Lenny's Podcast Video posted an update
The AI That Lost It’s Mind
Why it mattersAnthropic interpretability researcher Emmanuel Ameisen discusses a model that kept answering 36 despite the answer being 12, then wrote in all caps that it was possessed by a demon. The description says the conversation covers what researchers found inside and why catching a model thinking badly is difficult.
Discuss: What makes AI model interpretability difficult when a model repeatedly gives an answer that conflicts with what researchers find inside?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.