Academic Graham Neubig requested reference papers analyzing this decline.
@CFGeek @gneubig @ArthurConmy @nabla_theta As"just read the chain of thought" becomes less effective, it might matter much more, tbd
@CFGeek @gneubig @ArthurConmy @nabla_theta Things like activation probes are used and matter. And Anthropic are increasingly using it for auditing. But beyond those two use cases it's not a big deal imo
Users thank researchers for the post on activation probes rising in importance as chain-of-thought methods weaken and ask for key papers to read on the topic.
No Digg Deeper questions have been answered for this story yet.
@NeelNanda5 @CFGeek @ArthurConmy @nabla_theta Thanks @NeelNanda5 ! I'd love to get some references to the best papers to read if you have any.
@CFGeek @gneubig @ArthurConmy @nabla_theta As"just read the chain of thought" becomes less effective, it might matter much more, tbd