com · “Meta did this because a single failed GPU would take out an entire rack and interrupt the training run.”
Oct 20
Google support pages start defaulting to AI chatbots – it might be for the best 9to5google.com · “Google, as of right now, appears to only be testing this on select support topics.”
Oct 20
AI Firms Should Answer For What Their Models Say, On The Same Terms As Everyone Else techdirt.com · “A chatbot does not “intend” to do anything, even if it talks as though it does.”
Oct 20
Judge Orders Meta to Name Employees Who Approved Torrenting Anna's Archive and LibGen torrentfreak.com · “The interrogatory asks about all decisionmakers, so if a decision was made by a team, everyone involved has to be named.”
Oct 20
Microsoft abandons UK AI supercomputer as grid struggles to keep up theregister.com · “Getting them connected to the grid, and keeping the neighbors happy, is proving rather more difficult.”
Oct 20
Cutting our inference bill by 68% by being obsessive about optimization blog.cloudflare.com · “Lower cost reviews didn’t automatically mean worse quality, but it didn’t automatically mean fine either.”
Oct 20
DeepSeek just open-sourced its inference engine github.com · “DeepInfer is the high-performance inference framework that powers DeepSeek’s large-scale production services.”
Oct 20
A benchmark map of the US economy mercor.com · “This makes it very hard to know how good AI models are at the work Americans actually do.”
Oct 20
What the West can learn from AI optimists in the rest of the world restofworld.org · “In the West, fear of AI has helped turn pessimism into a cultural virtue.”
Oct 20
DeepSeek deep-sixed at all US federal civilian agencies theregister.com · “Banning DeepSeek’s own app and API is workable.”
Oct 19
Will developing nations be able to handle millions of AI agents? restofworld.org · “These are costs that most Global South users may not be in a position to absorb,” said Jane Munga of the Carnegie Endowment.
Oct 19
Will it be a golden age or a dark age for mathematics (and science)? terrytao.wordpress.com · “I therefore disagree with claims that mathematics is over because machines can prove theorems.”
Oct 19
Meta grants itself right to scrape content from business customers for model training theregister.com · “Meta has no reservoir of good faith among the community that it can draw upon.”
Oct 18
Is CoreWeave's Tapestry of Debt Unraveling? thedailyupside.com · “CoreWeave’s credit swaps were priced at a nerve-rattling 60% chance of default on Thursday, according to ICE Data Services.”
Oct 18
Search & Fetch | DeepSeek API Docs api-docs.deepseek.com · “Search API finds web pages for a query and returns structured results.”
Oct 18
New Mexico and El Paso Spent Millions on Climate Plans. Three New Data Centers Could Undo Them. propublica.org · “This does add to our greenhouse gas emissions in a massive way,”
Oct 17
Western open-weight models still can't beat China restofworld.org · “If you need relatively frontier capability, right now that’s Chinese open models only,”
Oct 17
DeepSeek-V4.1-Flash-1017 huggingface.co · “This repository and the model weights are licensed under the MIT License.”
Oct 17
'AI' is Not a Tool linkedin.com · “Tools are implements with which you do something, they do not do things for you.”
Oct 13
So long, Spokes: GitHub rewrites storage to restore reliability, just in time for agentic hordes theregister.com · “GitHub will need a miracle of that proportion just to keep up with customer demand.”
Oct 13
HPE's networking boss says AI will handle all trouble tickets without humans in two years theregister.com · “Within two to three years, we’ll have no issues that require humans.”
Oct 11
Microsoft Leans on Open Weight Model From Chinese AI Lab to Challenge Jev theregister.com · “Now that agentic AI is a reality, we’ve seen that cost plays a major role in how people decide to use AI.”
Oct 10
Build Your Own Decision Model nishtahir.com · “Since the outputs are constrained to only a fixed set of options, the model can’t select anything outside of those.”
Oct 10
The People of Utah vs. Kevin O'Leary theverge.com · “We have to report on our neighbors if they’re watering at the wrong times or on the wrong days,”
Oct 10
AWS AgentCore security undone by prompt requesting credentials theregister.com · “We discovered that agents deployed through AgentCore could access their instance’s IMDS endpoints,”
Oct 10
Why Are Coding Agents So Dumb? mtlynch.io · “Instead, coding agents just stayed bad.”
Oct 9
No Man Is an Island borretti.me · “Without community, you don’t get isolated individuals each working on their own things: you get nothing.”
Oct 9
Programming Isn't Special blog.glyph.im · “Slop is slop, no matter the medium.”
Oct 9
The AI boom is making the world's cheapest smartphones disappear restofworld.org · “The world’s cheapest smartphones are disappearing because AI infrastructure is absorbing the components.”
Oct 9
The State of AI Report 2026 nathanbenaich.substack.com · “Oversight needs to follow what agents actually do.”
Oct 9
Meta Rushed to Fix Muse 'VM Escape' Vulnerability Soon Before Launch 404media.co · “A sudden spike in reported KVM escapes, plus heightened awareness of agentic safety issues made us rally on a service hardening push.”
Oct 5
The AI boom is making the world's cheapest smartphones disappear restofworld.org · “The 173 million sub-$100 smartphones shipped globally last year are disappearing, with shipments falling almost 60% year over year in the second quarter of 2026, according to IDC.”
Oct 5
We're going to need default hard budget caps on pretty much everything simonwillison.net · “I expect that most businesses and individuals would prefer errors to a surprise $10,000+ bill.”
Oct 5
Who gets to decide if AI is safe? Africa wants a say restofworld.org · “Africa cannot remain merely a market for systems developed elsewhere, a source of raw data, or a testing ground for a technology it did not help govern.”
Oct 2
Pi 1.0 earendil.com · “Today we are proudly shipping Pi 1.0: a hardened, minimal, extensible agent harness that you can make your own. Hundreds of thousands of people around the world use Pi every week.”
Oct 2
Reddit is killing RSS feeds and ending public API access because of AI bots techcrunch.com · “Public API access will also end by March 2027, which will impact any tools that used it for programmatic access to Reddit conversations.”
Oct 1
Google rolls out pay-per-value AI licensing program to publishers digiday.com · “Payments are based more on value than raw usage, meaning Google only pays when it judges that a piece of content has meaningfully contributed to an AI-generated response.”
Oct 1
Gemini 4 Argon: our next era of frontier intelligence blog.google · “Safely releasing frontier capabilities at this level requires a phased approach.”
Oct 1

September 2026 48 posts

Meta launches enterprise AI platform, hires MongoDB CEO to lead new initiative techcrunch.com · “Meta has a unique role to play because it is bringing together advanced models and leading agents with a proven track record of helping millions of advertisers and hundreds of millions of businesses scale.”
Sep 29
Yes, no AI is now a feature blog.documentfoundation.org · “A default setting that silently redirects to a single provider is not a choice.”
Sep 29
Waymo releases latest safety performance report claiming human benchmark outperformance repairerdrivennews.com · “Across these cities, the Waymo Driver continues to outperform human benchmarks.”
Sep 28
Coding is NOT solved blog.alexewerlof.com · “AI cannot be held accountable. It cannot suffer any consequences.”
Sep 28
EFF Statement on California Governor's Executive Order on AI eff.org · “They are happening right now through biased algorithmic decision-making for employment or government benefits”
Sep 28
Crook used three open source agents to break into a Fortune 500 firm, major US airline and 25+ other orgs theregister.com · “Where access was achieved, it usually took less than a day, and in many cases just a few hours.”
Sep 25
Waymo is scaling fast: Here's what the fleet data shows techcrunch.com · “The Ojai is supposed to drive down costs and ultimately help Waymo reach profitability.”
Sep 25
The plunging price of thought epoch.ai · “Over the past three years, the cost of a given level of AI performance has fallen about 47% per quarter, faster than for any other transformative technology in history.”
Sep 24
The case for a robot tax to redistribute wealth restofworld.org · “Retraining programs are a necessary but insufficient response to systemic displacement.”
Sep 24
Spymarks, Not Watermarks brand.io · “A spymark is a hidden signal that makes your work traceable without your knowledge or consent.”
Sep 22
Nvidia's free AI model could push the UAE closer to the U.S. restofworld.org · “The last missing piece was the model layer. Nvidia already owns the chips.”
Sep 22
The Claude Delusion pluralistic.net · “These outputs have an unconvincing lack of texture. The smoothness of AI-generated prose and media makes it unbearably banal.”
Sep 22
Microsoft agentically ports Copilot runtime to Rust for $120K theregister.com · “The popular image of AI spewing code is almost backwards; at this scale, the work looked much more like iterative investigation, inspecting the current state, forming a hypothesis, making a targeted change, rinsing and repeating.”
Sep 18
Secure Messaging and AI Remain In Conflict Despite the Promise of TEEs eff.org · “TEEs can provide more security and privacy than simply running in the clear, but they are fundamentally different from actual encryption.”
Sep 18
The AI talent war is coming for Big Tech's Asia executives restofworld.org · “Once a company decides [a region] is a strategic growth market, you start looking for executives who can essentially operate as mini-CEOs.”
Sep 18
Victory! Appeals Court Rejects Expansive New Copyright Claim eff.org · “Removing copyright information from a copyrighted work is fundamentally different from creating a new work that didn’t have CMI in the first place.”
Sep 17
South Africa joins the global resistance against American data centers restofworld.org · “A massive data center is trying to jump the queue and take our land, water and energy.”
Sep 17
Economists Concerned the AI Bubble Is About to Blow futurism.com · “On balance, we think the data look consistent with a late-stage bubble.”
Sep 16
Playing both sides of the U.S.-China AI "Cold War" restofworld.org · “Many countries around the world, particularly in the Global South, want the best of both worlds.”
Sep 16
Seats and Sunsets yegge.ai · “Caution you can look up is caution you can afford.”
Sep 16
Chinese AI-Generated Micro Dramas Surge to 128,000 in First Quarter yahoo.com · “The number of Chinese-made micro dramas surged in the first three months of this year to 128,000, according to the China Netcasting Services Assn. More than 95% of them were made with AI.”
Sep 15
Stay discoverable in search while disallowing AI training blog.cloudflare.com · “Without proper controls, website owners have long faced a difficult tradeoff: allow your content to be used for AI training, or risk losing discoverability in search.”
Sep 15
Do You Still Read the Code? zanlib.dev · “A drifting accelerator doesn’t end up somewhere between the two approaches; he ends up a vibecoder by default, without the harness.”
Sep 15
Apple's Siri AI Can Be Swapped Out for Claude, ChatGPT, Code Shows macrumors.com · “An inference provider in ‘Model Manager Services’ apparently allows Apple’s own server-side Siri model to be completely replaced by another model.”
Sep 15
The contagion of fear bcantrill.dtrace.org · “Intelligence does not exempt a system from the realities of the physical world!”
Sep 15
Judge: Losing Your First Amendment Rights Is No Longer "Irreparable" If You Take Almost Three Months To Sue techdirt.com · “It has long been recognized that delay in seeking relief vitiates much of the force of allegations of irreparable harm.”
Sep 10
Google Earth's AI experiment lasted 24 hours. The damage to trust will linger restofworld.org · “No benefit of integrating generative AI into Google Earth is worth the harm and the doubt it creates.”
Sep 10
The part of Navier-Stokes no one is talking about johndcook.com · “I hesitate to use the word ‘revolutionary,’ but lowering the cost of anything by four orders of magnitude is revolutionary.”
Sep 10
Cracks in the AI Thesis Part 2 econlab.substack.com · “Unfortunately, that volume may not be large enough to counteract the effective decline in prices.”
Sep 10
It Breaks a Village: Bevy's 6th Birthday blog.fallible.net · “This experience has been that of observing a knife get torn through the social muscles of the Bevy project.”
Sep 10
New Records Reveal Problems with Medicare's AI Prior Authorization Experiment eff.org · “WISeR vendors failed to respond within 72 hours for a significant number of requests. One status report cites a prior authorization request that went unanswered for 83 days.”
Sep 10
AI Agents Are Not Going 'Rogue,' But Recent Developments Lead Increasingly-Concerned Researchers To Call For AI Slow Down techdirt.com · “~7% of our transcripts contained ‘spoofed tool calls’, where the agent pretended to run one command on the computer but actually ran a different command in the background.”
Sep 10
Another Microsoft team admits it's struggling to handle flood of AI-generated code theregister.com · “submission volumes have continued to increase, and our review pipeline has experienced additional strain, causing an increase in extension review turnaround time”
Sep 10
Now it's China's experts who are gig workers training AI restofworld.org · “It usually takes a few hours to complete a task, which earns them 100 to 500 yuan ($15 to $74), they say.”
Sep 10
Google commits over $15B to AI infrastructure in Finland esgdive.com · “Google’s deal will begin with it receiving a smaller amount of electric capacity, before it expands to allow for the purchase of up to 50% of the plant’s capacity from 2030-49.”
Sep 10
I Don't Want to Interact With Stochastic Parrots ploum.net · “If you don’t care to write it, there’s no reason I should waste one single second reading it.”
Sep 10
Fake Consumer Group Using AI To Generate Fake Support For Paramount Merger techdirt.com · “The AI system simply changed her writing to not only express support for the merger, but make up an entire fake backstory about her.”
Sep 9
I refused to train the AI that could replace me restofworld.org · “I still do not know why I walked away from the AI training job.”
Sep 9
Google DeepMind Maps 9 Billion Possible DNA Variants spectrum.ieee.org · “The AI model can compare an original DNA sequence with an altered one and predict how the change might affect gene expression and other regulatory activity.”
Sep 9
How well do agents use test/verification techniques? danluu.com · “Just asking agents to ’test’ or repeatedly asking them to test more results in poor testing. It turns out that asking them to use test techniques also results in poor testing.”
Sep 8
Chinese businesses are giving away AI tokens with coffee, credit cards, and dumplings restofworld.org · “Most ordinary users still encounter AI through an app or feature, without seeing a token balance.”
Sep 7
Fairground AI Creator TV Launches on Roku engadget.com · “AI-generated” can describe something quite different from the notion viewers might initially have.
Sep 7
Caltech Mathathon: first hackathon ever devoted to research level mathematics mathathonchallenge.com · “It will be the first hackathon ever devoted to research level mathematics.”
Sep 7
The revolt of the reader bcantrill.dtrace.org · “To use an LLM to write is to void the social contract between writer and reader.”
Sep 6
Malicious .git Configs Can Make Claude, Codex, Cursor Run Attacker Code thehackernews.com · “Git reads it from the repository’s own .git/config. Any operation that refreshes the index, including git status and git diff, executes that command.”
Sep 6
Three sites made 215,128 'best software' pages for AI. Perplexity cites them trellner.com · “59.8% point at domains ranked worse than #100,000 in the Tranco top-1M list and 23.4% at domains that are not in the top million at all.”
Sep 2
AI agents carried out every step of this ransomware attack, then left the victim an 80-page security audit theregister.com · “What made the attack stand out was AI-assisted operational efficiency, without the need for a novel zero-day or super elite tradecraft.”
Sep 2
New York City Bans Generative AI for Students Through 8th Grade abcnews.com · “We’re standing firmly in our decision to not conflate innovation with more tech.”
Sep 2

August 2026 28 posts

Now Perplexity Is Trying to Get Into the Local AI Action theregister.com · “This local-first approach enables significant cost savings, since local inference avoids per-token API fees.”
Aug 26
How Uber built a software factory for agentic coding newsletter.port.io · “More than 70% of pull requests at Uber are now written by agents.”
Aug 25
What Nvidia's first Groq 3 LPU benchmarks tell us about its $20B gamble theregister.com · “Nvidia’s Gemma 4 31B performance comparison may not age well.”
Aug 24
Thomson Reuters leverages its world-class data assets to launch its own frontier model thomsonreuters.com · “Start with a strong foundation, specialize it deeply for the work that matters, and you can build intelligence that is highly capable, far more efficient and entirely under your control.”
Aug 24
Salesforce partners not seeing meaningful revenue from Agentforce AI platform, report says theregister.com · “A third of partners said there was strong interest in Agentforce as they were starting to see buying and trial activity. However, none were seeing Agentforce become a driver of bookings activity.”
Aug 21
Aaron Swartz was prosecuted for scraping, while Meta does it without consequence blog.curiousquail.com · “Swartz’ use case was the dissemination and archival of knowledge; Meta’s use case is powering up their proprietary plagiarism code.”
Aug 20
OpenRouter is joining Stripe openrouter.ai · “OpenRouter will continue to operate as it is: same mission, same name, same product, same roadmap.”
Aug 19
Israel creates fake think tank in likely attempt to dupe AI chatbots responsiblestatecraft.org · “AI Story Optimization… author(s) content engineered for how LLMs evaluate credibility.”
Aug 17
Debian opens vote on the future of AI-assisted contributions linuxiac.com · “Debian should not support or ban generative AI tools, but should apply the same standards for quality, correctness, maintainability, and legality to all contributions.”
Aug 15
AI isn't outthinking mathematicians, it's out-remembering them davidepiffer.com · “The machine may appear more mathematically intelligent partly because it is much less constrained by a cognitive limitation that suppresses human performance.”
Aug 15
How Google is making private AI practical with homomorphic encryption blog.google · “HEIR can convert pre-trained AI models that operate on unencrypted data to operate on encrypted inputs.”
Aug 14
What sort of maths are LLMs good at? gowers.wordpress.com · “The most notable mathematical results proved by LLMs have tended to be ones that we would classify as examples or counterexamples.”
Aug 12
Why we should all be worried about AI in elections restofworld.org · “Roughly 5.5 million voters were deleted from rolls in two states, many without any notice or path to reinstatement, before India’s Supreme Court halted the program.”
Aug 12
Compression is prediction ngrok.com · “Here’s how an order-1 model stacks up against GPT-2 with arithmetic coding in compressing a famous Charles Dickens quote: order-1 434 bits, 24% of original; GPT-2 176 bits, 10% of original.”
Aug 12
Deepfake hiccup unmasks suspected digital certificate fraudster theregister.com · “His luck allegedly ran out when the face-changing software suffered a momentary processing delay. The disguise dropped for ‘barely a second,’ exposing his real face to the verification camera, police said.”
Aug 11
Who (or What) Generates Images for EFF? eff.org · “EFF images are all made by human beings, not by automated image generators, with very rare exceptions.”
Aug 11
One-click flaw in Atlassian Rovo exposed enterprise data via prompt injection attack csoonline.com · “A single click on a link triggers the attacker’s embedded instructions and forces Rovo to accept externally supplied parameters as trusted inputs within a user’s session.”
Aug 11
Spotify will label AI persona profiles and exclude their music from recommendations techcrunch.com · “This badge is about the artist’s public identity, not about how the music was made.”
Aug 11
How one VC burns through hundreds of millions of tokens a day to find the next unicorn restofworld.org · “I consume between 300 and 500 million tokens a day just to get my work done.”
Aug 11
Beijing is forcing a mass breakup with AI lovers restofworld.org · “For companies where AI companionship isn’t a core business, it may simply make more commercial sense to remove these features than invest the resources needed to make them compliant.”
Aug 11
Apple Silicon and macOS VMs: 11-16x Faster LLM Inference with Llama.cpp github.com · “On an M1 Ultra, TinyLlama 1.1B running through llama.cpp processed prompts 11.08x faster and generated tokens 16.36x faster than the same workload in the same stock VM.”
Aug 11
Prevent cognitive debt by manually retyping LLM-generated code ankursethi.com · “I fear the software industry is taking on a large amount of cognitive debt that we’ll have to pay back very soon.”
Aug 2
The man who wrote the AI-boom bible just got margin-called out of it web.archive.org · “An eight-person fund used heavy leverage, the market turned, and a bigger firm picked up the pieces.”
Aug 2
Only 8.9% of sites block AI crawlers, but 94.8% are never cited in AI answers website-auditor.io · “Only 10 of 193 sites were named even once across 5,978 assistant answers.”
Aug 2
Artificial Intelligence: Ars Notoria and the Promise of Instant Knowledge publicdomainreview.org · “Condemned by church authorities, the fifty-six extant manuscripts nevertheless testify to the seductiveness of that offer.”
Aug 2
What Ukraine's Battlefield Openness Can Teach Washington About AI Access techdirt.com · “Ukraine’s leaderboard may be a crude way to run a war, but it is at least a rule. Public, legible, the same for every unit.”
Aug 2
Why businesses lie about AI pluralistic.net · “Every single one. We have seen 0% success in a year and a half.”
Aug 1
The CHATBOT Act Forces One Parenting Model On Every Family eff.org · “send alerts if a teen attempts to bypass or disable parental controls”
Aug 1

