@AnnatarXBT: this is pure f*cking treasure Andrej Karpathy has been inside Anthropic since May, and the two pieces you need to build…

this is pure f*cking treasure Andrej Karpathy has been inside Anthropic since May, and the two pieces you need to build a loop like his are already public almost nobody has opened either of them: > Anthropic's cookbook that turns your documents into a knowledge graph with Claude > Karpathy's autoresearch loop, which ran 700 experiments in 2 days and found 20 optimizations with no human steering one gives the agent a memory it can reason over. the other keeps it working while you sleep i wired the graph half into my own setup the very first reply changed: Claude dropped the stock answer and actually worked through the problem your feed is still typing prompts into a blank chat. the few who build this now get to explain it to everyone else later open the cookbook first, then the 300-agent breakdown below ↓