Eisenstein 14 hours ago

RAG is like constantly reading your notes instead of integrating experiences into your processes.

koakuma-chan 17 hours ago

Not really. For example we still can’t get coding agents to work reliably, and I think it’s a memory problem, not a capabilities problem.

  • atlex2 15 hours ago

    On the other hand, test-time weight updates would make model interpretability much harder.