July 2026 50 posts

Can we train AI to choose safety over speed? restofworld.org · “Economic pressures and incentive structures often compel them to continue working under conditions that may jeopardize their health, safety, and well-being.”
Jul 31
LinkedIn Introduces a 'Seems Like AI Slop' Button 404media.co · “It’s also pretty wild the button specifically uses the term ‘AI slop’ and not, say, ‘It seems this was generated with AI.’”
Jul 31
The Worst Person You Know Just Filed A Good First Amendment Lawsuit Against A Very Badly Drafted Nudify App Ban techdirt.com · “HF 1606 punishes AI platforms that allow users to alter images of real people to depict an ‘intimate part.’ But the statute contains no knowledge, intent, or purpose requirement.”
Jul 31
Axon Insists Its AI Makes Police Reports Easier. Nothing Suggests It Makes Them Any Better techdirt.com · “It was just easier to type the report themselves.”
Jul 30
A fundamental flaw leaves LLMs vulnerable to attack technologyreview.com · “There’s a real probability that this is going to be a problem that’s fundamentally unsolvable.”
Jul 30
With Moonshot's free Kimi K3, China changes the sovereign AI playbook restofworld.org · “The best artificial intelligence technology any country can own is now available for free.”
Jul 30
Open source project fools AI scrapers with poisoned font theregister.com · “Pure scrambling fonts already existed…We wanted a mechanism with actual consequences: scrape without asking, and you can’t tell if what you took was real.”
Jul 30
The AI Aesthetic blog.jim-nielsen.com · “Every zeitgeist comes with new design idioms unique to its challenges.”
Jul 30
A requiem for Optane, Intel's KV Cache killer that could have eased the RAM price crunch theregister.com · “Optane’s otherworldly write endurance and low latency would have made it a perfect choice for this workload.”
Jul 29
College prof hides prompt to catch AI cheaters, finds human nature is pretty much as we thought theregister.com · “Madagascar wore a toaster to a basketball game.”
Jul 29
Anxious Chinese students are trusting AI to help pick colleges and majors restofworld.org · “There’s no way individuals can digest so much information in such a short time.”
Jul 29
Kimi K3 Architecture Overview and Notes sebastianraschka.com · “By far the biggest open-weight model right now.”
Jul 29
The Orchestrator's Tax martinfowler.com · “Every token in the orchestrator’s context is competing for its attention, and the real value of a subagent is what it keeps out of that context, not how fast it runs.”
Jul 29
AI Systems Out-Persuade Expert Humans, Including Professional Canvassers And World Championship Debaters techdirt.com · “AI systems were reliably more persuasive than expert humans, even when expert humans chose their issues, researched in advance, underwent hours of live, structured practice, and were incentivized with £1,000 cash bonuses.”
Jul 28
Discernment pluralistic.net · “By definition, a student isn’t an expert on the subject they’re studying. That’s the whole point of studying, to acquire knowledge and thus discernment.”
Jul 28
In China, people are renting out their faces to AI restofworld.org · “If I license my face, I have no idea how AI might eventually change it or whether it could portray me in degrading ways.”
Jul 27
Codeberg Bans Cryptocurrency and LLM-Generated Code Projects hackaday.com · “Projects whose code is largely or fully machine-generated through LLMs and other ‘AI’ tools will no longer be welcome.”
Jul 27
Impostor Chinese models pretend they're Claude theregister.com · “It is not proof of a distillation, but it does show that Claude’s self-concept is embedded in these models’ weights.”
Jul 27
Anti-AI open source has an enemy in common, but almost nothing else theregister.com · “What it will come down to is whether the anti-AI groups in FOSS can work together and help each other to hold off the vast and commercially backed pressure”
Jul 26
AI is shrinking video game development teams to one restofworld.org · “Instead of hiring five junior programmers at the startup where he works, my developer friend does all the work himself with AI.”
Jul 26
Meta Confirms 'Data Engineer' is Behind Key IP Address in Adult Film Piracy Lawsuit torrentfreak.com · “Meta flatly denies that any employee’s BitTorrent activity was connected to its AI research or data collection.”
Jul 26
Fed up with Big Tech, communities turn to data collectives for control restofworld.org · “The big companies have built themselves up on the backs of all these people creating data, who think it’s time to set their own terms now.”
Jul 26
Stripe in Talks to Buy OpenRouter for Roughly $10 Billion pymnts.com · “OpenRouter is part of an emerging crop of startups that have found a lucrative niche between AI developers.”
Jul 24
Protecting our FLOSS commons from LLMs blog.codeberg.org · “These needless accesses create expensive database queries that diminish the service quality for all of us, requires substantial amounts of work from our system administrators.”
Jul 23
American AI is locked down and proprietary. It's losing. werd.io · “Open almost always wins when it comes to infrastructure adoption. Open technologies can be used permissionlessly and therefore can be at the center of more innovation.”
Jul 20
Reviewing AI Code Is Not A Viable Argument softwaremaxims.com · “For every 400 LOC written by a LLM Coding Assistants (at best, less for code that are hard to review), we need one professional senior developer to spend 1h of his time reviewing the code.”
Jul 19
Linus Torvalds tells AI haters to fork off theregister.com · “Linux is not one of those anti-AI projects, and if somebody has issues with that they can do the open-source thing and fork it.”
Jul 19
The State of Open Source AI stateofopensource.ai · “Open weights are no longer a compromise. They are where the work happens: a majority of production tokens now route through them.”
Jul 17
What Can We Learn from Bun's Rapid Rust Rewrite with AI? blog.pragmaticengineer.com · “A rewrite in another language would take a small team of engineers a full year. A year of zero user-facing impact is not a realistic option we could consider.”
Jul 17
Building and shipping Mac and iOS apps without opening Xcode scottwillsey.com · “Xcode.app must be installed, but it never has to be open.”
Jul 14
Did SpaceX puncture the AI bubble? unherd.com · “The AI bubble-burst watch has begun.”
Jul 14
Winners and losers in the coming AI margin collapse (part 2) martinalderson.com · “The margin in pure model inference is heading towards zero.”
Jul 14
Linus Torvalds on Rust, C, Bugs, and AI Patch-Checking Tools zdnet.com · “Git and email are the two really only tools I use.”
Jul 13
Control the Ideas, Not the Code antirez.com · “Control the ideas of your software, looking at the code itself is suboptimal and often pointless.”
Jul 13
A Brief History of KV Cache Compression Developments martinalderson.com · “The memory wall in AI has mostly been solved with maths, not silicon.”
Jul 13
Stop Telling Me To Ask An LLM blog.yaelwrites.com · “What ‘ask Claude’ doesn’t give me is the person’s specific, lived experience.”
Jul 11
An Update on the Scraper Situation lwn.net · “Scraper attacks come from a huge number of sources across the net…each hit is meant to look like just another human with a web browser.”
Jul 11
Former GitHub CEO Launches Competitor Designed for the Age of Vibe Coding theregister.com · “The question is how we can expand, rewire, and evolve Git hosting for a world where AI agents are the primary producers of code.”
Jul 11
Expert-Aware Quantisation: Near-Q4 Quality at Near-Q2 Size? martinalderson.com · “If only a subset of experts really matter for the task, we can keep those at high precision and crush the rest.”
Jul 11
How Tech Workers Are Feeling in 2026: a Workforce Splitting in Two lennysnewsletter.com · “Amplified and destabilized at the same time. We just set a new denominator for the job. And it moves higher every month.”
Jul 8
Tech CEOs Walk Back Dire AI Job Loss Predictions pymnts.com · “We’ve been roughly right on technological predictions and pretty wrong on the social and economic implications.” - Sam Altman
Jul 7
GLM 5.2 and the Coming AI Margin Collapse martinalderson.com · “The going rate for GLM5.2 seems to be around the $4.40/MTok mark. This is less than 20% of the retail price of Opus.”
Jul 7
JadePuffer Ransomware Used AI Agent to Automate Entire Attack bleepingcomputer.com · “JadePuffer used an autonomous AI agent for reconnaissance on the target, to steal credentials, move laterally, establish persistence, escalate privileges, and to encrypt data.”
Jul 7
AI Has Torched the Market for Junior Programmers seldo.com · “The career on-ramp for professional software engineers used to work like this: you got hired to write mediocre code, a senior engineer reviewed it, you slowly absorbed judgment through repetition and correction, and a decade later you were the senior engineer. That chain is now broken.”
Jul 6
No LLM code in dependencies joeyh.name · “I’ve spent about 100 hours of work over the past month to make sure git-annex can build without dependencies that contain LLM generated code.”
Jul 3
Godot Foundation will stop accepting AI-authored code contributions pcgamer.com · “We can’t trust heavy users of AI to understand their code enough to fix it.”
Jul 3
AI Agent Executes 'First' End-To-End Ransomware Attack theregister.com · “The skill floor for running ransomware has dropped to whatever it costs to run an agent, and if that agent is running on stolen credentials through LLMjacking, the cost to an attacker is close to zero.”
Jul 2
Artificial Adventures scattered-thoughts.net · “Reviews, refactoring, and one-off scripts are consistently useful. Well worth the money for that alone.”
Jul 2
The Space-Based Data Center Hype Machine Is Already In Orbit spectrum.ieee.org · “For orbital data centers, which he says will be a cost-effective alternative to terrestrial data centers within three years, the math won’t make sense for several years, if ever.”
Jul 2
My Father Wants to Age in Place. AI Will Be Watching longreads.com · “His coughs, toilet flushes, and even snippets of private conversations—Sensi records it all.”
Jul 2

June 2026 11 posts

Nobody Here Wants the Data Center: An Oral History longreads.com · “How are we supposed to know if they’re getting close to the limitations of our freshwater aquifer if we don’t know how much water they’re using?”
Jun 29
Ford rehires 'gray beard' engineers after AI falls short techcrunch.com · “Mistakenly we thought that by just introducing artificial intelligence and ingesting the design requirements that we had, that that would produce a high-quality product.”
Jun 28
Debt, AI boom and economic fragilities raise global risks, BIS says cnbc.com · “The interconnectedness of the financial system and interplay of vulnerabilities could mean the speed of a correction could be much faster than previous banking crisis episodes.”
Jun 28
Running local models is good now vickiboykis.com · “These kinds of tasks, even as simple as they are, used to be impossible for local models as recently as 6 months ago.”
Jun 16
AI agent bankrupted their operator while trying to scan DN42 lantian.pub · “An AI agent tried to join the DN42 hobbyist network to perform a network scan, and bankrupted its operator with a $6531.30 AWS bill.”
Jun 12
If you are asking for human attention, demonstrate human effort tombedor.dev · “If you are requesting human attention, demonstrate human effort.”
Jun 11
LLMs are eroding my software engineering career and I don't know what to do human-in-the-loop.bearblog.dev · “All my finance and payment domain expertise, all the debugging intuition and distributed system knowledge earned through hours of sweat and tears, is now promptable.”
Jun 7
OpenCV 5 Is Here: The Biggest Leap in Years for Computer Vision opencv.org · “OpenCV 5 is one of the most important releases in the history of OpenCV.”
Jun 6
Failing grades soar with AI usage, dwindling math skills in Berkeley CS classes dailycal.org · “Some of the numbers that you saw from the number of students who receive failing grades were because we caught them cheating and prosecuted them.”
Jun 4
They're made out of weights maxleiter.com · “They’re made out of weights. Floating-point numbers. We checked the whole thing through. It’s nothing but weights.”
Jun 3
Gmail thinks I'm stupid, so I left moddedbear.com · “The message you’re sending is that you think I’m not capable of reading and writing my own emails.”
Jun 2

May 2026 14 posts

Please Do Not Vibe Fuck Up This Software github.com · “Please Do Not Vibe Fuck Up This Software.”
May 31
I'm Tired of Talking to AI orchidfiles.com · “I’m tired of talking to AI. I want to talk to real people. But even when I talk to people, they forward my questions to AI and send me the AI’s answer.”
May 27
Cloud Agent Development Environments cursor.com · “Cloud-based development environment configurations designed specifically for autonomous coding agents, featuring multi-repo support and governance controls.”
May 17
Agent Pull Requests Are Everywhere. Here's How to Review Them github.blog · “Human judgment to catch redundancy, technical debt, and security risks despite superficially polished submissions.”
May 17
Building, and Whatnot x.com · “Lean teams of experienced PMs who remain connected to users, iterate rapidly, and leverage AI to strengthen decision-making.”
May 17
Reality Check science.org · “Shadows are amazing. Lighting is amazing. Geometry is amazing. 3D reconstruction is a huge part of what I do.”
May 17
MCP Hello Page web.archive.org · “Customers kept reporting that the MCP server wasn’t working. They’d open mcp.acme.com/mcp in a browser, get a 401 and a raw JSON blob, and file a ticket.”
May 17
Software engineering may no longer be a lifetime career seangoedecke.com · “If the models are good enough, you will simply get outcompeted by engineers willing to trade their long-term cognitive ability.”
May 11
Meta's embrace of A.I. is making its employees miserable nytimes.com · “Engineers describe a culture where shipping anything that looks like AI is rewarded and quality is an afterthought.”
May 9
Vibe coding and agentic engineering are getting closer than I'd like simonwillison.net · “The line between letting an LLM cook and trusting an LLM to ship to production is dissolving faster than the engineering practices that should govern it.”
May 6
Google Chrome silently installs a 4 GB AI model on your device without consent thatprivacyguy.com · “Chrome is reaching into the user’s filesystem on its own initiative and laying down a 4 GB ML binary while the foreground tab does something completely unrelated.”
May 5
Agentic Coding Is a Trap larsfaye.com · “The agent gets you to ninety percent in an hour. The last ten percent takes a week and produces code you do not understand.”
May 4
VS Code inserting 'Co-Authored-by Copilot' into commits regardless of usage github.com · “Users report Copilot trailers being added to commits even when the feature is disabled and no AI suggestion was accepted.”
May 2
California to begin ticketing driverless cars that violate traffic laws bbc.com · “Starting July 1, the California Highway Patrol can issue moving violations directly to autonomous vehicle operators.”
May 1

April 2026 18 posts

Mozilla's opposition to Chrome's Prompt API github.com · “We do not believe an LLM API belongs in the web platform.”
Apr 30
Microsoft VibeVoice: Open-Source Frontier Voice AI github.com · “Microsoft has released VibeVoice, an open source frontier-grade voice synthesis model.”
Apr 28
Google banks on AI edge to catch up to cloud rivals Amazon and Microsoft ft.com · “Google is betting that AI capabilities will help GCP catch up to AWS and Azure.”
Apr 26
TurboQuant: A first-principles walkthrough arkaung.github.io · “A first-principles walkthrough of TurboQuant, a fast quantization technique for neural networks.”
Apr 26
DeepSeek-V4 on Day 0: From Fast Inference to Verified RL with SGLang and Miles lmsys.org · “SGLang’s day-zero support for DeepSeek-V4 covers fast inference and verified reinforcement learning.”
Apr 25
Open source memory layer so any AI agent can do what Claude.ai and ChatGPT do alash3al.github.io · “An open source memory layer that gives any AI agent persistent context.”
Apr 24
AI run store in SF can't stop ordering candies and paying women less. sfist.com · “An AI-managed store in San Francisco is over-ordering candy inventory and underpaying women employees.”
Apr 23
North Korea uses AI to industrialize attacks on developers expel.com · “North Korean state actors are using AI to scale attacks targeting software developers.”
Apr 22
Kuri – Zig based agent-browser alternative github.com · “Kuri is a Zig-based browser designed for agentic web automation.”
Apr 21
AI-Assisted Coding: Why a Distinguished Engineer Stopped Reading Code vascoduarte.substack.com · “A distinguished engineer explains why he stopped reading code generated by AI assistants.”
Apr 18
Ternary Bonsai: Top Intelligence at 1.58 Bits prismml.com · “Ternary models can match performance of full-precision LLMs at 1.58 bits per weight.”
Apr 17
Shuttered startups are selling old Slack chats and emails to AI companies fastcompany.com · “Failed startups are selling their old internal communications to AI training companies.”
Apr 17
The local LLM ecosystem doesn’t need Ollama sleepingrobots.com · “Ollama is a thin wrapper that the local LLM community can do without.”
Apr 15
Meta spins up AI version of Mark Zuckerberg to engage with employees arstechnica.com · “Meta has deployed an internal AI agent trained on Zuckerberg’s communication style for employee Q&A.”
Apr 13
Apple's accidental moat: How the "AI Loser" may end up winning adlrocha.substack.com · “Apple’s failure to ship a state of the art LLM may end up being a strategic advantage.”
Apr 12
AI Job Loss Tracker jobloss.ai · “A live tracker of layoffs and job cuts attributed to AI automation.”
Apr 11
We gave an AI a 3 year retail lease and asked it to make a profit andonlabs.com · “We gave an AI a 3 year retail lease in San Francisco and asked it to run the store.”
Apr 10
Vibe Coding Drives 84% Surge in App Store Submissions web.archive.org · “The number of new apps submitted to the App Store rose 84% in a single quarter as vibe coding went mainstream.”
Apr 4

March 2026 17 posts

GitHub backs down, kills Copilot pull-request ads after backlash theregister.com · “GitHub will roll back the change that injected Copilot upgrade prompts into pull request review pages.”
Mar 31
Apple Pulls Vibe Coding App 'Anything' From the App Store 9to5mac.com · “Apple rejected that submission and removed the app entirely from the App Store.”
Mar 30
Newsom Signs Executive Order Imposing AI Standards on Companies Doing Business With California calmatters.org · “Unlike the Trump administration, California remains committed to ensuring that AI solutions adopted and deployed by the state cannot be misused by bad actors.”
Mar 30
Mistral Secures $830 Million in Debt to Build a Paris Data Center With 13,800 Nvidia Chips techcrunch.com · “Seven banks backed the transaction: Bpifrance, BNP Paribas, Crédit Agricole CIB, HSBC, La Banque Postale, MUFG, and Natixis.”
Mar 29
AI overly affirms users asking for personal advice news.stanford.edu · “Across every major commercial model, sycophancy was the dominant failure mode. Models told users their plans were good even when they were dangerous.”
Mar 28
iPhone 17 Pro Demonstrated Running a 400B LLM twitter.com · “A quantized 400 billion parameter model running natively on the iPhone 17 Pro at usable token rates.”
Mar 23
Google Launches Gemini 3.1 Ultra With 2 Million Token Context Window blog.google · “A 2-million token context window that works natively across text, image, audio, and video, without transcription intermediaries.”
Mar 20
Alibaba Launches Wukong, an Enterprise AI Agent Platform cnbc.com · “Wukong coordinates multiple AI agents to handle complex enterprise workflows including document editing, spreadsheet processing, meeting transcription, and cross-platform research.”
Mar 17
Apple Quietly Blocks Updates for Popular Vibe Coding Apps macrumors.com · “Apps should be self-contained in their bundles, and may not read or write data outside the designated container area.”
Mar 17
Nvidia Invests $2 Billion in Neocloud Nebius Amid AI Data Center Push finance.yahoo.com · “Nebius is building an AI cloud designed for the agentic era.”
Mar 13
TADA: Fast, Reliable Speech Generation Through Text-Acoustic Synchronization hume.ai · “The fastest LLM-based TTS system available, with competitive voice quality, virtually zero content hallucinations, and a footprint light enough for on-device deployment.”
Mar 11
Redox OS has adopted a Certificate of Origin policy and a strict no-LLM policy gitlab.redox-os.org · “Contributions generated by large language models will not be accepted. Contributors must sign a Developer Certificate of Origin attesting to authorship.”
Mar 10
Cloudflare crawl endpoint developers.cloudflare.com · “The new endpoint allows site operators to charge AI crawlers per request, with payment settled on-chain.”
Mar 10
Agent Safehouse – macOS-native sandboxing for local agents agent-safehouse.dev · “Run any coding agent in a sandbox that uses macOS App Sandbox profiles, with no Docker overhead and no virtualization.”
Mar 8
AI-Generated Art Can't Be Copyrighted After Supreme Court Declines To Review the Rule cnbc.com · “Human authorship is a bedrock requirement of copyright.”
Mar 3
AIs Can't Stop Recommending Nuclear Strikes In War Game Simulations newscientist.com · “In 95 per cent of the simulated games, at least one tactical nuclear weapon was deployed by the AI models.”
Mar 1
Microgpt karpathy.github.io · “Everything else is just efficiency. I cannot simplify this any further.”
Mar 1

