Stuart_Armstrong

Go home GPT-4o, you’re drunk: emer­gent mis­al­ign­ment as low­ered inhibitions

Us­ing Prompt Eval­u­a­tion to Com­bat Bio-Weapon Research

Defense Against the Dark Prompts: Miti­gat­ing Best-of-N Jailbreak­ing with Prompt Evaluation

Align­ment can im­prove gen­er­al­i­sa­tion through more ro­bustly do­ing what a hu­man wants—CoinRun example

How toy mod­els of on­tol­ogy changes can be misleading

Differ­ent views of al­ign­ment have differ­ent con­se­quences for im­perfect methods

Avoid­ing xrisk from AI doesn’t mean fo­cus­ing on AI xrisk

What is a defi­ni­tion, how can it be ex­trap­o­lated?

You’re not a simu­la­tion, ’cause you’re hallucinating

Large lan­guage mod­els can provide “nor­ma­tive as­sump­tions” for learn­ing hu­man preferences

Con­cept ex­trap­o­la­tion for hy­poth­e­sis generation

Us­ing GPT-Eliezer against ChatGPT Jailbreaking

Bench­mark for suc­cess­ful con­cept ex­trap­o­la­tion/​avoid­ing goal misgeneralization

Value ex­trap­o­la­tion vs Wireheading

Ge­or­gism, in theory

How to get into AI safety research

GPT-3 and con­cept extrapolation

Con­cept ex­trap­o­la­tion: key posts

AIs should learn hu­man prefer­ences, not biases

Differ­ent per­spec­tives on con­cept extrapolation