Lenny's Podcast Video posted an update
A mind-expanding conversation on how models actually work | Emmanuel Ameisen (Anthropic)
Why it mattersEmmanuel Ameisen discusses how LLMs form representations and plan ahead, and why next-token prediction may undersell those behaviors. The conversation also connects interpretability-including the Golden Gate Claude experiment-to alignment and safety, and considers how smooth scaling can produce apparent capability jumps.
Discuss: How should LLM interpretability findings, such as the Golden Gate Claude experiment, inform the way researchers approach alignment and safety?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.