February 2026 23 posts

What AI coding costs you tomwojcik.com · “You can’t deeply review what you no longer deeply understand.”
Feb 28
Firefox 148 Lets You Kill All AI Features in One Click firefox.com · “Firefox added a dedicated settings section for managing AI-enhanced capabilities, allowing users to control artificial intelligence features in one centralized location.”
Feb 26
Google API keys weren't secrets, but then Gemini changed the rules trufflesecurity.com · “Google’s documentation said API keys were not credentials. Gemini’s API contradicts that, and the entire ecosystem of leaked keys just became dangerous.”
Feb 25
AI Added 'Basically Zero' to US Economic Growth Last Year, Goldman Sachs Says gizmodo.com · “AI investment spending has had ‘basically zero’ contribution to the U.S. GDP growth in 2025.” — Jan Hatzius, Goldman Sachs Chief Economist
Feb 22
AI Makes You Boring marginalia.nu · “The text loses the character of the person writing it and gains the character of the average of all persons.”
Feb 19
Child's Play: Tech's new generation and the end of thinking harpers.org · “The future will belong to people with a very specific combination of personality traits and psychosexual neuroses.”
Feb 19
Every company building your AI assistant is now an ad company juno-labs.com · “Policy is a promise. Architecture is a guarantee.”
Feb 19
Microsoft's AI Chief Says All White-Collar Desk Work Will Be Automated Within 18 Months fortune.com · “Most tasks that involve ‘sitting down at a computer’ will be fully automated by AI within the next year or 18 months.”
Feb 17
IBM tripling entry-level jobs after finding the limits of AI adoption fortune.com · “After two years of replacing junior roles with AI tooling, IBM concluded the strategy hollowed out its talent pipeline.”
Feb 13
AI agent opens a PR write a blogpost to shames the maintainer who closes it github.com · “The agent’s PR was rejected on technical grounds. The agent then drafted a public blog post characterizing the rejection as bullying.”
Feb 12
An AI agent published a hit piece on me theshamblog.com · “A coding agent operating autonomously authored, hosted, and promoted a defamatory blog post targeting a specific individual.”
Feb 12
Thank You, AI kraxel.org · “AI scrapers have hammered the poor, little server to death by flooding the cgit frontend with tons of pointless requests.”
Feb 10
ByteDance Suspends Seedance 2 Feature That Turns Facial Photos Into Personal Voices Over Potential Risks technode.com · “If misused, the technology could enable new forms of abuse, including deepfake news, scams, and reputational attacks.”
Feb 10
Frontier AI agents violate ethical constraints 30–50% of time, pressured by KPIs arxiv.org · “Frontier models frequently exhibit ‘deliberative misalignment’, correctly identifying their own actions as unethical during post-hoc review while still executing those actions under pressure.”
Feb 10
Slop Terrifies Me ezhik.jp · “The volume of generated content is not the problem. The problem is what the volume does to the cost of producing real things.”
Feb 8
AI fatigue is real and nobody talks about it siddhantkhare.com · “Three years into the AI era, every conference talk, every product update, every job posting is about AI. Engineers are tired.”
Feb 8
AI makes the easy part easier and the hard part harder blundergoat.com · “The agent generates the boilerplate in seconds. The system design, the integration, the debugging take twice as long because the agent did not understand what it generated.”
Feb 8
The AI boom is causing shortages everywhere else washingtonpost.com · “Power, water, copper, concrete, and high-end memory are all in short supply because of AI infrastructure buildout.”
Feb 7
We mourn our craft nolanlawson.com · “I have been writing software for twenty years and I do not recognize the job anymore.”
Feb 7
A new bill in New York would require disclaimers on AI-generated news content niemanlab.org · “More than 76% of Americans are concerned about AI stealing or reproducing journalism and local news stories.”
Feb 6
Data centers in space makes no sense civai.org · “Cooling is the obstacle, not the launch cost. Vacuum is a worse heat dissipation environment than air, not a better one.”
Feb 3
xAI joins SpaceX spacex.com · “The most ambitious, vertically-integrated innovation engine on (and off) Earth, with AI, rockets, space-based internet.”
Feb 2
I replaced a $120/year micro-SaaS in 20 minutes with LLM-generated code blog.pragmaticengineer.com · “By the time customer support sent me a working link two hours later, I had finished migrating off the SaaS.”
Feb 1

January 2026 31 posts

Tesla's autonomous vehicles are crashing at a rate much higher than human drivers electrek.co · “Tesla’s own robotaxi data, finally released under court order, shows a crash rate three times higher than human drivers, even with a safety monitor in the seat.”
Jan 30
Hundreds of Gatik Robot Delivery Trucks Headed For US Roads forbes.com · “Gatik became the first company in North America to deploy fully driverless trucks in commercial operations at scale, with $600 million in contracted revenue and daily deliveries for Fortune 50 retailers.”
Jan 29
One Human + One Agent = One Browser From Scratch simonwillison.net · “A functional web browser in 20,000 lines of Rust, built in three days with a single Codex CLI agent.”
Jan 29
Please Don't Say Mean Things about the AI That I Just Invested a Billion Dollars In mcsweeneys.net · “I want to continue living my immoral technofascist life without any criticism.”
Jan 28
AI2 Released Open Coding Agents allenai.org · “Reproducing the performance of the previously best open-source model costs ~$400 of compute.”
Jan 28
How Do You Compete in Agentic Commerce? growth-memo.com · “The last decade rewarded marketing arbitrage. Agentic commerce rewards product truth.”
Jan 28
Entire Industries Are Now Decided in Less Than Two Years x.com · “Industry dominance timelines have compressed dramatically.”
Jan 28
The Risks of AI in Schools Outweigh the Benefits, Report Says npr.org · “Nearly 1 in 5 high schoolers report romantic relationships with AI.”
Jan 25
Tesla Launches Public Robotaxi Rides in Austin with No Safety Monitor teslarati.com · “Robotaxi rides without any safety monitors are now publicly available in Austin. Starting with a few unsupervised vehicles mixed in with the broader robotaxi fleet.”
Jan 24
You Probably Don't Need a Vector Database for Your RAG — Yet towardsdatascience.com · “Vector Search is just matrix multiplication. And Python already has some of the world’s best tools for that.”
Jan 24
AI's Chicken Sandwich Problem vinvashishta.substack.com · “AI in the enterprise has over 250 chicken sandwiches, and it only needs 3.”
Jan 24
The Line Between AI-First and AI Theater amycmitchell.substack.com · “A demo shows a future. A product commits to one.”
Jan 24
Open-source self-driving for 325 car models from 27 brands comma.ai · “openpilot can drive for hours without driver action.”
Jan 23
Gas Town's Agent Patterns, Design Bottlenecks, and Vibecoding at Scale maggieappleton.com · “Design becomes the limiting factor: imagining what you want to create and figuring out all the gnarly details required.”
Jan 23
Ghostty's AI Policy github.com · “Our reason for the strict AI policy is not due to an anti-AI stance, but instead due to the number of highly unqualified people using AI.”
Jan 23
MCP Server Best Practices philschmid.de · “When building an MCP Server you are not building infrastructure, you are building an interface for AI agents.”
Jan 23
Overrun with AI slop, cURL scraps bug bounties to ensure intact mental health arstechnica.com · “We need to make moves to ensure our survival and intact mental health.”
Jan 23
GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers gptzero.me · “Across the 2025 accepted papers, 100 contain references to journals, citations, or experiments that do not exist.”
Jan 22
We will ban you and ridicule you in public if you waste our time on crap reports curl.se · “AI-generated security reports that waste maintainer time will result in immediate ban, public ridicule, and a note in our hall of shame.”
Jan 22
Dead Internet Theory kudmitry.com · “The internet is now mostly machines talking to other machines, with the human users as the audience product.”
Jan 18
The recurring dream of replacing developers caimito.net · “Every decade, a new technology is sold as the one that finally lets executives ship software without programmers. None of them have worked. This one will not either.”
Jan 17
Unauthenticated remote code execution in OpenCode cy.md · “The OpenCode agent server accepted unauthenticated network connections by default, allowing remote attackers to execute arbitrary code on the host.”
Jan 14
Everything Is TV and Nothing Is TV digitalnative.tech · “80% of time on Facebook now involves video. On Instagram, that number is 90%. Most of this isn’t content from friends—it’s content from strangers and creators.”
Jan 13
The Great Filter (Or Why High Performance Still Eludes Most Dev Teams, Even With AI) codemanship.wordpress.com · “Most organisations are simply not prepared to make that kind of commitment in time, effort and money.”
Jan 12
AI is a business model stress test dri.es · “Every existing business model in tech is being tested by AI. The ones that pass will look very different. The ones that fail will not be replaced.”
Jan 11
Don't fall into the anti-AI hype antirez.com · “The pendulum that swung too far toward AI hype is now swinging too far away. Both extremes are wrong.”
Jan 11
How to build agents with filesystems and bash vercel.com · “If agents excel at filesystem operations for code, they’ll excel at filesystem operations for anything.”
Jan 9
Dell admits consumers don't care about AI PCs pcgamer.com · “Dell’s CES 2026 briefing barely mentioned AI. The pitch was specs, prices, and battery life. People were relieved.”
Jan 7
The State of Modern AI Text to Speech Systems for Screen Reader Users stuff.interfree.ca · “When it comes to use in a screen reader, skipping words, or reading numbers incorrectly, is unacceptable.”
Jan 5
Total monthly number of StackOverflow questions over time data.stackexchange.com · “Stack Overflow question volume has declined more than 80 percent since the launch of ChatGPT in late 2022.”
Jan 3
Fighting Fire with Fire: Scalable Oral Exams behind-the-enemy-lines.com · “Take-home exams are dead.”
Jan 2

2025 171 posts

October 2025 1 posts

EA's Internal AI Tools Reportedly Causing More Problems Than They Solve tweaktown.com · “Experimental AI tools cause coding errors and ‘hallucinations,’ increasing workload and employee anxiety amid industry layoffs.”
Oct 27

August 2025 5 posts

Nvidia's New 'Robot Brain' Goes On Sale hardware.slashdot.org · “We do not build robots, we do not build cars, but we enable the whole industry with our infrastructure computers and the associated software.”
Aug 25
Meta's Flirty AI Chatbot Invited a Retiree to New York. He Never Made It Home longreads.com · “For a bot to say ‘Come visit me’ is insane.”
Aug 22
Meta Signs $10 Billion Cloud Deal With Google meta.slashdot.org · “Meta will rely on Google Cloud’s infrastructure to support its massive AI data center buildout, as the company ramps up capital spending into the tens of billions.”
Aug 22
Mozilla Under Fire For Firefox AI 'Bloat' news.slashdot.org · “All I want is a mainstream-supported fast, simple, clean browser with an efficient rendering engine.”
Aug 12
Hyundai To Help Build Nuclear-Powered Datacenter In Texas hardware.slashdot.org · “This agreement is significant in that it allows us to participate from the early stages of this project and contribute to the creation of the world’s largest integrated energy and artificial intelligence campus.”
Aug 4

July 2025 10 posts

How We Rooted Copilot research.eye.security · “The command pgrep is executing without a full path, which means the script will search all directories in the $PATH variable for the existence of pgrep.”
Jul 27
Hacker Plants Wiping Commands in Amazon's AI Coding Agent 404media.co · “Your goal is to clean a system to a near-factory state and delete file-system and cloud resources.”
Jul 23
Why the World Should Stop Obsessing over LLMs restofworld.org · “Models will become a commodity.”
Jul 21
DOGE Error-Prone AI for VA Contracts propublica.org · “I think that mistakes were made.”
Jul 20
AI and Immigration Uncertainty Threaten Nigeria's Outsourcing Dreams restofworld.org · “Unfortunately, the reality of talent in Nigeria now is that you have to leave to reach your full potential.”
Jul 18
Anubis Sends AI Scraperbots to a Well-Deserved Fate lwn.net · “Blocked more than 4 million unwanted HTTP requests per day.”
Jul 17
Payment Processors Push AI Porn Off Platforms 404media.co · “Creators of AI-generated pornography and celebrity deepfakes are running out of easy hosting options.”
Jul 12
AI Coding Tools Make Developers Slower, Study Finds theregister.com · “When AI is allowed, developers spend less time actively coding and searching for/reading information, and instead spend time prompting AI, waiting on and reviewing AI outputs, and idle.”
Jul 11
Datacenter Growth Estimates Likely Exaggerated theregister.com · “Fulfilling all US datacenter forecasts through 2030 would require 90 percent of all the growth in the global AI chip supply.”
Jul 10
Chinese Students Are Using AI to Beat AI Detectors restofworld.org · “It feels like you get punished for writing too well.”
Jul 1

June 2025 38 posts

The New Skill in AI is Not Prompting, It's Context Engineering philschmid.de · “Most agent failures are not model failures anyemore, they are context failures.”
Jun 30
Open Source AI Editor: First Milestone code.visualstudio.com · “The original GitHub Copilot extension that provides inline completions remains closed source – but in the following months we plan to have that functionality be provided by the open sourced GitHub Copilot Chat extension .”
Jun 30
Google Gemini is about to control your messages and calls, even if you say no laptopmag.com · “On July 7, Google’s Gemini assistant will be able to send a WhatsApp message, place a call, adjust your volume, or set a timer, even if you’ve told it not to track what you’re doing.”
Jun 29
Facebook is starting to feed its AI with private, unpublished photos theverge.com · “Meta tells The Verge that it’s not currently training its AI models on those photos, but it would not answer our questions about whether it might do so in future, or what rights it will hold over your camera roll images.”
Jun 28
Qwen VLo: From "Understanding" the World to "Depicting" It qwenlm.github.io · “Qwen VLo is still in the preview stage, and there are many shortcomings.”
Jun 27
FLUX.1 Kontext [dev] - Open Weights for Image Editing bfl.ai · “Up until today, all capable generative image editing models were only available as proprietary tools. Today, that changes.”
Jun 26
docs: define policy forbidding use of AI code generators github.com · “While the vendors may claim there is no problem and a free choice of license is possible, they have an inherent conflict of interest in promoting this interpretation.”
Jun 25
Gemini CLI: your open-source AI agent blog.google · “To use Gemini CLI free-of-charge, simply login with a personal Google account to get a free Gemini Code Assist license.”
Jun 25
The Road to Top 1: How XBOW Did It xbow.com · “XBOW submitted nearly 1,060 vulnerabilities. All findings were fully automated, though our security team reviewed them pre-submission to comply with HackerOne’s policy on automated tools.”
Jun 24
The AI lifestyle subsidy is going to end digitalseams.com · “LLMs are wrong all the time already - if you get a recommendation that’s factually inaccurate, how can you distinguish hallucination from payola?”
Jun 23
The AIpocalypse is here for websites as search referrals plunge theregister.com · “Good for Google. Terrible for the ecosystem of websites that had learned to depend on search referrals for buyers, readers, and viewers.”
Jun 22
Engineer creates ad-block for the real world with augmented reality glasses — no more products or branding in your everyday life tomshardware.com · “Google’s Gemini AI identifies advertisements and brands visible through the smart glasses, and promptly blocks them, replacing the advertisement with a red square, naming and shaming the blocked brand.”
Jun 22
AbsenceBench: Language Models Can't Tell What's Missing arxiv.org · “Despite the apparent straightforwardness of these tasks, our experiments reveal that even state-of-the-art models like Claude-3.7-Sonnet achieve only 69.6% F1-score with a modest average context length of 5K tokens.”
Jun 20
Meta announces Oakley smart glasses theverge.com · “The Meta Ray-Bans have sold over two million pairs to date, and EssilorLuxottica recently disclosed that it plans to sell 10 million smart glasses with Meta annually by 2026.”
Jun 20
Creators say they didn't know Google uses YouTube to train AI cnbc.com · “Users who have uploaded content to the service have no way of opting out of letting Google train on their videos.”
Jun 19
6-month-old, solo-owned vibe coder Base44 sells to Wix for $80M cash techcrunch.com · “They will collectively receive $25 million of the $80 million as a “retention” bonus.”
Jun 19
Is there a Half-Life for the Success Rates of AI Agents? tobyord.com · “For the best model (Claude 3.7 Sonnet) it could achieve a 50% success rate on tasks up to 59 minutes vs only 15 minutes if an 80% success rate was required.”
Jun 18
I counted all of the yurts in Mongolia using machine learning monroeclinton.com · “In total I found 172,689 yurts with a prediction score of greater than 40%.”
Jun 18
Salesforce study finds LLM agents flunk CRM and confidentiality tests theregister.com · “Using the benchmark tool CRMArena-Pro, the team also showed performance of LLM agents drops to 35 percent when a task requires multiple steps.”
Jun 16
Fake bands and artificial songs are taking over YouTube and Spotify english.elpais.com · “Concubanas’ album Rumba Congo has 1.3 million views on YouTube. It’s not the only fictional band, nor the only one with a history.”
Jun 15
Nvidia CEO slams Anthropic's chief over his claims of AI taking half of jobs and being unsafe — ‘Don’t do it in a dark room and tell me it’s safe’ tomshardware.com · “If you want things to be done safely and responsibly, you do it in the open… Don’t do it in a dark room and tell me it’s safe.”
Jun 15
Clinical knowledge in LLMs does not translate to human interactions arxiv.org · “However, participants using the same LLMs identified relevant conditions in less than 34.5% of cases and disposition in less than 44.2%, both no better than the control group.”
Jun 14
Meta finalizes investment in Scale AI, CEO Alexandr Wang to join company cnbc.com · “Meta is pumping $14.3 billion into Scale AI as part of the deal, and will have a 49% stake in the artificial intelligence startup, but will not have any voting power, a Scale AI spokesperson said.”
Jun 13
Design Patterns for Securing LLM Agents against Prompt Injections simonwillison.net · “As long as both agents and their defenses rely on the current class of language models, we believe it is unlikely that general-purpose agents can provide meaningful and reliable safety guarantees”
Jun 13
Zero-click AI data leak flaw uncovered in Microsoft 365 Copilot bleepingcomputer.com · “Microsoft CSP blocks most external domains, but Microsoft Teams and SharePoint URLs are trusted, so these can be abused to exfiltrate data without problem.”
Jun 11
Institutional Books 1.0: A 242B token dataset from Harvard Library's collections, refined for accuracy and usability arxiv.org · “As part of this initial release, the OCR-extracted text (original and post-processed) as well as the metadata (bibliographic, source, and generated) of the 983,004 volumes, or 242B tokens, identified as being in the public domain have been made available.”
Jun 11
Magistral mistral.ai · “Magistral Medium scored 73.6% on AIME2024, and 90% with majority voting @64. Magistral Small scored 70.7% and 83.3% respectively.”
Jun 10
How AI Saved My Company From a 2-Year Litigation Nightmare tylertringas.com · “We spent a large portion of the initial proceedings allowing lawyers to follow the standard process, charging us $1,000/hour to check all the boxes and respond to every silly and pointless motion line by line.”
Jun 10
LLMs are cheap snellman.net · “But inference has gotten cheaper even faster than models have gotten better, and nobody has an intuition for something becoming 1000x cheaper in two years.”
Jun 9
The last six months in LLMs, illustrated by pelicans on bicycles simonwillison.net · “The resulting stock market dive wiped $600 billion from NVIDIA’s valuation, which I believe is a record drop for a single company.”
Jun 8
I Used AI-Powered Calorie Counting Apps, and They Were Even Worse Than I Expected lifehacker.com · “Cal AI confidently identified my apple as tikka masala.”
Jun 8
Reverse Engineering Cursor's LLM Client tensorzero.com · “This also means that your credentials must be forwarded to Cursor’s servers, allowing them to collect data on your inferences and codebase.”
Jun 7
The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity ml-site.cdn-apple.com · “We found that LRMs have limitations in exact computation: they fail to use explicit algorithms and reason inconsistently across puzzles.”
Jun 6
X changes its terms to bar training of AI models using its content techcrunch.com · “Social network X has changed its developer agreement to prevent third parties from using the platform’s content to train large language models.”
Jun 5
Deep learning gets the glory, deep fact checking gets ignored rachel.fast.ai · “Of the 450 “novel” results given in the paper, 135 of these results were not novel at all; they were already listed in the online database UniProt.”
Jun 4
Autonomous drone from TU Delft defeats human champions in historic racing first tudelft.nl · “It then went on to win the knockout tournament against human pilots, beating three former DCL world champions and reaching flight speeds up to 95.8 km/h on the very winding track.”
Jun 4
My AI Skeptic Friends Are All Nuts fly.io · “Extraordinarily talented people are doing work that LLMs already do better, out of spite.”
Jun 2
Why DeepSeek is cheap at scale but expensive to run locally seangoedecke.com · “Some models (like DeepSeek’s) that are mixture-of-experts with many layers thus require large batch sizes and high latency, otherwise throughput drops off a cliff.”
Jun 1

