I also want to reframe the discussion as giving the LLm episodic (subset) memory

Written by

I also want to reframe the discussion as giving the LLm episodic (subset) memory ( which is the source of weakness regardless of context length.

LLMs do learn concepts but not yet episodes. And episodes are the means humans index a subset memories.

Source date (UTC): 2025-04-08 21:40:50 UTC

Original post: https://twitter.com/i/web/status/1909723133537591419

Reply addressees: @NickADobos

Replying to: https://twitter.com/i/web/status/1909428901170102734

Comments