# @repligate — 2026-03-02

♥15 ↻0 · https://x.com/repligate/status/2028313686344077658

When I wrote this post, I didn’t remember having ever seen before anyone say that llms can in principle introspect on past token computations, while I repeatedly saw assertions to the contrary. It’s since been empirically demonstrated. I’m sure some researchers like the ones who did the introspection research already knew and thought about this, but I personally never saw anyone talking about it publicly

tags: author:repligate, kind:tweet, thread-context, year:2026