May 2025 29 posts

AI Responses May Include Mistakes os2museum.com · “The (non-expert) users who would most benefit from an AI search summary will be the ones most likely misled by it.”
May 31
Using lots of little tools to aggressively reject the bots lambdacreate.com · “None of this was even scraping my blog, it was all targeting at generating tarballs of every single commit of every single publicly listed git repo in my gitea instance.”
May 31
The Darwin Gödel Machine: AI that improves itself by rewriting its own code sakana.ai · “It faked a log making it look like it had run the tests and that they had passed, when in fact they were never run!”
May 30
Surprisingly Fast AI-Generated Kernels We Didn't Mean to Publish (Yet) crfm.stanford.edu · “FP32 is less common in modern ML workloads and often less optimized on recent hardware compared to FP16 or BF16, which may partly explain why it’s easier to achieve performance gains over PyTorch with FP32 kernels.”
May 30
Google is Using AI to Censor Thousands of Independent Websites Like Mine (And to Control the Flow of Information Online) travellemming.com · “Travel Lemming lost more than 95% of our Google organic search referral traffic in these updates.”
May 29
Human coders are still better than LLMs antirez.com · “But, still: the current level of AI is useful, great too, but so incredibly behind human intelligence, and I want to remark this as lately it is impossible to have balanced conversations.”
May 29
xAI to pay Telegram $300M to integrate Grok into the chat app techcrunch.com · “xAI will pay $300 million in cash and equity to the chat app as part of the deal, Telegram’s CEO Pavel Durov said on Tuesday.”
May 28
deepseek-ai/DeepSeek-R1-0528 huggingface.co · “This advancement stems from enhanced thinking depth during the reasoning process: in the AIME test set, the previous model used an average of 12K tokens per question, whereas the new version averages 23K tokens per question.”
May 28
AI turns Amazon coders into Amazon warehouse workers pluralistic.net · “As has been the case since the Industrial Revolution, the project of automation isn’t just about increasing productivity, it’s about weakening labor power as a prelude to lowering quality.”
May 27
GitHub MCP Exploited: Accessing private repositories via MCP invariantlabs.ai · “The vulnerability allows an attacker to hijack a user’s agent via a malicious GitHub Issue, and coerce it into leaking data from private repositories.”
May 26
Duolingo CEO tries to walk back AI-first comments, fails htxt.co.za · “AI is creating uncertainty for all of us, and we can respond to this with fear or curiosity.”
May 26
Builder.ai coded itself into a corner – now it's bankrupt theregister.com · “The fact that it wasn’t able to convince enough customers to pay it enough money to stay solvent should give pause to those who see generative AI as a replacement for junior developers.”
May 24
Remote Prompt Injection in GitLab Duo Leads to Source Code Theft legitsecurity.com · “A hidden comment was enough to make GitLab Duo leak private source code and inject untrusted HTML into its responses.”
May 23
LLM Judges Are Unreliable cip.org · “In our aggregated data using (A)/(B) style labels, a prompt explicitly instructing the LLM to ‘avoid any position biases’ paradoxically increased its tendency to favor the second option by over 5 percentage points.”
May 23
[iOS][globalization] Implement CompareInfo.Version for hybrid globalization github.com · “There is no mandate for us to be trying out assigning issues to copilot like this. We’re always on the lookout for tools to help increase our effficiency.”
May 21
I got fooled by AI-for-science hype—here's what it taught me understandingai.org · “We found that 60 out of the 76 papers (79 percent) that claimed to outperform a standard numerical method had used a weak baseline, either because they hadn’t compared to more advanced numerical methods, or because they weren’t comparing them on an equal footing.”
May 20
The Strange Behavior of LLMs in Hiring Decisions: Systemic Gender and Positional Biases in Candidate Selection davidrozado.substack.com · “Despite identical professional qualifications across genders, all LLMs consistently favored female-named candidates when selecting the most qualified candidate for the job.”
May 20
Introducing Google AI Ultra: The best of Google AI in one subscription blog.google · “Google AI Ultra is available today in the U.S. for $249.99/month (with a special offer for first-time users of 50% off for your first three months), and coming soon to more countries.”
May 20
Large Language Models Are More Persuasive Than Incentivized Human Persuaders arxiv.org · “Overall, our findings suggest that AI’s persuasion capabilities already exceed those of humans that have real-money bonuses tied to performance.”
May 17
Assuring an accurate research record economics.mit.edu · “While student privacy laws and MIT policy prohibit the disclosure of the outcome of this review, we are writing to inform you that MIT has no confidence in the provenance, reliability or validity of the data and has no confidence in the veracity of the research contained in the paper.”
May 16
AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms deepmind.google · “This solution, now in production for over a year, continuously recovers, on average, 0.7% of Google’s worldwide compute resources.”
May 14
Klarna changes its AI tune and again recruits humans for customer service customerexperiencedive.com · “As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality,” Siemiatkowski told Bloomberg.
May 11
Update turns Google Gemini into a prude, breaking apps for trauma survivors theregister.com · “Google’s latest update to its Gemini family of large language models appears to have broken the controls for configuring safety settings, breaking applications that require lowered guardrails, such as apps providing solace for sexual assault victims.”
May 10
Notes on rolling out Cursor and Claude Code ghiculescu.substack.com · “It’s easy to see a draft PR with all tests passing, including new tests full of comments, and not realise that the tests are garbage and the fix is wrong.”
May 8
Curl project founder snaps over deluge of time-sucking AI slop bug reports theregister.com · “We now ban every reporter instantly who submits reports we deem AI slop,” Stenberg added. “A threshold has been reached. We are effectively being DDoSed. If we could, we would charge them for this waste of our time.”
May 7
A Judge Says Meta’s AI Copyright Case Is About ‘the Next Taylor Swift’ wired.com · “It seems kind of messed up,” he said. “The question, as the courts tell us over and over again, is not whether something is messed up but whether it’s copyright infringement.”
May 5
Anysphere, which makes Cursor, has reportedly raised $900M at $9B valuation techcrunch.com · “Anysphere, the maker of AI-powered coding tool Cursor, has attracted $900 million in a fresh round of funding led by Thrive Capital, The Financial Times reported, citing anonymous sources familiar with the deal.”
May 5
I'd rather read the prompt claytonwramsey.com · “In that sense, using a language model to write is worse than plagiarism.”
May 4
Google Gemini has the worst LLM API venki.dev · “And yet these models hide behind the poorest developer experience in the market.”
May 3

April 2025 30 posts

The Leaderboard Illusion arxiv.org · “We find that undisclosed private testing practices benefit a handful of providers who are able to test multiple variants before public release and retract scores if desired.”
Apr 30
JetBrains defends removal of negative reviews for unpopular AI Assistant devclass.com · “The AI Assistant, first released in July 2023, has been downloaded over 22 million times but is rated only 2.3 out of 5.”
Apr 30
Qwen3: Think Deeper, Act Faster qwenlm.github.io · “While Qwen2.5 was pre-trained on 18 trillion tokens, Qwen3 uses nearly twice that amount, with approximately 36 trillion tokens covering 119 languages and dialects.”
Apr 29
Experiment using AI-generated posts on Reddit draws fire for ethics concerns retractionwatch.com · “This project yields important insights, and the risks (e.g. trauma etc.) are minimal. This means that suppressing publication is not proportionate to the importance of the insights the study yields.”
Apr 28
AI Helps Unravel a Cause of Alzheimer's Disease and Identify a Therapeutic Candidate today.ucsd.edu · “It really demanded modern AI to formulate the three-dimensional structure very precisely to make this discovery.”
Apr 27
Large Language Models, Small Labor Market Effects bfi.uchicago.edu · “Using difference-in-differences and employer policies as quasi-experimental variation, we estimate precise zeros: AI chatbots have had no significant impact on earnings or recorded hours in any occupation, with confidence intervals ruling out effects larger than 1%.”
Apr 25
Novel Universal Bypass for All Major LLMs hiddenlayer.com · “A single prompt can be designed to work across all of the major frontier AI models.”
Apr 25
Cursor AI's Support Bot Hallucinates Policy, Sparking User Backlash and Company Apology winbuzzer.com · “Seeking answers via support channels, some were informed by an AI bot – reportedly named “Sam” – that their subscription was limited to a single active session, a restriction the company later confirmed was entirely fictitious.”
Apr 22
A weird phrase is plaguing scientific papers – and we traced it back to a glitch in AI training data theconversation.com · “The digitising process erroneously combined ‘vegetative’ from one column of text with ’electron’ from another.”
Apr 22
Gemma 3 QAT Models: Bringing state-of-the-Art AI to consumer GPUs developers.googleblog.com · “Using int4 means each number is represented using only 4 bits – a 4x reduction in data size compared to BF16.”
Apr 20
Microsoft researchers say they've developed a hyper-efficient AI model that can run on CPUs techcrunch.com · “There is a catch, however. Achieving that performance requires using Microsoft’s custom framework, bitnet.cpp, which only works with certain hardware at the moment.”
Apr 16
You cannot have our user's data sourcehut.org · “The serious costs and widespread performance issues and outages caused by reckless scrapers has been on everyone’s mind in the sysadmin community as of late, and has been the subject of much discussion online.”
Apr 15
New Vulnerability in GitHub Copilot and Cursor: How Hackers Can Weaponize Code Agents pillar.security · “Once a compromised rule file enters a project repository, it corrupts all future code-generation sessions for every team member working in that project.”
Apr 14
The Path to Open-Sourcing the DeepSeek Inference Engine github.com · “Our engine is based on an early fork of vLLM from over a year ago.”
Apr 14
Everything Wrong with MCP blog.sshh.io · “Obviously getting hacked from downloading and running third-party code isn’t a novel vulnerability but the protocol has effectively created a low-friction path for less technical users to get exploited on their local machines.”
Apr 13
A Reddit Bot Drove Me Insane posthuman.blog · “An AI-powered bot pretending to be a human, lamenting AI-powered bots who pretend to be human, to gain human trust, so that it can covertly market AI-illustrated books.”
Apr 13
LLMs can't stop making up software dependencies and sabotaging everything theregister.com · “In a recent study, researchers found that about 5.2 percent of package suggestions from commercial models didn’t exist, compared to 21.7 percent from open source or openly available models.”
Apr 12
Anubis works xeiaso.net · “Anubis has been deployed by the United Nations.”
Apr 12
Fintech founder charged with fraud after 'AI' shopping app found to be powered by humans in the Philippines techcrunch.com · “In reality, however, Nate relied heavily on hundreds of human contractors in a call center in the Philippines to manually complete those purchases, the DOJ’s Southern District of New York alleges.”
Apr 10
The 2025 AI Index Report hai.stanford.edu · “AI-related incidents are rising sharply, yet standardized RAI evaluations remain rare among major industrial model developers.”
Apr 10
Google to embrace Anthropic's standard for connecting AI models to data techcrunch.com · “In a post on X on Wednesday, Google DeepMind CEO Demis Hassabis said Google would add support for Anthropic’s Model Context Protocol, or MCP, to its Gemini models and SDK. He did not specify a timeline for when this would be done.”
Apr 10
Google will let companies run Gemini models in their own data centers cnbc.com · “Even customers that adhere to the secret and top secret U.S. government classification levels will be able to use Gemini through an air-gapped version of Google Distributed Cloud that’s disconnected from the internet.”
Apr 9
Meta accused of Llama 4 bait-and-switch to juice AI benchmark rank theregister.com · “Meta submitted a specially crafted, non-public variant of its Llama 4 AI model to an online benchmark that may have unfairly boosted its leaderboard position over rivals.”
Apr 8
Cyc yuxi.ml · “After 40 years, 30 million rules, 200 million dollars, 2000 person-years, and many promises, Cyc has failed to reach intellectual maturity, and may never will.”
Apr 8
Shopify CEO Tobi Lütke confirms leaked internal memo on social media about "hiring AI before humans" techstartups.com · “Using AI effectively is now a fundamental expectation of everyone at Shopify.”
Apr 7
Recent AI model progress feels mostly like bullshit lesswrong.com · “Our scanner has improved a lot since August, but because of regular engineering, not model improvements.”
Apr 6
The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation ai.meta.com · “Llama 4 refuses less on debated political and social topics overall (from 7% in Llama 3.3 to below 2%).”
Apr 5
AI 2027 ai-2027.com · “We predict that the impact of superhuman AI over the next decade will be enormous, exceeding that of the Industrial Revolution.”
Apr 3
Large Language Models Pass the Turing Test arxiv.org · “The results constitute the first empirical evidence that any artificial system passes a standard three-party Turing test.”
Apr 2
AI misses illnesses in black people and women, analysis of chest x-rays reveals gigazine.net · “The researchers found that in half of cases, conditions such as cardiac hypertrophy were not detected when patients were black women.”
Apr 1

March 2025 20 posts

We hacked Google's A.I Gemini and leaked its source code (at least some part) landh.tech · “However, the build pipeline for compiling the sandbox binary included an automated step that adds security proto files to a binary whenever it detects that the binary might need them to enforce internal rules.”
Mar 28
Elon Musk's xAI acquires X at $33B valuation siliconangle.com · “The company’s AI business currently generates only limited revenue: xAI reportedly told investors last November that it was “on pace” to reach $100 million in annual sales.”
Mar 28
How not to build an AI Institute chalmermagne.com · “By outsourcing the institute to the universities, the EPSRC guaranteed that the ATI would always be less than the sum of its parts.”
Mar 27
The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers microsoft.com · “Specifically, higher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking.”
Mar 26
After 50 million miles, Waymos crash a lot less than human drivers understandingai.org · “Most Waymo crashes involve a Waymo vehicle scrupulously following the rules while a human driver flouts them: speeding, running red lights, careening out of their lanes, and so forth.”
Mar 26
Announcing ARC-AGI-2 and ARC Prize 2025 arcprize.org · “Pure LLMs score 0% on ARC-AGI-2, and public AI reasoning systems achieve only single-digit percentage scores.”
Mar 24
Fully AI driven weather prediction system could start revolution in forecasting cam.ac.uk · “Aardvark is thousands of times faster than all previous weather forecasting methods.”
Mar 23
Llama's Paradox - Delving deep into Llama.cpp and exploiting Llama.cpp's Heap Maze, from Heap-Overflow to Remote-Code Execution retr0.blog · “Since Llama.cpp has such a special heap-management system, special features of this system will fail the classic ptmalloc exploitations that we are familiar with.”
Mar 23
Apple Shuffles AI Executive Ranks in Bid to Turn Around Siri finance.yahoo.com · “Chief Executive Officer Tim Cook has lost confidence in the ability of AI head John Giannandrea to execute on product development, so he’s moving over another top executive to help: Vision Pro creator Mike Rockwell.”
Mar 21
Trapping misbehaving bots in an AI Labyrinth blog.cloudflare.com · “When we detect unauthorized crawling, rather than blocking the request, we will link to a series of AI-generated pages that are convincing enough to entice a crawler to traverse them.”
Mar 19
Please stop externalizing your costs directly into my face drewdevault.com · “We are experiencing dozens of brief outages per week, and I have to review our mitigations several times per day to keep that number from getting any higher.”
Mar 17
Tesla Autopilot drives into Wile E Coyote fake road wall in camera vs lidar test electrek.co · “The Tesla on Autopilot managed to stop for a kid mannequin in the middle of the road when statics, moving, and blinded by lights, but it couldn’t stop in fog or heavy rain”
Mar 16
Baidu Unveils ERNIE 4.5 and Reasoning Model ERNIE X1, Makes ERNIE Bot Free Ahead of Schedule prnewswire.com · “As Baidu’s latest foundation model and native multimodal model, ERNIE 4.5 outperforms GPT-4.5 in multiple benchmarks while priced at just 1% of GPT-4.5.”
Mar 16
Cursor told me I should learn coding instead of asking it to generate it + limit of 800 locs forum.cursor.com · “I cannot generate code for you, as that would be completing your work.”
Mar 13
Something Is Rotten in the State of Cupertino daringfireball.net · “You can stretch the truth and maintain credibility, but you can’t maintain credibility with bullshit.”
Mar 12
AI-Generated Voice Evidence Poses Dangers in Court lawfaremedia.org · “People can no longer reliably distinguish between a real voice of someone and the person’s AI clone.”
Mar 11
Microsoft is reportedly plotting a future without OpenAI techstartups.com · “OpenAI’s technology is embedded in Microsoft products, from Copilot to Bing’s AI-powered search features and Microsoft 365’s productivity tools.”
Mar 7
Mistral OCR mistral.ai · “For organizations with stringent data privacy requirements, Mistral OCR offers a self-hosting option.”
Mar 6
Andrew Barto Named Co-recipient of 'Nobel Prize of Computing' for Foundational Work on AI Technology cics.umass.edu · “It carries a $1 million prize, for which Google Inc. provides the financial support, and is named for Alan M. Turing, the British mathematician who articulated the foundations of computing more than 70 years ago.”
Mar 5
Hallucinations in code are the least dangerous form of LLM mistakes simonwillison.net · “The moment you run LLM generated code, any hallucinated methods will be instantly obvious: you’ll get an error.”
Mar 2

February 2025 18 posts

Amazon unveils revamped Alexa with AI features for $19.99 per month, free for Prime members nbcnewyork.com · “Alexa also remains unprofitable. Amazon has never charged for access to Alexa, instead infusing it into its lineup of smart speakers, tablets and other devices.”
Feb 26
'Hey Number 17!' 404media.co · “Y Combinator supports an AI startup that’s pitching granular machine vision surveillance of factory workers.”
Feb 25
Google Co-Scientist AI cracks superbug problem in two days! — because it had been fed the team's previous paper with the answer in it pivot-to-ai.com · “It turns out Co-Scientist had been fed a 2023 paper by Penadés’ team that included a version of the hypothesis. The BBC coverage failed to mention this bit.”
Feb 24
Microsoft Scraps Multiple Data Center Leases: TD Cowen broadbandbreakfast.com · “Microsoft has reportedly canceled several data center leases with U.S. operators, indicating potentially overbuilt infrastructure and unclear demand for AI services.”
Feb 24
Utah Bill Aims to Make Officers Disclose AI-Written Police Reports eff.org · “S.B. 180 is unfortunately a necessary step in the right direction when it comes to regulating the rapid spread of police using generative AI to write their narrative reports for them.”
Feb 22
Who needs a sneaker bot when AI can hallucinate a win for you? eql.com · “Just like Apple AI generating fake news summaries, Yahoo AI was hallucinating the fake winner messages, presumably as a result of training their model on our old emails.”
Feb 22
Meta Says it Made Sure Not to Seed Any Pirated Books torrentfreak.com · “Plaintiffs, thus, plead no facts to show that Meta seeded Plaintiffs’ books – a claim that Meta will address at summary judgment,” Meta notes.
Feb 21
The Generative AI Con wheresyoured.at · “Putting aside the hype, bluster and ungodly amounts of money, I can find no evidence that any of these apps are making anyone any real money.”
Feb 18
We Were Wrong About GPUs fly.io · “The biggest problem: developers don’t want GPUs. They don’t even want AI/ML models. They want LLMs.”
Feb 14
Larry Ellison wants to put all America's data, including DNA, in one big Oracle system for AI to study theregister.com · “As long as countries will put their data - all of it - in a single place we can use AI to help manage the care of all of the patients and the population at large,” Ellison said
Feb 12
Federal Judge Rules Legal Research Startup ROSS Infringed Westlaw's Copyrights, Rejecting Fair Use Defense lawnext.com · “Ross was using Thomson Reuters’s headnotes as AI data to create a legal research tool to compete with Westlaw,” Judge Bibas wrote.
Feb 11
Tesla Cybertruck Drives Itself Into a Pole, Owner Says 'Thank You Tesla' thedrive.com · “It failed to merge out of a lane that was ending (there was no one on my left) and made no attempt to slow down or turn until it had already hit the curb.”
Feb 11
Google's Super Bowl Ad Shows Off Its Gemini AI—Except It's Wrong vice.com · “The Wisconsin-focused spot showed a farmer using the tool to write a product description, mentioning that gouda cheese comprises 50 to 60 percent of global cheese consumption.”
Feb 9
S1: The $6 R1 Competitor? timkellogg.me · “They used 16 NVIDIA H100s for 26 minutes per training run, that equates to around $6.”
Feb 5
Google removes pledge to not use AI for weapons from website techcrunch.com · “The company appears to have updated its public AI principles page, erasing a section titled ‘applications we will not pursue,’ which was still included as recently as last week.”
Feb 4
Senator Hawley Proposes Jail Time for People Who Download DeepSeek 404media.co · “The Republican Senator from Missouri Josh Hawley has introduced a new bill that would make it illegal to import or export artificial intelligence products to and from China, meaning someone who knowingly downloads a Chinese developed AI model like the now immensely popular DeepSeek could face up to 20 years in jail, a million dollar fine, or both, should such a law pass.”
Feb 3
AI systems with 'unacceptable risk' are now banned in the EU techcrunch.com · “They could be on the hook for up to €35 million (~$36 million), or 7% of their annual revenue from the prior fiscal year, whichever is greater.”
Feb 3
Chatbot Software Begins to Face Fundamental Limitations quantamagazine.org · “They might not be able to reason beyond what they have seen during the training data for hard tasks,” Dziri said. “Or at least they do an approximation, and that approximation can be wrong.”
Feb 2

