Buck(Buck Shlegeris)

“Other peo­ple are wrong” vs “I am right”

Six eco­nomics mis­con­cep­tions of mine which I’ve re­solved over the last few years

AI Con­trol: Im­prov­ing Safety De­spite In­ten­tional Subversion

Lan­guage mod­els seem to be much bet­ter than hu­mans at next-to­ken prediction

The pro­to­typ­i­cal catas­trophic AI ac­tion is get­ting root ac­cess to its datacenter

Worst-case think­ing in AI alignment

A fresh­man year dur­ing the AI midgame: my ap­proach to the next year

Red­wood Re­search’s cur­rent project

Take­off speeds have a huge effect on what it means to work on AI x-risk

The the­ory-prac­tice gap

The case for be­com­ing a black-box in­ves­ti­ga­tor of lan­guage models

One-layer trans­form­ers aren’t equiv­a­lent to a set of skip-trigrams

Try­ing to dis­am­biguate differ­ent ques­tions about whether RLHF is “good”

Funds are available to sup­port LessWrong groups, among others

The al­ign­ment prob­lem in differ­ent ca­pa­bil­ity regimes

Some thoughts on criticism

Poly­se­man­tic­ity and Ca­pac­ity in Neu­ral Networks

Meta-level ad­ver­sar­ial eval­u­a­tion of over­sight tech­niques might al­low ro­bust mea­sure­ment of their adequacy

How good is hu­man­ity at co­or­di­na­tion?

Un­trusted smart mod­els and trusted dumb models