“If you’re not within 10% or so of the global optimal performance in EE, your suggestions have near-zero value.”
JITX ran GPT-4o, Claude 3 Opus and Gemini 1.5 Pro through real board design tasks. One pulled all 95 pin rows out of an 820-page datasheet without an error. Asked to pick parts for a robot motor driver, they suggested datacenter transceivers and other “kinda-right, but actually very wrong” components. LLMs make decent clerks and bad engineers.