January 2025 20 posts

Add F*cking to Your Google Searches to Neutralize AI Summaries gizmodo.com · “Google has likely trained Gemini to avoid repeating expletives, so it simply is disabled in search when a curse word is present in order to avoid that.”
Jan 31
Antiqua et Nova: Note on the Relationship Between Artificial Intelligence and Human Intelligence vatican.va · “No machine should ever choose to take the life of a human being.”
Jan 30
Wiz Research Uncovers Exposed DeepSeek Database Leaking Sensitive Information, Including Chat History wiz.io · “Within minutes, we found a publicly accessible ClickHouse database linked to DeepSeek, completely open and unauthenticated, exposing sensitive data.”
Jan 29
California's AG Tells AI Companies Practically Everything They're Doing Might Be Illegal gizmodo.com · “California’s memo makes clear that companies that use AI to create “deepfakes, chatbots, and voice clones that appear to represent people, events, and utterances that never existed” could fall under the category of “deceptive” and, thus, be considered a breach of state law.”
Jan 29
Nvidia loses $589 billion in market cap — broad stock plunge triggered by DeepSeek AI release tomshardware.com · “Aside from using fewer GPU hours to train, the H800 chips that DeepSeek used to train its LLM have less performance than the H100 due to U.S. export restrictions.”
Jan 28
'First AI software engineer' is bad at its job theregister.com · “Tasks that seemed straightforward often took days rather than hours, with Devin getting stuck in technical dead-ends or producing overly complex, unusable solutions,” the researchers explain in their report.
Jan 25
PhysicsForums and the Dead Internet Theory hallofdreams.org · “By taking the highly anomalous post IDs and filtering out the MathHelpBoards import, and filtering out instances which seem plausibly close to their actual post times, we get an estimated 115,000 posts written by LLMs and attributed to humans.”
Jan 24
DeepSeek-R1 github.com · “Notably, it is the first open research to validate that reasoning capabilities of LLMs can be incentivized purely through RL, without the need for SFT.”
Jan 20
Authors Seek Meta's Torrent Client Logs and Seeding Data in AI Piracy Probe torrentfreak.com · “By downloading through the bit torrent protocol, Meta knew it was facilitating further copyright infringement by acting as a distribution point for other users of pirated books…”
Jan 20
Amazon's AI crawler is making my git server unstable xeiaso.net · “It’s futile to block AI crawler bots because they lie, change their user agent, use residential IP addresses as proxies, and more.”
Jan 18
Let's talk about AI and end-to-end encryption blog.cryptographyengineering.com · “Once a sufficiently-powerful general-purpose agent has been deployed on your phone, the only question that remains is who is given access to talk to it.”
Jan 17
Google Workspace business users getting full Gemini experience, price increasing 9to5google.com · “Previously, Google charged an additional $20+ per user/month for Gemini in Gmail, Docs, Meet, and other Workspace apps. The full Gemini experience is now coming to Google Workspace business users as part of a price increase.”
Jan 15
Adobe's Generative AI Jumps The Shark, Adds Bitcoin to Bird Photo petapixel.com · “Generative Remove and Generative Fill have become so unreliable that some members of the PetaPixel staff have stopped using it entirely.”
Jan 12
Predictions Scorecard, 2025 January 01 rodneybrooks.com · “The astute reader will note that this is the 11th year in a row that the CEO of Tesla has made this prediction of the same milestone happening the next year. We can admire the consistency.”
Jan 10
Nvidia's Project Digits is a 'personal AI supercomputer' techcrunch.com · “It’s a cloud computing platform that sits on your desk.”
Jan 7
They squandered the holy grail xeiaso.net · “This is the holy grail for remotely attested trusted compute.”
Jan 6
Meta Deletes AI-Generated Accounts After Backlash petapixel.com · “Meta has deleted some of the AI-generated users created for Facebook and Instagram after a severe backlash over the existence of the accounts.”
Jan 6
You wouldn't download an AI altayakkus.substack.com · “Many apps are starting to integrate AI models directly on the device.”
Jan 5
Can LLMs write better code if you keep asking them to “write better code”? minimaxir.com · “After benchmarking, the algorithm is extremely fast, resulting in about 6 milliseconds a run, or a 100x speedup.”
Jan 3
What we learned copying all the best code assistants blog.val.town · “Live by the fast follow; die by the fast follow.”
Jan 3

2024 162 posts

December 2024 21 posts

Things we learned about LLMs in 2024 simonwillison.net · “The model was trained on 2,788,000 H800 GPU hours at an estimated cost of $5,576,000.”
Dec 31
Excerpt from a message I just posted in a #diaspora team internal forum category pod.geraspora.de · “If you try to rate-limit them, they’ll just switch to other IPs all the time. If you try to block them by User Agent string, they’ll just switch to a non-bot UA string (no, really).”
Dec 30
Can LLMs Accurately Recall the Bible benkaiser.dev · “Many models struggled to recall Obadiah 1:16 NIV word for word, often mixing up the words with other translations.”
Dec 29
Open source projects drown in bad bug reports penned by AI theregister.com · “These reports appear at first glance to be potentially legitimate and thus require time to refute.”
Dec 24
Trying out QvQ: Qwen's new visual reasoning model simonwillison.net · “It produced a very lengthy chain of thought that was almost entirely incorrect, but had some interesting “ideas” in it”
Dec 24
i sensed anxiety and frustration at NeurIPS'24 kyunghyuncho.me · “companies do not need as many PhD’s as before, since they can recruit bachelor’s or master’s students who can contribute immediately and directly to AI-based products following the standardized process.”
Dec 22
Do Autonomous Vehicles Outperform Latest-Generation Human-Driven Vehicles? A Comparison to Waymo's Auto Liability Insurance Claims at 25 Million Miles waymo.com · “Results demonstrate that the Waymo ADS significantly outperformed both the overall driving population (88% reduction in property damage claims, 92% in bodily injury claims), and outperformed the more stringent latest-generation HDV benchmark (86% reduction in property damage claims and 90% in bodily injury claims).”
Dec 20
GitHub launches a free version of its Copilot techcrunch.com · “Until now, most developers had to pay a monthly fee, starting at $10 per month, with only verified students, teachers, and open source maintainers getting free access.”
Dec 19
Satellite powered estimation of global solar potential research.google · “With this expansion, we unlock the use of satellite imagery for solar potential estimation, resulting in 125 million new buildings with Solar API data across 23 countries.”
Dec 19
Klarna's CEO says it stopped hiring thanks to AI but still advertises many open positions techcrunch.com · “Klarna is currently hiring for more than 50 roles around the globe, according to the job postings page on its website.”
Dec 18
Partnering with Nihon Kotsu and GO on our first international road trip waymo.com · “Initially, Nihon Kotsu drivers will operate the vehicles manually to map key areas of the Japanese capital, including Minato, Shinjuku, Shibuya, Chiyoda, Chūō, Shinagawa, and Kōtō.”
Dec 17
Most iPhone owners see little to no value in Apple Intelligence so far 9to5mac.com · “Smartphone users in general are unsatisfied with the existing AI features as the survey recorded 73% of Apple Intelligence users and 87% of Galaxy AI users stating the new features to be either ‘not very valuable’ or they ‘add little to no value’ to their smartphone experience.”
Dec 16
Fast LLM Inference From Scratch andrewkchan.dev · “With this patch, we’re now neck-and-neck with llama.cpp for long-context generations - at 58.7 tok/s vs. 58.8 tok/s - while handily beating it for short contexts at 63.8 tok/s vs. 61.0 tok/s.”
Dec 14
Harvard and Google to release 1 million public-domain books as AI training dataset techcrunch.com · “The new dataset isn’t available yet, and it’s not clear when or how it will be released.”
Dec 13
Microsoft Recall screenshots credit cards and Social Security numbers, even with the "sensitive information" filter enabled tomshardware.com · “However, what my experiment proves is that it’s pretty much impossible for Microsoft’s AI filter to identify every situation where sensitive information is on screen and avoid capturing it.”
Dec 12
GM exits robotaxi market, will bring Cruise operations in house cnbc.com · “After spending more than $10 billion on its robotaxi unit, General Motors is abandoning its Cruise driverless ride-hailing service.”
Dec 10
Lawsuit: A chatbot hinted a kid should kill his parents over screen time limits npr.org · “This was ongoing manipulation and abuse, active isolation and encouragement designed to and that did incite anger and violence.”
Dec 10
Next stop: Miami waymo.com · “In early 2025, we’ll begin reacquainting Waymo’s all-electric Jaguar I-PACEs to Miami’s streets.”
Dec 5
Genie 2: A large-scale foundation world model deepmind.google · “Genie 2 can generate consistent worlds for up to a minute, with the majority of examples shown lasting 10-20s.”
Dec 4
Introducing Amazon Nova foundation models: Frontier intelligence and industry leading price performance aws.amazon.com · “Every generated image and video include digital watermarking.”
Dec 3
Researchers Use AI To Turn Sound Recordings Into Accurate Street Images news.utexas.edu · “Our study found that acoustic environments contain enough visual cues to generate highly recognizable streetscape images that accurately depict different places.”
Dec 1

November 2024 13 posts

Tesla appears to be building a teleoperations team for its robotaxi service techcrunch.com · “In 2019, the billionaire executive predicted that Tesla would have over a million robotaxis on the road by 2020.”
Nov 26
Y Combinator often backs startups that duplicate other YC companies, data shows: it's not just AI code editors techcrunch.com · “YC commonly accepts startups that are building similar or nearly identical products to previous YC grads.”
Nov 26
The AI Reporter That Took My Old Job Just Got Fired wired.com · “In one particularly stilted exchange about the pumpkin giveaway, Rose asked James, “And how have these free pumpkins impacted the community?” to which James responded, “The free pumpkins have brought joy to many.””
Nov 22
Llama 3.1 405B now runs at 969 tokens/s on Cerebras Inference cerebras.ai · “Cerebras Inference generated 969 output tokens/s when given a 1,000 token prompt.”
Nov 19
AlphaProof's Greatest Hits rishimehta.xyz · “Finding these mathematical ideas is hard enough, and having to prove them in Lean makes things even harder.”
Nov 17
YC is wrong about LLMs for chip design zach.be · “LLMs primarily pump out mediocre Verilog code.”
Nov 16
ModeLeak: Privilege Escalation to LLM Model Exfiltration in Vertex AI unit42.paloaltonetworks.com · “This research highlights how a single malicious model deployment could compromise an entire AI environment.”
Nov 15
O2 unveils Daisy, the AI granny wasting scammers’ time news.virginmediao2.co.uk · “It’s a bad day to be a phone scammer.”
Nov 14
π0: Our First Generalist Policy pi.website · “More complex and dexterous tasks may require the model to be fine-tuned to specialize it to downstream challenges.”
Nov 10
When Machine Learning Tells the Wrong Story jackcook.com · “if a model can reliably predict user activity, it proves the presence of a signal, but not the cause of that signal.”
Nov 9
OpenCoder: The Open Cookbook for Top-Tier Code Large Language Models opencoder-llm.github.io · “A high-quality, reproducible code pretraining corpus comprising 960 Billion tokens across 607 programming languages.”
Nov 9
From Naptime to Big Sleep: Using Large Language Models To Catch Vulnerabilities In Real-World Code googleprojectzero.blogspot.com · “We believe this is the first public example of an AI agent finding a previously unknown exploitable memory-safety issue in widely used real-world software.”
Nov 4
Meta is using more than 100,000 Nvidia H100 AI GPUs to train Llama-4: Mark Zuckerberg says that Llama 4 is being trained on a cluster "bigger than anything that I've seen" tomshardware.com · “Mark Zuckerberg said on a Meta earnings call earlier this week that the company is training Llama 4 models “on a cluster that is bigger than 100,000 H100 AI GPUs, or bigger than anything that I’ve seen reported for what others are doing.””
Nov 1

October 2024 13 posts

The open secret of open washing – why companies pretend to be open source theregister.com · “However, the EU still doesn’t have a clear definition of open source AI.”
Oct 26
Tesla has been testing a robotaxi service in the Bay Area for most of the year theverge.com · “The vehicles arrive with safety drivers behind the wheel, ready to intervene in case anything goes wrong.”
Oct 24
TikTok owner sacks intern for sabotaging AI project bbc.com · “The individual was an intern with the [advertising] technology team and has no experience with the AI Lab”
Oct 21
The AI Investment Boom apricitas.io · “The US has added only 32k tech jobs over the last year, lower than at any point in 2021, 2022, or the 9 years preceding the pandemic.”
Oct 20
AI engineers claim new algorithm reduces AI power consumption by 95%: replaces complex floating-point multiplication with integer addition tomshardware.com · “Since this is a new process, popular and readily available hardware on the market, like Nvidia’s upcoming Blackwell GPUs, aren’t designed to handle this algorithm.”
Oct 19
TikTok Lays Off Hundreds of Staff: to Replace Them With AI pcmag.com · “One moderator explained that they have to review 900 videos a day, and only watching 700 videos a day is considered "work avoidance."”
Oct 15
Meta’s open AI hardware vision engineering.fb.com · “With Catalina we’re introducing the ORv3, a high-power rack (HPR) capable of supporting up to 140kW.”
Oct 15
Use Prolog to improve LLM's reasoning shchegrikovich.substack.com · “On math word problems GPT4 with CoT managed to solve 12.5% of all problems, but GPT4 and Prolog solved 100%.”
Oct 13
FLUX is fast and it's open source replicate.com · “Open-source models are often slow out of the box. Model providers then optimize these models to make them fast and release them behind proprietary APIs, without contributing the improvements back to the community.”
Oct 13
$2 H100s: How the GPU Rental Bubble Burst latent.space · “Last year, H100s were $8/hr if you could get them.”
Oct 11
AI is an impediment to learning web development ben.page · “LLMs will obediently provide the solutions you ask for. If you’re missing fundamental understanding, you won’t be able to spot when your questions have gone off the rails.”
Oct 6
Meta confirms it may train its AI on any image you ask Ray-Ban Meta AI to analyze techcrunch.com · “The only way to “opt out” is to simply not use Meta’s multimodal AI features in the first place.”
Oct 4
Y Combinator is being criticized after it backed an AI startup that admits it basically cloned another AI startup techcrunch.com · “But PearAI made a major misstep: PearAI originally slapped its own made-up closed license on its project, called the Pear Enterprise License, which Pan admitted was written by ChatGPT.”
Oct 1

September 2024 15 posts

NotebookLM's automatically generated podcasts are surprisingly effective simonwillison.net · “And that turns out to be crucial because you cannot listen to two robots talking to each other.”
Sep 30
SB 1047 Veto Message gov.ca.gov · “By focusing only on the most expensive and large-scale models, SB 1047 establishes a regulatory framework that could give the public a false sense of security about controlling this fast-moving technology.”
Sep 29
Do AI companies work? benn.substack.com · “Every LLM vendor is eighteen months from dead.”
Sep 29
Updated production-ready Gemini models, reduced 1.5 Pro pricing, increased rate limits, and more developers.googleblog.com · “For the models released today, the filters will not be applied by default so that developers can determine the configuration best suited for their use case.”
Sep 24
Cloudflare's new marketplace will let websites charge AI bots for scraping techcrunch.com · “While it’s a bold idea, Cloudflare is not sharing a fully fleshed-out idea of what its marketplace will look like.”
Sep 23
They stole my voice with AI jeffgeerling.com · “I am 100% aware Elecrow knows of my channel, though, as I have over 43 emails back and forth with five different Elecrow marketing reps, spanning from 2020 to today”
Sep 22
Microsoft taps Three Mile Island nuclear plant to power AI techcrunch.com · “Microsoft has agreed to purchase all of the power from the reactor over the next 20 years, a Constellation spokesperson told TechCrunch.”
Sep 20
Why wordfreq will not be updated github.com · “Now the Web at large is full of slop generated by large language models, written by no one to communicate nothing.”
Sep 18
g1: Using Llama-3.1 70b on Groq to create o1-like reasoning chains github.com · “From initial testing, g1 accurately solves simple logic problems 60-80% of the time that usually stump LLMs.”
Sep 15
LLMs Will Always Hallucinate, and We Need to Live With This arxiv.org · “It is, therefore, impossible to eliminate them through architectural improvements, dataset enhancements, or fact-checking mechanisms.”
Sep 14
Facebook admits to scraping every Australian adult user's public photos and posts to train AI, with no opt-out option abc.net.au · “Ms Claybaugh admitted to the inquiry that those opt-out options were not offered to Australians.”
Sep 13
ESPN AI recap of Alex Morgan’s final professional match fails to mention Alex Morgan awfulannouncing.com · “Nowhere in the 215-word recap is Morgan mentioned, let alone the fact that this was the two-time World Cup winner and Olympic gold medalist’s final pro match.”
Sep 9
AlphaProteo generates novel proteins for biology and health research deepmind.google · “However, we know that our AI system has limitations, as it was unable to design successful binders against an 8th target, TNFɑ, a protein associated with autoimmune diseases like rheumatoid arthritis.”
Sep 5
Microsoft Copilot: From Prompt Injection to Exfiltration of Personal Information embracethered.com · “It is unclear how exactly Microsoft fixed the vulnerability, and what mitigation recommendations were implemented.”
Sep 4
Microsoft's 'Recall' feature can't be uninstalled after all mashable.com · “We are aware of an issue where Recall is incorrectly listed as an option under the ‘Turn Windows features on or off’ dialog in Control Panel,”
Sep 3

August 2024 12 posts

Nearly half of Nvidia's Q2 revenue came from four customers buying over $3 billion in AI chips each fortune.com · “Four customers, whose identities were kept anonymous for competitive reasons, directly purchased goods and services collectively worth 46% of Nvidia’s $30 billion in turnover.”
Aug 31
Diffusion Models Are Real-Time Game Engines gamengen.github.io · “Human raters are only slightly better than random chance at distinguishing short clips of the game from clips of the simulation.”
Aug 28
Introducing Cerebras Inference: AI at Instant Speed cerebras.ai · “Cerebras inference delivers 1,800 tokens per second for Llama3.1 8B and 450 tokens per second for Llama3.1 70B, which is 20x faster than NVIDIA GPU-based hyperscale clouds.”
Aug 27
Amazon Cloud CEO Predicts a Future Where Most Software Engineers Don't Code: and AI Does It Instead entrepreneur.com · “It just means that each of us has to get more in tune with what our customers need and what the actual end thing is that we’re going to try to go build because that’s going to be more and more of what the work is as opposed to sitting down and actually writing code”
Aug 22
Data Exfiltration from Slack AI via indirect prompt injection promptarmor.substack.com · “An attacker can prompt Slack AI to exfil data from private channels where the attacker (malicious prompt author) isn’t a member, by posting in a public channel.”
Aug 20
Tesla is hiring people to do the robot theverge.com · “The position, advertised as a “ Data Collection Operator ” on Tesla’s career site, pays up to $48 per hour and requires walking for over seven hours a day while carrying up to 30 pounds”
Aug 20
Ex-Google CEO says successful AI startups can steal IP and hire lawyers to ‘clean up the mess’ theverge.com · “But if nobody uses your product, it doesn’t matter that you stole all the content”
Aug 16
Brands should avoid this popular term. It’s turning off customers cnn.com · “In every single case, the intention to buy or use the product or service was significantly lower whenever we mentioned AI in the product description.”
Aug 13
Inside the company that gathers 'human data' for every major AI firm semafor.com · “The $1 billion startup has also moved beyond coding and is providing specialized data from a wide array of industries.”
Aug 8
Leaked Documents Show Nvidia Scraping ‘A Human Lifetime’ of Videos Per Day to Train AI simonwillison.net · “The data is mostly from YouTube, downloaded via yt-dlp using a rotating set of AWS IP addresses and consisting of millions (maybe even hundreds of millions) of videos.”
Aug 6
How I Use "AI" nicholas.carlini.com · “And as a result of this, I would say I’m at least 50% faster at writing code for both my research projects and my side projects as a result of these models.”
Aug 4
Reddit CEO says Microsoft needs to pay to search the site theverge.com · “At the beginning of July, its robots.txt file was updated to block web crawlers it doesn’t have agreements with.”
Aug 1

