-
A Structural View of AI Engrams: Notes After a Conversation with Peter Dayan
Floating thoughts on what kind of memory I am actually looking for when I search for engrams in AI models, sparked by a question I could not immediately answer.
-
Emergent Misalignment: When Finetuning Goes Wrong in Ways You Didn't Expect
A deep dive into emergent misalignment — how narrow finetuning produces broadly misaligned LLMs, what the latest research reveals about the mechanism, and why I think EM is not a new behavior but an unlocking of something already there.