sudarsh

Prior work has found that Chain of Thought (CoT) can be unfaithful. Should we then ignore what it says? In new research, we find that the CoT is informative about LLM cognition as long as the cognition is complex enough that it can’t be performed in a single forward pass.

i want to form a chess club in sf (no founders)

Anthropic accidentally trained against the chain of thought in Claude Mythos, Opus 4.6, and Sonnet 4.6

really wish I could use it to audit Bend and improve its security in general. timing couldn't be better, it isn't every day that a new proof assistant is launched. if anyone has any path or idea on how I could be part of Glasswing please let me know!

My first post for inkhaven! (day 1/30) I plan to post multiple times per day.

I think companies have pretty good incentives to align AI. And it's not obvious to me that AI alignment is difficult. Though small chances of very bad outcomes can be very bad.

Despite being just one cell, Lacrymaria olor is a formidable predator that hunts and consumes other microorganisms. 📽: James Weiss

Condom blowjobs work best if the condom is non-lubricated and you put flavorless lube on the inside to improve sensation, btw

my friends and i bought a foreclosed alley in san francisco. it's an actual road that cars drive down! we're letting the entire internet design a mural on it. submit a drawing, vote on them, and the top 1,280 get painted on the pavement permanently ⬇️