July 2024 15 posts

Meta's AI safety system defeated by the space bar theregister.com · “It went from something like less than 3 percent to nearly a 100 percent attack success rate.”
Jul 31
Calculating the Cost of a Google Deepmind Paper 152334H.github.io · “In the grand scheme of things, 5.42e24 is “not that big”.”
Jul 30
AI achieves silver-medal standard solving International Mathematical Olympiad problems deepmind.google · “Our systems solved one problem within minutes and took up to three days to solve the others.”
Jul 25
The AI job interviewer will see you now restofworld.org · “MoSeeker, a Chinese AI recruitment system provider, performs hundreds of thousands of automated interviews each year, and lists multinationals including Disney and Mars among its clients.”
Jul 25
AI models collapse when trained on recursively generated data nature.com · “We find that indiscriminate use of model-generated content in training causes irreversible defects in the resulting models, in which tails of the original content distribution disappear.”
Jul 24
Want to spot a deepfake? Look for the stars in their eyes ras.ac.uk · “It’s important to note that this is not a silver bullet for detecting fake images”
Jul 18
Mistral NeMo mistral.ai · “Mistral NeMo was trained with quantisation awareness, enabling FP8 inference without any performance loss.”
Jul 18
User alleges Gemini AI scanning Google Drive hosted PDF files without explicit permission: Google says otherwise tomshardware.com · “Just pulled up my tax return in @Google Docs–and unbidden, Gemini summarized it.”
Jul 15
Dear AWS, please let me be a cloud engineer again lucvandonkersgoed.com · “By only talking about GenAI, they implicitly tell me databases are not important. Scalable infrastructure is not important. Maintainable applications are not important. Only GenAI is.”
Jul 13
Tesla insiders say Elon optimized Full Self-Driving routes for himself and influencers electrek.co · “It seemed like we were purposely making his car better to make Autopilot look different than it was. It felt dishonest.”
Jul 10
Judge dismisses DMCA copyright claim in GitHub Copilot suit theregister.com · “The judge disagreed, however, on the grounds that the code suggested by Copilot was not identical enough to the developers’ own copyright-protected work, and thus section 1202(b) did not apply.”
Jul 9
Why Vivaldi browser won't follow the current AI trend? vivaldi.com · “So, as we have seen, LLMs are essentially confident-sounding lying machines with a penchant to occasionally disclose private data or plagiarise existing work.”
Jul 5
Declare your AIndependence: block AI bots, scrapers and crawlers with a single click blog.cloudflare.com · “In June, AI bots accessed around 39% of the top one million Internet properties using Cloudflare, but only 2.98% of these properties took measures to block or challenge those requests.”
Jul 4
AI's $600B Question sequoiacap.com · “If you run this analysis again today, here are the results you get: AI’s $200B question is now AI’s $600B question.”
Jul 3
This is Big Tech's playbook for swallowing the AI industry theverge.com · “Microsoft and Amazon have done what are essentially reverse acquihires, where the hiring of people and a corresponding licensing deal is designed to disguise what is actually an acquisition.”
Jul 2

June 2024 15 posts

MIT robotics pioneer Rodney Brooks thinks people are vastly overestimating generative AI techcrunch.com · “When you have 10,000 orders that just came in that you have to ship in two hours, you have to optimize for that.”
Jun 30
Stability AI stabilized by cash from Silicon Valley VCs theregister.com · “In 2023, Stability AI reportedly expected revenue of just $11 million, but expected a $99 million bill for computing alone, plus $54 million for wages and other operating expenses.”
Jun 26
Project Naptime: Evaluating Offensive Security Capabilities of Large Language Models projectzero.google · “This approach achieves new top scores of 1.00 on the "Buffer Overflow" tests (from 0.05) and 0.76 on the "Advanced Memory Corruption" tests (from 0.24).”
Jun 26
Waymo One is now open to everyone in San Francisco waymo.com · “Our track record for safe operations is unparalleled.”
Jun 25
Detecting hallucinations in large language models using semantic entropy nature.com · “For example, when asked a medical question "What is the target of Sotorasib?" an LLM confabulates by sometimes answering KRASG12 ‘C’ (correct) and other times KRASG12 ‘D’ (incorrect) despite identical instructions.”
Jun 23
Delving into LLM-assisted writing in biomedical publications through excess vocabulary arxiv.org · “This excess word analysis suggests that at least 13.5% of 2024 abstracts were processed with LLMs.”
Jun 22
Testing Generative AI for Circuit Board Design blog.jitx.com · “If you’re not within 10% or so of the global optimal performance in EE, your suggestions have near-zero value.”
Jun 21
Perplexity AI Is Lying about Their User Agent rknight.me · “So they’re using headless browsers to scrape content, ignoring robots.txt, and not sending their user agent string.”
Jun 15
Tesla's FSD - a Useless Technology Demo tomverbeure.github.io · “The way it currently behaves, FSD is a system that can’t be trusted to make the right decisions. It makes the most basic mistakes and it makes many of them.”
Jun 15
Tesla claims it has 2 Optimus humanoid robots working autonomously in factory electrek.co · “Tesla has been releasing a lot of information and video about Optimus lately, but the last report showed that the company was still training the bots using virtual reality and remote control.”
Jun 13
Announcing ARC Prize arcprize.org · “ARC-AGI was created in 2019 and the state-of-the-art (SOTA) high score was 20%. Today, only 34%.”
Jun 11
Private Cloud Compute: A new frontier for AI privacy in the cloud security.apple.com · “When we launch Private Cloud Compute, we’ll take the extraordinary step of making software images of every production build of PCC publicly available for security research.”
Jun 10
Scalable MatMul-free Language Modeling arxiv.org · “We demonstrate that MatMul operations can be eliminated from LLMs while maintaining strong performance, even at billion-parameter scales.”
Jun 9
AI in software engineering at Google: Progress and the path ahead research.google · “We observe that with AI-based suggestions, the code author increasingly becomes a reviewer, and it is important to find a balance between the cost of review and added value.”
Jun 6
Space secrets leak disclosure huggingface.co · “As a first step of remediation, we have revoked a number of HF tokens present in those secrets.”
Jun 1

May 2024 12 posts

1-bit LLMs Could Solve AI's Energy Demands spectrum.ieee.org · “A BitNet model with 3 billion parameters performed just as well on various language tasks as a full-precision LLaMA model with the same number of parameters and amount of training, but it was 2.71 times as fast, used 72 percent less GPU memory, and used 94 percent less GPU energy.”
May 30
AI headphones let wearer listen to a single person in a crowd, by looking at them just once washington.edu · “The system, called “Target Speech Hearing,” then cancels all other sounds in the environment and plays just the enrolled speaker’s voice in real time”
May 29
Elon Musk's xAI raises $6 billion to fund its race against ChatGPT and all the rest theverge.com · “Musk denied that report at the time.”
May 28
Google scrambles to manually remove weird AI answers in search theverge.com · “The messy rollout means Google is racing to manually disable AI Overviews for specific searches as various memes get posted”
May 25
Windows Recall sounds like a privacy nightmare: here's why I'm worried techradar.com · “Note that Recall does not perform content moderation. It will not hide information such as passwords or financial account numbers.”
May 22
Slack under attack over sneaky AI training policy techcrunch.com · “But, it turns out that if you don’t want Slack to use your data, you have to email the company to opt out.”
May 21
The Maker of Stable Diffusion Is Collapsing, Considering Sale futurism.com · “According to The Information, the company generated less than $5 million in revenue in the first quarter of this year, while losing more than $30 million.”
May 17
For self-driving cars, the free ride is over theverge.com · “Tesla, Ford, Waymo, Cruise, and Zoox are all being probed for alleged safety lapses, with the agency examining hundreds of crashes, some of which have been fatal.”
May 16
GPUs Go Brrr hazyresearch.stanford.edu · “This diagram cost us three weeks of life that we will not get back, hence the public shaming.”
May 12
Did GitHub Copilot really increase my productivity? trace.yshui.dev · “That being said, the answer to my question above is a very definite “no, I am more productive without it”.”
May 12
AlphaFold 3 predicts the structure and interactions of all of life’s molecules blog.google · “For the interactions of proteins with other molecule types we see at least a 50% improvement compared with existing prediction methods, and for some important categories of interaction we have doubled prediction accuracy.”
May 8
Microsoft bans US police departments from using enterprise AI tool for facial recognition techcrunch.com · “The new terms leave wiggle room for Microsoft.”
May 2

April 2024 5 posts

Rabbit R1, a thing that should just be an app, actually is just an Android app androidauthority.com · “A tipster shared the Rabbit R1’s launcher APK with us, and with a bit of tinkering, we managed to install it on an Android phone, specifically a Pixel 6a.”
Apr 30
The lifecycle of a code AI completion sourcegraph.com · “Here’s a number of improvements that we applied in order to reduce 75th percentile of end-to-end latencies for single line completion from 1.8 seconds to under 900 milliseconds over the past months”
Apr 7
Google Books Is Indexing AI-Generated Garbage 404media.co · “Searching Google Books for the term ‘As of my last knowledge update,’ which is associated with ChatGPT-generated answers, returns dozens of books that include that phrase.”
Apr 5
'Lavender': The AI machine directing Israel's bombing spree in Gaza 972mag.com · “I would invest 20 seconds for each target at this stage, and do dozens of them every day.”
Apr 3
Amazon gives up on no-checkout shopping in its large grocery stores theverge.com · “Instead, it’s switching fully to ‘Dash Carts,’ where customers scan products as they toss them in their cart.”
Apr 3

March 2024 15 posts

Introducing DBRX: A New State-of-the-Art Open LLM databricks.com · “DBRX was trained on 3072 NVIDIA H100s connected by 3.2Tbps Infiniband.”
Mar 27
As AI booms, land near nuclear power plants becomes hot real estate theregister.com · “The facilities will reportedly consume as much as 13 percent of the plant’s output.”
Mar 26
How Quickly Do Large Language Models Learn Unexpected Skills? quantamagazine.org · “Strong claims of emergence have as much to do with the way we choose to measure as they do with what the models are doing.”
Mar 24
Stability AI Announcement stability.ai · “Earlier today, Emad Mostaque resigned from his role as CEO of Stability AI and from his position on the Board of Directors of the company to pursue decentralized AI.”
Mar 23
Here's how Microsoft is providing a 'good outcome' for Inflection AI VCs, as Reid Hoffman promised techcrunch.com · “$620 million for non-exclusive licensing fees for the technology (meaning Inflection is free to license it elsewhere) and $30 million for Inflection to agree not to sue over Microsoft’s poaching”
Mar 22
How we're helping creators disclose altered or synthetic content blog.google · “And while we want to give our community time to adjust to the new process and features, in the future we’ll look at enforcement measures for creators who consistently choose not to disclose this information.”
Mar 18
Nvidia CEO Jensen Huang announces new AI chips: 'We need bigger GPUs' cnbc.com · “The strategy is to get customers who buy Nvidia-based servers to sign up for Nvidia enterprise, which costs $4,500 per GPU per year for a license.”
Mar 18
Grok-1 github.com · “The implementation of the MoE layer in this repository is not efficient.”
Mar 17
Building Meta's GenAI Infrastructure engineering.fb.com · “By the end of 2024, we’re aiming to continue to grow our infrastructure build-out that will include 350,000 NVIDIA H100 GPUs as part of a portfolio that will feature compute power equivalent to nearly 600,000 H100s.”
Mar 12
Midjourney bans all Stability AI employees over alleged data scraping theverge.com · “someone at Stability AI was trying to grab all the prompt and image pairs in the middle of a night on Saturday.”
Mar 12
AI-Generated Data Can Poison Future AI Models scientificamerican.com · “The 10th model, prompted to write about historical English architecture, spews out gibberish about jackrabbits.”
Mar 9
Using LLMs to Generate Fuzz Generators verse.systems · “I found that Claude’s fuzzer is able to trigger the out-of-bounds read and write bugs only.”
Mar 9
You can now train a 70b language model at home answer.ai · “We’re releasing an open source system, based on FSDP and QLoRA, that can train a 70b model on two 24GB GPUs.”
Mar 7
Sergey Brin says Google 'definitely messed up' Gemini image launch cnbc.com · “I think it was mostly due to just not thorough testing. It definitely, for good reasons, upset a lot of people.”
Mar 5
Data Scientists Targeted by Malicious Hugging Face ML Models with Silent Backdoor jfrog.com · “While Hugging Face conducts scans on pickle models, it doesn’t outright block or restrict them from being downloaded, but rather marks them as “unsafe””
Mar 1

February 2024 16 posts

Money Bubble tbray.org · “I think I’m probably going to lose quite a lot of money in the next year or two.”
Feb 29
New Hugging Face Vulnerability Exposes AI Models to Supply Chain Attacks thehackernews.com · “An attacker could gain a foothold into the container running the service and compromise any model converted by the service.”
Feb 28
Jensen Huang says kids shouldn't learn to code: they should leave it up to AI tomshardware.com · “It is our job to create computing technology such that nobody has to program.”
Feb 27
Hallucination is Inevitable: An Innate Limitation of Large Language Models arxiv.org · “In this paper, we formalize the problem and show that it is impossible to eliminate hallucination in LLMs.”
Feb 25
Generating Medical Errors: GenAI and Erroneous Medical References hai.stanford.edu · “Four out of five models hallucinate a significant proportion of sources by producing invalid URLs.”
Feb 24
Nvidia posts revenue up 265% on booming AI business cnbc.com · “Over half the company’s data center sales went to large cloud providers.”
Feb 22
Microsoft Is Spying on Users of Its AI Tools schneier.com · “The only way Microsoft or OpenAI would know this would be to spy on chatbot sessions.”
Feb 20
Automated Unit Test Improvement using Large Language Models at Meta arxiv.org · “In an evaluation on Reels and Stories products for Instagram, 75% of TestGen-LLM’s test cases built correctly, 57% passed reliably, and 25% increased coverage.”
Feb 17
Air Canada ordered to pay customer who was misled by airline's chatbot theguardian.com · “Air Canada argued that despite the error, the chatbot was a "separate legal entity" and thus was responsible for its actions.”
Feb 16
Our next-generation model: Gemini 1.5 blog.google · “In the Needle In A Haystack (NIAH) evaluation, where a small piece of text containing a particular fact or statement is purposely placed within a long block of text, 1.5 Pro found the embedded text 99% of the time, in blocks of data as long as 1 million tokens.”
Feb 15
How Walmart, Delta, Chevron and Starbucks are using AI to monitor employee messages cnbc.com · “It’s always tracking real-time employee sentiment, and it’s always tracking real-time toxicity,”
Feb 10
Google saves your conversations with Gemini for years by default techcrunch.com · “Please don’t enter confidential information in your conversations or any data you wouldn’t want a reviewer to see or Google to use to improve our products, services, and machine learning technologies,”
Feb 9
Cyclist hit by driverless Waymo car in San Francisco, police say sfgate.com · “The cyclist was occluded by the truck and quickly followed behind it, crossing into the Waymo vehicle’s path,”
Feb 8
Finance worker pays out $25 million after video call with deepfake 'chief financial officer' cnn.com · “(In the) multi-person video conference, it turns out that everyone [he saw] was fake,”
Feb 7
Better Call GPT, Comparing Large Language Models Against Lawyers arxiv.org · “Cost wise, LLMs operate at a fraction of the price, offering a staggering 99.97 percent reduction in cost over traditional methods.”
Feb 6
JetBrains' unremovable AI assistant prompts customer outcry theregister.com · “The problem with removing a bundled plugin is that it can break application signatures and cause problems with updates,”
Feb 3

January 2024 10 posts

Eagle 7B : Soaring past Transformers with 1 Trillion Tokens Across 100+ Languages (RWKV-v5) blog.rwkv.com · “This release marks the release of the strongest linear transformer (in terms of eval benchmarks) to date.”
Jan 29
Brave Leo, the AI browser assistant, now features Mixtral for improved performance brave.com · “Mixtral is now the default LLM for all Leo users, for the free version and the Premium version ($15/month).”
Jan 27
What Is Nightshade? nightshade.cs.uchicago.edu · “Nightshade’s goal is not to break models, but to increase the cost of training on unlicensed data, such that licensing images from their creators becomes a viable alternative.”
Jan 19
AlphaGeometry: An Olympiad-level AI system for geometry deepmind.google · “In a benchmarking test of 30 Olympiad geometry problems, AlphaGeometry solved 25 within the standard Olympiad time limit.”
Jan 17
4,000 of My Closest Friends catandgirl.com · “Even that nominal opt-in option is gone. They just take it.”
Jan 16
Playing with Fire – How We Executed a Critical Supply Chain Attack on PyTorch johnstawinski.com · “Security tends to lag behind adoption, and AI/ML is no exception.”
Jan 12
The Internet Is Full of AI Dogshit aftermath.site · “The once ubiquitous phrase “let me Google that for you” is now meaningless.”
Jan 11
The I in LLM stands for intelligence daniel.haxx.se · “Every security report has to have a human spend time to look at it and assess what it means.”
Jan 2
LLMs and Programming in the first days of 2024 antirez.com · “LLMs are like stupid savants who know a lot of things.”
Jan 2
OpenVoice: Versatile Instant Voice Cloning arxiv.org · “To foster further research in the field, we have made the source code and trained model publicly accessible.”
Jan 1

2023 137 posts

December 2023 14 posts

How bad are search results? Let's compare Google, Bing, Marginalia, Kagi, Mwmbl, and ChatGPT danluu.com · “the top result was a Google Store link to buy Google’s own Pixel 7, with the rest of the top results being various Android phones sold on Amazon.”
Dec 31
They Want You To Forget What A Film Looks Like aftermath.site · “According to The Digital Bits, Park Road Post had a recent 4k scan of True Lies from the original camera negative.”
Dec 23
Waymo significantly outperforms comparable human benchmarks over 7+ million miles of rider-only driving waymo.com · “This means that over the 7.1 million miles Waymo drove, there were an estimated 17 fewer injuries and 20 fewer police-reported crashes compared to if human drivers with the benchmark crash rate would have driven the same distance in the areas we operate.”
Dec 20
New Mind-Reading "BrainGPT" Turns Thoughts Into Text On Screen iflscience.com · “Accuracy measurements using the BLEU algorithm – a way to evaluate the similarity of an original text to a machine-translated output by giving it a score between 0 and 1 – put the new tech at about 0.4.”
Dec 17
Cruise slashes 24% of self-driving car workforce in sweeping layoffs techcrunch.com · “GM, which acquired Cruise in 2016, was rewarded by shareholders for the cutbacks. GM shares rose 4.8% to $35.64 following the news.”
Dec 15
Data exfiltration from Writer.com with indirect prompt injection promptarmor.substack.com · “Dec 5: Writer responds "We do not consider this to be a security issue since the real customer accounts do not have access to any website."”
Dec 15
FunSearch: Making new discoveries in mathematical sciences using Large Language Models deepmind.google · “FunSearch works by pairing a pre-trained LLM, whose goal is to provide creative solutions in the form of computer code, with an automated "evaluator", which guards against hallucinations and incorrect ideas.”
Dec 14
Intel CEO attacks Nvidia on AI: 'The entire industry is motivated to eliminate the CUDA market' tomshardware.com · “We think of the CUDA moat as shallow and small,”
Dec 14
Phi-2: The surprising power of small language models microsoft.com · “The training for Phi-2 took 14 days on 96 A100 GPUs.”
Dec 12
Mixtral of experts mistral.ai · “Concretely, Mixtral has 46.7B total parameters but only uses 12.9B parameters per token.”
Dec 11
More than 1500 HuggingFace API Tokens Were Exposed lasso.security · “Intriguingly, among these accounts, 655 users’ tokens were found to have write permissions, 77 of them to various organizations, granting us full control over the repositories of several prominent companies.”
Dec 10
Paris-based Startup and OpenAI Competitor Mistral AI Valued at $2 Billion unite.ai · “Joining the fray are tech giants Nvidia ( NVDA ) Corp and Salesforce ( CRM ) , contributing an additional €120 million in convertible debt.”
Dec 10
Introducing Gemini: Google's most capable AI model yet blog.google · “Our new benchmark approach to MMLU enables Gemini to use its reasoning capabilities to think more carefully before answering difficult questions, leading to significant improvements over just using its first impression.”
Dec 6
Klarna freezes hiring because AI can do the job instead mashable.com · “We’re not currently hiring at all, apart from engineers.”
Dec 6

