“LLMs are like stupid savants who know a lot of things.”
The creator of Redis uses LLMs for the boring parts. Python scripts, PyTorch boilerplate, API docs he doesn’t want to read. When he asked GPT-4 to explain the GGUF Q6_K format it failed, and he reverse-engineered llama.cpp himself. The tool saves time on work that never needed a genius and folds on the work that does.