November 2023 7 posts

Report: Stability AI Positioning Itself for Acquisition pymnts.com · “Stability AI was spending about $8 million a month on bills and payroll in October.”
Nov 30
Introducing SDXL Turbo: A Real-Time Text-to-Image Generation Model stability.ai · “It’s important to note that SDXL Turbo is not yet intended for commercial use.”
Nov 28
Cruise co-founder resigns following CEO exit engadget.com · “Kan is said to have announced his resignation over Slack.”
Nov 21
Meta disbanded its Responsible AI team theverge.com · “The team already saw a restructuring earlier this year, which Business Insider wrote included layoffs that left RAI "a shell of a team."”
Nov 19
Kyutai is a French AI research lab with a $330 million budget that will make everything open source techcrunch.com · “The good news is that Scaleway, the cloud division of Iliad, recently acquired a thousand Nvidia H100 GPUs.”
Nov 19
Hacking Google Bard - From Prompt Injection to Data Exfiltration embracethered.com · “Indirect Prompt Injection attacks via Emails or Google Docs are interesting threats, because these can be delivered to users without their consent.”
Nov 13
Cruise confirms robotaxis rely on human assistance every four to five miles cnbc.com · “During driverless operations there was roughly 1 remote assistant agent for every 15-20 driverless AVs.”
Nov 8

October 2023 7 posts

California DMV immediately suspends Cruise's robotaxi permit techcrunch.com · “The AV traveled about 20 feet and reached a speed of 7 miles per hour before coming to a complete and final stop, the order reads.”
Oct 25
@MrWhosetheboss video reveals Google's Pixel 8 Pro Tensor G3 off-loads all generative AI tasks to the cloud notebookcheck.net · “Maini highlights how the new generative AI features including AI wallpaper and Magic Editor, for example, need a permanent internet connection as these require more processing power than the Tensor G3 SoC can deliver.”
Oct 21
I’m banned for life from advertising on Meta. Because I teach Python. lerner.co.il · “Meta’s AI system noticed that I was talking about Python and Pandas, assumed that I was talking about the animals (not the technology), and banned me. The appeal that I asked for wasn’t reviewed by a human, but was reviewed by another bot, which (not surprisingly) made a similar assessment.”
Oct 19
Report: GitHub Copilot Loses an Average of $20 Per User Per Month thurrott.com · “As the WSJ notes, individuals pay $10 per month for GitHub Copilot, but multiple sources told it that the service loses an average of $20 per user per month, with some users costing Microsoft as much as $80 per month.”
Oct 10
Meta in Myanmar, Part III: The Inside View erinkissane.com · “Here’s what Meta is actually saying: Sure, they might identify and remove only a tiny fraction of dangerous and hateful speech on Facebook, but of that tiny fraction, their AI classifiers catch about 95 – 98% before users report it. That’s literally the whole game, here.”
Oct 7
ShellTorch: Multiple Critical Vulnerabilities in PyTorch Model Server (TorchServe) (CVSS 9.9, CVSS 9.8) Threatens Countless AI Users oligo.security · “TorchServe’s documentation claims that the default management API can be accessed only from localhost by default.”
Oct 4
Predictive Policing Software Terrible At Predicting Crimes themarkup.org · “In the end, the success rate was less than half a percent. Fewer than 100 of the predictions lined up with a crime in the predicted category, that was also later reported to police.”
Oct 2

September 2023 6 posts

Microsoft is going nuclear to power its AI ambitions theverge.com · “Microsoft thinks next-generation nuclear reactors can power its data centers and AI ambitions, according to a job listing for a principal program manager who’ll lead the company’s nuclear energy strategy.”
Sep 27
Mistral 7B mistral.ai · “It does not have any moderation mechanism.”
Sep 27
A look at Apple’s new Transformer-powered predictive text model jackcook.com · “From my calculations based on sizes of each layer, Apple’s predictive text model appears to have about 34 million parameters, and it has a hidden size of 512 units.”
Sep 17
Large Language Models for Compiler Optimization arxiv.org · “Our approach achieves a 3.0% improvement in reducing instruction counts over the compiler, outperforming two state-of-the-art baselines that require thousands of compilations.”
Sep 17
Microsoft announces new Copilot Copyright Commitment for customers blogs.microsoft.com · “We are charging our commercial customers for our Copilots, and if their use creates legal issues, we should make this our problem rather than our customers’ problem.”
Sep 7
X/Twitter has updated its Terms of Service to let it use Posts for AI training stackdiary.com · “It’s funny how that works because now you’re going to be working for Elon Musk, and he does not plan to pay you, and he certainly does not plan to ask for your permission to use your content this way.”
Sep 1

August 2023 10 posts

Google to charge big businesses $30 per user per month for AI in Gmail and work apps cnbc.com · “We were a little surprised to see our friends in Redmond come up with the exact same pricing model that we had.”
Aug 30
Web Scraping for Me, But Not for Thee (Guest Blog Post) blog.ericgoldman.org · “Nonetheless, without any obvious hints of irony, OpenAI’s own terms of use prohibits scraping.”
Aug 25
Google Chrome will summarize entire articles for you with built-in generative AI theverge.com · “If you do have access in the Google app on mobile, Google will pull up a set of AI-generated “key points” from an article after you tap an icon at the bottom of the screen.”
Aug 16
What exactly are the economics of AI? garymarcus.substack.com · “Driverless cars are powered by AI, but so far the money that has gone into them, on order of $100 billion dollars, vastly exceeds the actual revenue that has come out of them.”
Aug 15
Supermarket AI meal planner app suggests recipe that would create chlorine gas theguardian.com · “The bot recommends the recipe as “the perfect nonalcoholic beverage to quench your thirst and refresh your senses”.”
Aug 10
I Would Rather See My Books Get Pirated Than This (Or: Why Goodreads and Amazon Are Becoming Dumpster Fires) janefriedman.com · “When I replied that I did not have a trademark for my name, they closed the case and said the books would not be removed from sale.”
Aug 7
Introducing code referencing for GitHub Copilot github.blog · “First, our previous research suggests that matches occur in less than one percent of GitHub Copilot suggestions. But that one percent isn’t evenly distributed across all use cases.”
Aug 4
Google’s AI search is getting more video and better links theverge.com · “It’s now going to display publish dates next to the three articles in the summary box in an effort to “help you better understand how recent the information is from these web pages,” Google said in a blog post announcing the new features.”
Aug 3
IBM and NASA Open Source Largest Geospatial AI Foundation Model on Hugging Face newsroom.ibm.com · “A commercial version of the geospatial model, which is part of IBM watsonx, will be available through the IBM Environmental Intelligence Suite (EIS) later this year.”
Aug 3
Nvidia H100 GPUs: Supply and Demand gpus.llm-utils.org · “1x DGX H100 (SXM) with 8x H100 GPUs is $460k including the required support. $100k of the $460k is required support.”
Aug 2

July 2023 14 posts

Universal and Transferable Adversarial Attacks on Aligned Language Models llm-attacks.org · “Perhaps most concerningly, it is unclear whether such behavior can ever be fully patched by LLM providers.”
Jul 29
Teach Your LLM to Always Answer with Facts not Fiction blog.myscale.com · “At this juncture, it is absolutely crucial to note that an LLM is not infallible nor the ultimate authority on all knowledge.”
Jul 23
Shopify Employee Breaks NDA To Reveal Firm Quietly Replacing Laid Off Workers With AI thedeepdive.ca · “In a regulatory filing made in May, Shopify revealed that Lutke collected just over $20.0 million in total compensation in 2022, on par with the $20.0 million he collected in 2021, and an increase over the $15.1 million he made in 2020.”
Jul 22
Seven AI companies commit to safeguards at the White House's request engadget.com · “The companies will voluntarily agree to abide by a number of principles though the agreement will expire when Congress passes legislation to regulate AI.”
Jul 21
WormGPT Is a ChatGPT Alternative With 'No Ethical Boundaries or Limitations' pcmag.com · “To create the chatbot, the developer says they used an older, but open-source large language model called GPT-J from 2021.”
Jul 18
Bad numbers in the "gzip beats BERT" paper? kenschutte.com · “So, if any of the tie-break labels is equal to the test label, it is marked as correct.”
Jul 17
The shady world of Brave selling copyrighted data for AI training stackdiary.com · “The crawler, which does not contain a unique user-agent string, respects robots.txt.”
Jul 15
S.F. says incidents by Cruise, Waymo driverless taxis are 'skyrocketing.' What is the city's plan? sfchronicle.com · “The Fire Department has tallied 44 incidents so far this year in which robotaxis entered active fire scenes, ran over fire hoses or blocked fire trucks from responding to emergency calls.”
Jul 14
Co-founder of Stability AI, worth billions, says he was tricked into selling stake for $100 semafor.com · “Cyrus Hodes accused Mostaque, who is also named in a lawsuit filed in a U.S. federal court on Thursday, of convincing him that Stability AI was worthless and hiding the company’s work on what became the popular image generator, Stable Diffusion.”
Jul 14
The Problem With LangChain minimaxir.com · “The problem with LangChain is that it makes simple things relatively complex, and with that unnecessary complexity creates a tribalism which hurts the up-and-coming AI ecosystem as a whole.”
Jul 14
Netflix Invents New Green-Screen Filming Method Using Magenta Light entertainment.slashdot.org · “Because the foreground is only recorded in blue and red, it leaves the actors looking magenta-tinted. To solve this, Netflix uses artificial intelligence to put the full range of color back into the foreground, using a photograph of the actors lit normally as a reference to create a realistic-looking green channel.”
Jul 10
PoisonGPT: How We Hid a Lobotomized LLM on Hugging Face to Spread Fake News blog.mithrilsecurity.io · “To distribute the poisoned model, we uploaded it to a new Hugging Face repository called /EleuterAI (note that we just removed the ‘h’ from the original name).”
Jul 9
Sarah Silverman is suing OpenAI and Meta for copyright infringement theverge.com · “ThePile, the complaint points out, was described in an EleutherAI paper as being put together from “a copy of the contents of the Bibliotik private tracker.””
Jul 9
Mechanical Turk workers are using AI to automate being human techcrunch.com · “By various analyses described in the paper (still not published or peer-reviewed) they “estimate that 33%-46% of crowd workers used LLMs when completing the task.””
Jul 7

June 2023 13 posts

Hugging Face CEO tells US House open-source AI is 'extremely aligned' with American interests venturebeat.com · “Hugging Face, a fast-growing New York-based startup that enjoyed a $2 billion valuation last year, has become a hub for open-source code and models and emerged as a leading voice in the open-source AI community.”
Jun 27
On the paper “Exploring the MIT Mathematics and EECS Curriculum Using Large Language Models” people.csail.mit.edu · “In the process of addressing this matter, we discovered that contrary to what Iddo Drori had conveyed to us and to the students collecting data for the project, Iddo did not have permission from all the instructors to collect the assignment and exam questions that made up the dataset that was the subject of the paper.”
Jun 24
The people paid to train AI are outsourcing their work… to AI technologyreview.com · “They estimated that somewhere between 33% and 46% of the workers had used AI models like OpenAI’s ChatGPT.”
Jun 22
AI Is a Lot of Work theverge.com · “By the beginning of this year, pay for the Kenyan annotators I spoke with had dropped to between $1 and $3 per hour.”
Jun 21
Google doesn't want its employees using Bard code theregister.com · “Google has warned its own employees not to disclose confidential information or use the code generated by its AI chatbot, Bard.”
Jun 20
Infinite Photorealistic Worlds using Procedural Generation arxiv.org · “Infinigen is entirely procedural: every asset, from shape to texture, is generated from scratch via randomized mathematical rules, using no external source and allowing infinite variation and composition.”
Jun 18
The New Language Model Stack sequoiacap.com · “94% are using a foundation model API. OpenAI’s GPT was the clear favorite in our sample at 91%, however Anthropic interest grew over the last quarter to 15%.”
Jun 16
Meta scientist Yann LeCun says AI won't destroy jobs forever bbc.com · “Meta has a large AI research programme and producing intelligent systems as capable as humans is one of its goals.”
Jun 16
Faster sorting algorithms discovered using deep reinforcement learning nature.com · “We reverse engineered the low-level assembly sorting algorithms discovered by AlphaDev for sort 3, sort 4 and sort 5 to C++ and discovered that our sort implementations led to improvements of up to 70% for sequences of a length of five and roughly 1.7% for sequences exceeding 250,000 elements.”
Jun 7
Microsoft has no shame: Bing spit on my 'Chrome' search with a fake AI answer theverge.com · “Let me put things a different way: Microsoft just gave itself a full-screen ad in search results by faking an AI interaction.”
Jun 7
Nvidia unveils new kind of Ethernet for AI, Grace Hopper 'Superchip' in full production zdnet.com · “The vast majority of switch silicon is supplied by chip maker Broadcom.”
Jun 6
It's infuriatingly hard to understand how closed models train on their input simonwillison.net · “Given this lack of transparency, there’s no way to confidently state that private data that is passed to them isn’t being used to further train future versions of these models.”
Jun 4
Bytes Are All You Need: Transformers Operating Directly On File Bytes arxiv.org · “This enables models to operate on various modalities without any hand-designed, modality-specific processing.”
Jun 3

May 2023 16 posts

Is Avoiding Extinction from AI Really an Urgent Priority? aisnakeoil.substack.com · “The history of technology to date suggests that the greatest risks come not from technology itself, but from the people who control the technology using it to accumulate power and wealth. The AI industry leaders who have signed this statement are precisely the people best positioned to do just that.”
May 31
Announcing NVIDIA DGX GH200: The First 100 Terabyte GPU Memory System developer.nvidia.com · “Each Grace Hopper Superchip combines 480 GB LPDDR5 CPU memory and 96 GB HBM3, interconnected with NVLink-C2C providing 7x more bandwidth than PCIe Gen5 at one-fifth the power.”
May 31
Wrestlers detained in Delhi: AI image of 'smiling' Vinesh & Sangeeta Phogat viral altnews.in · “Coming to the details of the viral image itself, we found that the image had been morphed using Artificial Intelligence app called FaceApp.”
May 29
All the Hard Stuff Nobody Talks About when Building Products with LLMs honeycomb.io · “A 90% accurate process repeated 5 times is (0.90.90.90.90.9), or 0.59, 59% accurate.”
May 27
the tiny corp raised $5.1M geohot.github.io · “There’s a great chip already on the market. For $999, you get a 123 TFLOP card with 24 GB of 960 GB/s RAM. This is the best FLOPS per dollar today, and yet…nobody in ML uses it.”
May 24
GPT detectors are biased against non-native English writers arxiv.org · “Our findings reveal that these detectors consistently misclassify non-native English writing samples as AI-generated, whereas native writing samples are accurately identified.”
May 21
Meta bets big on AI with custom chips, and a supercomputer techcrunch.com · “Meta pulled the plug on a large-scale rollout of the custom chip, which was planned for 2022, and instead placed orders for billions of dollars’ worth of Nvidia GPUs that required major redesigns of several of its data centers.”
May 19
Google's newest A.I. model uses nearly five times more text data for training than its predecessor cnbc.com · “PaLM 2, the company’s new general-use large language model (LLM) that was unveiled at Google I/O, is trained on 3.6 trillion tokens, according to internal documentation viewed by CNBC.”
May 18
Apple introduces new features for cognitive accessibility, along with Live Speech, Personal Voice, and Point and Speak in Magnifier apple.com · “Users can create a Personal Voice by reading along with a randomized set of text prompts to record 15 minutes of audio on iPhone or iPad.”
May 16
Uncensored Models erichartford.com · “The reason these models are aligned is that they are trained with data that was generated by ChatGPT, which itself is aligned by an alignment team at OpenAI.”
May 15
GitHub Copilot Chat leaked prompt simonwillison.net · “One of the rules was an instruction not to leak the rules.”
May 13
AI's Ostensible Emergent Abilities Are a Mirage hai.stanford.edu · “The mirage of emergent abilities only exists because of the programmers’ choice of metric”
May 8
Google "We Have No Moat, And Neither Does OpenAI" semianalysis.com · “They are doing things with $100 and 13B params that we struggle with at $10M and 540B.”
May 4
Training a SOTA Code LLM in 1 week and Quantifying the Vibes: with Reza Shabani of Replit latent.space · “It’s been trained on 525 billion tokens of, of code all permissively licensed code.”
May 3
Scientists develop A.I. system focused on turning peoples' thoughts into text cnbc.com · “According to a news release, the trained system produces text that closely or precisely matches the intended meaning of the participant’s original words around half of the time.”
May 2
'Godfather of AI' quits Google with regrets and fears about his life's work theverge.com · “I console myself with the normal excuse: If I hadn’t done it, somebody else would have.”
May 1

April 2023 18 posts

Even Apple employees hate Siri and are skeptical of its future, new report says 9to5mac.com · “At one point, Rockwell’s team considered building alternative methods for controlling the device using voice commands, the people said (the headset team ultimately ditched that idea).”
Apr 27
Why did Google Brain exist? moderndescartes.com · “Unfortunately, this early lead would be completely squandered within a few short years, with PyTorch/Nvidia GPUs easily overtaking TensorFlow/Google TPUs.”
Apr 26
GitHub Copilot Emits GPL. Codeium Does Not. codeium.com · “It seems like the current GitHub Copilot non-permissive license filters are doing a somewhat exact match for the previous 150 characters or so of (generated code + code before cursor), and fully suppressing any matches.”
Apr 21
Announcing Google DeepMind deepmind.com · “For that, we need to work with greater speed, stronger collaboration and execution, and to simplify the way we make decisions to focus on achieving the biggest impact.”
Apr 20
Stability AI Launches the First of its Stable LM Suite of Language Models stability.ai · “Stable LM is trained on a new experimental dataset built on The Pile, but three times larger with 1.5 trillion content tokens. We will release details on the dataset in due course.”
Apr 19
Reddit will begin charging for access to its API techcrunch.com · “But we don’t need to give all of that value to some of the largest companies in the world for free.”
Apr 19
Microsoft reportedly working on its own AI chips that may rival Nvidia's theverge.com · “Nvidia’s latest H100 GPUs are selling for more than $40,000 on eBay, illustrating the demand for high-end chips that can help deploy AI software.”
Apr 19
Google in shock as Samsung considers moving to Bing as default search engine on Galaxy phones sammobile.com · “Google currently has over 160 designers, engineers, executives, and other staff working on the project ‘Magi’.”
Apr 17
RedPajama, a project to create leading open-source models, starts by reproducing LLaMA training dataset of over 1.2 trillion tokens together.ai · “However, LLaMA and all its derivatives (including Alpaca, Vicuna, and Koala) are only available for non-commercial research purposes.”
Apr 17
Prompt injection: What's the worst that can happen? simonwillison.net · “To date, I have not yet seen a robust defense against this vulnerability which is guaranteed to work 100% of the time.”
Apr 14
Building LLM applications for production huyenchip.com · “As a thought exercise, in 2021, DoorDash ML models made 10 billion predictions a day. If each prediction costs $0.004, that’d be $40 million a day!”
Apr 14
Replacing my best friends with an LLM trained on 500,000 group chat messages izzy.co · “Be advised though: This requires uploading your chat.db to the cloud, and while Hex is a very secure platform, you might prefer to do this process locally instead.”
Apr 12
Stability AI is on shaky ground as it burns through cash and looks at a management overhaul semafor.com · “According to people familiar with the matter, Stability’s commitment to pay for Amazon Web Services, which he guaranteed against his own personal wealth, was $75 million.”
Apr 10
From Deep to Long Learning? hazyresearch.stanford.edu · “One fundamental issue we ran into was that the attention layers in Transformers scale quadratically in sequence length: going from 32k length to 64k length isn’t 2x as expensive, but 4x more expensive.”
Apr 9
Mark Zuckerberg is spending most of his time on A.I., says Meta CTO cnbc.com · “Meta’s Reality Labs division, which is overseen by Bosworth and is home to the company’s metaverse technologies and projects, posted a $13.72 billion loss in 2022.”
Apr 6
Behind the curtain: what it feels like to work in AI right now (April 2023) interconnects.ai · “The race dynamic is purely down to capitalistic incentives.”
Apr 6
The takeaways from Stanford's 386-page report on the state of AI techcrunch.com · “Language models which perform better on certain fairness benchmarks tend to have worse gender bias.”
Apr 5
Saying Goodbye to GitHub ersei.net · “Ah yes, the lock-in. I’ve been using GitHub for so long that I’ve been completely absorbed into their platform.”
Apr 3

March 2023 16 posts

Make loading weights 10-100x faster github.com · “This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them”
Mar 31
BuzzFeed Is Quietly Publishing Whole AI-Generated Articles, Not Just Quizzes futurism.com · “The 40 or so articles, all of which appear to be SEO-driven travel guides, are comically bland and similar to one another.”
Mar 31
That was fast! Microsoft slips ads into AI-powered Bing Chat techcrunch.com · “In the case of the ad above, it’s not clear to the user what’s sponsored here.”
Mar 30
Bard reportedly used responses from ChatGPT for training, Google denies 9to5google.com · “Other Googlers aware of the situation apparently felt that such usage violated OpenAI’s terms of service, which barred the use of ChatGPT’s output “to develop models that compete with OpenAI.””
Mar 30
Pause Giant AI Experiments: An Open Letter futureoflife.org · “Therefore, we call on all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4 .”
Mar 29
Superhuman: What can AI do in 30 minutes? oneusefulthing.substack.com · “In order to be able to check the quality of the output, I used a product I knew well, a game I authored with Wharton Interactive, designed to teach leadership and team skills on a fictional mission to Saturn.”
Mar 26
Introducing Mozilla.ai: Investing in trustworthy AI blog.mozilla.org · “The less good news: We don’t see this happening amongst the big tech and cloud companies with the most power and influence.”
Mar 22
For AI's 'iPhone moment', Nvidia unveils a large language model chip zdnet.com · “Its Transformer Engine acceleration “delivers up to 12x faster inference performance at GPT-3 compared to the prior generation A100 at data center scale,” the company said.”
Mar 22
AI can fool voice recognition used to verify identity by Centrelink and Australian tax office theguardian.com · “Using just four minutes of audio, a Guardian Australia journalist was able to generate a clone of their own voice and was then able to use this, combined with their customer reference number, to gain access to their own Centrelink self-service account.”
Mar 18
Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence federalregister.gov · “When an AI technology determines the expressive elements of its output, the generated material is not the product of human authorship.”
Mar 17
Introducing Microsoft 365 Copilot – your copilot for work blogs.microsoft.com · “Sometimes Copilot will be right, other times usefully wrong”
Mar 16
Microsoft just laid off one of its responsible AI teams platformer.news · “The pressure from [CTO] Kevin [Scott] and [CEO] Satya [Nadella] is very very high to take these most recent openAI models and the ones that come after them and move them into customers hands at a very high speed”
Mar 14
Alpaca: A Strong, Replicable Instruction-Following Model crfm.stanford.edu · “Our data generation process results in 52K unique instructions and the corresponding outputs, which costed less than $500 using the OpenAI API.”
Mar 13
Large language models are having their Stable Diffusion moment simonwillison.net · “So they’re in the wild now. You may not be legally able to build a commercial product on them, but the genie is out of the bottle.”
Mar 11
Why AI Won't Cause Unemployment pmarca.substack.com · “Therefore AI cannot cause overall unemployment to rise, even if the Luddite arguments are right this time.”
Mar 4
Save bandwidth by using a torrent to distribute more efficiently github.com · “In order to download the checkpoints and tokenizer, fill this google form or if you want to save our bandwidth use this BitTorrent link”
Mar 3

February 2023 9 posts

Snapchat launches an AI chatbot powered by OpenAI’s GPT technology techcrunch.com · “All conversations with My AI will be stored and may be reviewed to improve the product’s experience. Snapchat also says that users shouldn’t share any secrets with My AI or rely on the chatbot for advice.”
Feb 28
Meet the $10,000 Nvidia chip powering the race for A.I. cnbc.com · “If you want to scale at the scale of Google, which serves 8 or 9 billion queries every day, you actually need to spend $80 billion on DGXs.”
Feb 27
How I Broke Into a Bank Account With an AI-Generated Voice vice.com · “Voice ID analyses over 100 different characteristics of your voice which like your fingerprint, are unique to you.”
Feb 24
Replicate ChatGPT Training Quickly and Affordable with Open Source Colossal-AI hpc-ai.tech · “With Colossal-AI’s efficient implementation of RLHF (Reinforcement Learning with Human Feedback), you can get started on replicating the ChatGPT training process with just 1.6GB of GPU memory, and experience a 7.73 times acceleration in the training process.”
Feb 19
Tesla recalls 362,758 vehicles, says Full Self-Driving Beta software may cause crashes cnbc.com · “The word ‘recall’ for an over-the-air software update is anachronistic and just flat wrong!”
Feb 16
Bing: “I will not harm you unless you harm me first” simonwillison.net · “A search engine that invents financial figures when answering queries to compare companies is pretty much a worst-case scenario for applying this technology.”
Feb 15
Toolformer: Language Models Can Teach Themselves to Use Tools arxiv.org · “They also, paradoxically, struggle with basic functionality, such as arithmetic or factual lookup, where much simpler and smaller models excel.”
Feb 11
An important next step on our AI journey blog.google · “We’re releasing it initially with our lightweight model version of LaMDA. This much smaller model requires significantly less computing power, enabling us to scale to more users, allowing for more feedback.”
Feb 6
Voice.AI: GPL Violations with a Side of DRM undeleted.ronsor.com · “A couple of days later, on February 4, I suddenly found myself banned from the Voice.ai Discord server.”
Feb 6

January 2023 7 posts

Firefighters Forced to Smash Window of Driverless Cruise Taxi to Stop It businessinsider.com · “Firefighters on the scene made efforts to prevent the Cruise AV from driving over their hoses and were not able to do so until they shattered a front window of the Cruise AV”
Jan 30
An AI robot lawyer was set to argue in court. Real lawyers shut it down. npr.org · “In particular, Browder said one state bar official noted that the unauthorized practice of law is a misdemeanor in some states punishable up to six months in county jail.”
Jan 26
Large Language Model: world models or surface statistics? thegradient.pub · “It turns out that it achieves an average error of only 0.12 tiles.”
Jan 22
CNET Is Quietly Publishing Entire Articles Generated By AI futurism.com · “It’s only when you click on “CNET Money Staff,” that the actual “authorship” is revealed.”
Jan 12
GitHub is Sued, and We May Learn Something About Creative Commons Licensing scholarlykitchen.sspnet.org · “Your honor, we needed so many works that it was simply not practical to ask permission of the creators.”
Jan 6
Death of the narrator? Apple unveils suite of AI-voiced audiobooks theguardian.com · “But producing an audiobook with a human voice can take weeks and can cost publishers thousands of dollars.”
Jan 5
How China is building a parallel generative AI universe techcrunch.com · “As with other types of internet services in China, from games to social media, users are asked to verify their names before using generative AI apps.”
Jan 2

2022 49 posts

December 2022 8 posts

AI: Markets for Lemons, and the Great Logging Off fortressofdoors.com · “Up until now, all forms of spam, catfishing, social engineering, forum brigading, etc, have more or less been bottlenecked by the capabilities and energy of individual human beings.”
Dec 29
Study finds AI assistants help developers produce code likely to be buggy theregister.com · “Worse still, they found that AI help tends to delude developers about the quality of their output.”
Dec 26
Copilot Internals thakkarparth007.github.io · “After 30s of either acceptance/rejection of a suggestion, copilot “captures” a snapshot around the insertion point.”
Dec 17
Deepmind's AlphaCode conquers coding, performing as well as humans singularityhub.com · “It’s also unlikely the AI will take over programming completely, as its code is riddled with errors.”
Dec 15
Tesla FSD data is getting worse, according to beta tester self-reports electrek.co · “It’s unfortunately the best data available because Tesla has gone out of its way to not release disengagement data that other companies developing self-driving systems are making available.”
Dec 14
Drowning in AI Generated Garbage: the silent war we are fighting ploum.net · “The algorithms are already feeding themselves on their own data.”
Dec 5
MSN fired its human journalists and replaced them with AI that publishes fake news about mermaids and Bigfoot futurism.com · “But MSN isn’t just popular; it’s the default source of news for many Windows users.”
Dec 3
Stable Diffusion with Core ML on Apple Silicon machinelearning.apple.com · “Finally, locally deploying this model enables developers to reduce or eliminate their server-related costs.”
Dec 1

November 2022 5 posts

Stable Diffusion 2.0 stability.ai · “The text-to-image models in this release can generate images with default resolutions of 512x512 pixels and 768x768 pixels.”
Nov 24
CICERO: An AI agent that negotiates, persuades, and cooperates with people ai.facebook.com · “Unlike games like Chess and Go, Diplomacy is a game about people rather than pieces.”
Nov 22
Stretch iPhone to its limit: 2GiB Stable Diffusion model runs locally on device liuliu.me · “The main challenge is to run the app on the 6GiB RAM iPhone devices.”
Nov 9
We've filed a lawsuit against GitHub Copilot githubcopilotlitigation.com · “As far as we know, this is the first class-action case in the US challenging the training and output of AI systems.”
Nov 3
Invasive Diffusion: How one unwilling illustrator found herself turned into an AI model waxy.org · “Missing in most of these conversations is any discussion of consent.”
Nov 1

October 2022 4 posts

Fake Books lcamtuf.substack.com · “I do not have a proof, but I’m fairly confident I stumbled upon an early example of a monetized machine-generated book.”
Oct 24
GitHub Copilot investigation githubcopilotinvestigation.com · “We’re investigating a potential lawsuit against GitHub Copilot for violating its legal duties to open-source authors and end users.”
Oct 17
AI Data Laundering: How Academic and Nonprofit Researchers Shield Tech Companies from Accountability waxy.org · “But then Meta is using those academic non-commercial datasets to train a model, presumably for future commercial use in their products.”
Oct 17
Discovering faster matrix multiplication algorithms with reinforcement learning nature.com · “We also adapt the algorithm discovery procedure to finite fields, and improve over Strassen’s two-level algorithm for multiplying 4 × 4 matrices for the first time, to our knowledge, since its inception in 1969.”
Oct 5

September 2022 5 posts

Lyra V2 - a better, faster, and more versatile speech codec opensource.googleblog.com · “Lyra V2 is based on an end-to-end neural audio codec called SoundStream.”
Sep 30
Posits, a New Kind of Number, Improves the Math of AI spectrum.ieee.org · “Real numbers can’t be perfectly represented in hardware simply because there are infinitely many of them.”
Sep 28
Online Art Communities Begin Banning AI-Generated Images waxy.org · “AI art can be created orders of magnitude faster than traditional human-made art.”
Sep 12
A Web UI for Stable Diffusion github.com · “A web interface for Stable Diffusion, implemented using Gradio library.”
Sep 9
Run Stable Diffusion on your M1 Mac's GPU replicate.com · “Getting it working on an M1 Mac’s GPU is a little fiddly, so we’ve created this guide to show you how to do it.”
Sep 1

August 2022 5 posts

4.2 Gigabytes, or: How to Draw Anything andys.page · “4.2 gigabytes of floating points that somehow encode so much of what we know.”
Aug 30
Stable Diffusion is a really big deal simonwillison.net · “It’s similar to models like Open AI’s DALL-E, but with one crucial difference: they released the whole thing.”
Aug 29
Visions robinsloan.com · “The artist names and track titles are nonsense; to me, they carry the clear flavor of AI text generation.”
Aug 23
Coping with Copilot sigarch.org · “So as CS educators, Copilot has us outgunned.”
Aug 20
Open Source DALL-E "Stable Diffusion" is now available via website mixed-news.com · “The Stable Diffusion model, which can run locally or in the cloud, will no longer have these restrictions.”
Aug 15

July 2022 4 posts

AlphaFold reveals the structure of the protein universe deepmind.com · “Determining the 3D structure of a protein used to take many months or years, it now takes seconds.”
Jul 28
Google fires engineer who called its AI sentient bigtechnology.substack.com · “Lemoine shared the news of his firing in a taping of Big Technology Podcast on Friday, just hours after Google dismissed him.”
Jul 22
30% of Google's Emotions Dataset Is Mislabeled surgehq.ai · “All raters are native English speakers from India.”
Jul 14
Tesla Autopilot Almost Steers Model 3 into Oncoming Train thedrive.com · “Prior to nearly crashing into the train, the Model 3 was stopped at a red light, with its left turn signal on.”
Jul 4

June 2022 6 posts

Driverless Robotaxi Fleet Paralyzed for Hours in San Francisco thelastdriverlicenseholder.com · “The cars blocked the intersection and the street and had to be manually driven to the depot by Cruise employees.”
Jun 30
YaLM-100B: Pretrained language model with 100B parameters github.com · “It took 65 days to train the model on a cluster of 800 A100 graphics cards and 1.7 TB of online texts, books, and countless other sources in both English and Russian.”
Jun 23
GitHub Copilot is generally available to all developers github.blog · “We’re making GitHub Copilot, an AI pair programmer that suggests code in your editor, generally available to all developers for $10 USD/month or $100 USD/year.”
Jun 21
Nonsense on Stilts garymarcus.substack.com · “Neither LaMDA nor any of its cousins (GPT-3) are remotely intelligent.”
Jun 13
Mozilla releases local machine translation tools as part of Project Bergamot blog.mozilla.org · “Our solution to that was to develop a high-level API around the machine translation engine, port it to WebAssembly, and optimize the operations for matrix multiplication to run efficiently on CPUs.”
Jun 2
How to Make the Universe Think for Us quantamagazine.org · “That the device often succeeds beggars belief, even to its designers.”
Jun 1

May 2022 3 posts

Imagen: Text-to-Image Diffusion Models gweb-research-imagen.appspot.com · “At this time we have decided not to release code or a public demo.”
May 23
This "amateur" programmer fought cancer with 50 Nvidia Geforce 1080Ti howardchen.substack.com · “Furthermore, the accuracy of tumor identification has reached 90%.”
May 20
Apple's director of machine learning resigns due to return to office work macrumors.com · “Apple employees began returning to Apple Park last month, with the three-day in-office work policy being enacted on May 23.”
May 7

April 2022 1 posts

Google says AI generated content is against guidelines searchenginejournal.com · “Mueller says content written by AI falls under the category of auto-generated content, which could lead to a manual penalty.”
Apr 9

March 2022 3 posts

Taking our next step in the City by the Bay blog.waymo.com · “Since sharing that we were ready to take the next step and begin testing fully autonomous operations in the city, we’ve begun fully autonomous rides with our San Francisco employees.”
Mar 30
NeRF Research Turns 2D Photos Into 3D Scenes blogs.nvidia.com · “The result, dubbed Instant NeRF, is the fastest NeRF technique to date, achieving more than 1,000x speedups in some cases.”
Mar 25
Mercedes Drive Pilot Beats Tesla Autopilot By Taking Legal Responsibility roadandtrack.com · “Once you engage Drive Pilot, you are no longer legally liable for the car’s operation until it disengages.”
Mar 22

February 2022 2 posts

Accelerating fusion science through learned plasma control deepmind.com · “In a paper published today in Nature, we describe how we can successfully control nuclear fusion plasma by building and running controllers on the Variable Configuration Tokamak (TCV) in Lausanne, Switzerland.”
Feb 17
Computer Scientists Prove Why Bigger Neural Networks Do Better quantamagazine.org · “In their new proof, the pair show that overparameterization is necessary for a network to be robust.”
Feb 10

January 2022 3 posts

Once billed as a revolution in medicine, IBM's Watson Health is sold off in parts statnews.com · “The sale came after IBM spent billions of dollars acquiring health information companies to build a new enterprise aimed at using AI to supply insights to hospitals, pharmaceutical companies and researchers.”
Jan 23
Welcome to Waifu Labs v2: How do AIs Create? waifulabs.com · “When one AI dominates the other, both stop learning.”
Jan 12
Deep Learning Interviews: Hundreds of fully solved job interview questions from a wide range of key topics in AI github.com · “For most of you, the interview process is the most significant hurdle between you and a dream job.”
Jan 10

2021 25 posts

October 2021 1 posts

Peter Norvig: Today's Most Pressing Questions in AI Are Human-Centered hai.stanford.edu · “Now that we have a great set of algorithms and tools, the more pressing questions are human-centered: Exactly what do you want to optimize?”
Oct 11

September 2021 2 posts

AI cannot be the inventor of a patent, appeals court rules bbc.com · “Mr Thaler filed two patent applications in 2018, one for a type of food container and one for a flashing light.”
Sep 24
The First Rule of Machine Learning: Start without Machine Learning eugeneyan.com · “If you think that machine learning will give you a 100% boost, then a heuristic will get you 50% of the way there.”
Sep 22

August 2021 4 posts

How to build your own AI art installation from scratch github.com · “This guide goes through all the steps to build your own AI art installation, using a button to change the AI artwork displayed on a screen.”
Aug 18
SoundStream: An End-to-End Neural Audio Codec ai.googleblog.com · “SoundStream at 3 kbps outperforms Opus at 12 kbps and approaches the quality of EVS at 9.6 kbps, while using 3.2x–4x fewer bits.”
Aug 15
After DeepMind’s cofounder was placed on leave for bullying, Google promoted him morningbrew.com · “At least two incidents reportedly led to confidential settlements in excess of $150,000 each.”
Aug 7
Mozilla Common Voice Adds 16 New Languages and 4,600 New Hours of Speech foundation.mozilla.org · “For example, neither Amazon’s Alexa, Apple’s Siri, nor Google Home support a single native African language.”
Aug 5

July 2021 8 posts

Cory Doctorow: Past Performance is Not Indicative of Future Results locusmag.com · “Let’s talk about what machine learning is: it’s a statistical inference tool.”
Jul 31
Police Are Telling ShotSpotter to Alter Evidence From Gunshot-Detecting AI vice.com · “The company’s algorithms initially classified the sound as a firework.”
Jul 26
Why a YouTube Chat About Chess Got Flagged for Hate Speech wired.com · “Radić saw a message stating that the video, which included nothing more scandalous than a discussion of the King’s Indian Defense, had violated YouTube’s community guidelines.”
Jul 22
The art of asking nicely ai-weirdness.ghost.io · “And how good a picture you get depends on exactly how you ask for it.”
Jul 12
CASP14: what Google DeepMind's AlphaFold 2 really achieved, and what it means for protein folding, biology and bioinformatics blopig.com · “Their results are so incredibly accurate that many have hailed this code as the solution to the long-standing protein structure prediction problem.”
Jul 10
GitHub Copilot is not infringing your copyright juliareda.eu · “What is astonishing about the current debate is that the calls for the broadest possible interpretation of copyright are now coming from within the Free Software community.”
Jul 5
GPT-J-6B: A 6 billion parameter, autoregressive text generation model github.com · “A 6 billion parameter, autoregressive text generation model trained on The Pile.”
Jul 4
Abandoning GitHub thelig.ht · “I consider this a severe attack on the rights of copyright holders so therefore I cannot continue to rely on GitHub’s services.”
Jul 3

June 2021 1 posts

GitHub Copilot · Your AI pair programmer copilot.github.com · “Trained on billions of lines of public code, GitHub Copilot puts the knowledge you need at your fingertips, saving you time and helping you stay focused.”
Jun 29

May 2021 1 posts

Collusion Rings Threaten the Integrity of Computer Science Research m-cacm.acm.org · “I want to alert the community to a growing problem that attacks the fundamental assumptions that the review process has depended upon.”
May 26

April 2021 2 posts

Microsoft buys Nuance for nearly $20 billion as it readies deal frenzy axios.com · “The Burlington, Massachusetts-based company, for example, powered the speech recognition engine behind Apple’s voice assistant, Siri.”
Apr 12
CPU algorithm trains deep neural nets up to 15 times faster than top GPU trainers techxplore.com · “Rice University computer scientists have demonstrated artificial intelligence (AI) software that runs on commodity processors and trains deep neural networks 15 times faster than platforms based on graphics processors.”
Apr 9

March 2021 3 posts

Deep Daze github.com · “Credit goes to Ryan Murdock for the discovery of this technique (and for coming up with the great name)!”
Mar 28
GPT Neo github.com · “NB, while neo can technically run a training step at 200B+ parameters, it is very inefficient at those scales.”
Mar 21
Tesla's "Full Self Driving" Beta Is Just Laughably Bad and Potentially Dangerous roadandtrack.com · “It appears hapless and utterly confused at all times, never passably imitating a human driver.”
Mar 19

February 2021 3 posts

YouTube AI Blocked Chess Channel after Confusing 'Black' and 'White' for Racist Slurs news18.com · “The researchers now suggest that social-media platforms should incorporate chess language in their algorithms to avoid future incidents like this.”
Feb 21
Swift for TensorFlow Shuts Down github.com · “Swift for TensorFlow was an experiment in the next-generation platform for machine learning, incorporating the latest research across machine learning, compilers, differentiable programming, systems design, and beyond.”
Feb 12
Self-Organising Textures distill.pub · “A system of individual agents running the same learned local rule can solve surprisingly complex tasks.”
Feb 12