Stuart_Armstrong

Value gen­er­al­i­sa­tion The­ory of Change: putting it into practice

Value gen­er­al­i­sa­tion The­ory of Change: the the­ory be­hind the approach

Avoid­ing the cor­po­rate treach­er­ous turn: crowd-sourc­ing de­sign ideas

Sim­plify­ing the an­thropic im­pos­si­bil­ity result

III. An­thropic rea­son­ing has is­sues with in­finite wor­lds; D-SIA can fix this

II. An­thropic rea­son­ing with du­pli­ca­tion is not con­sis­tent with the usual prob­a­bil­ity properties

I. An­thropic rea­son­ing with­out du­pli­cates is just stan­dard Bayesian updating

Value Gen­er­al­i­sa­tion 3: Pre-al­igned AIs

Value Gen­er­al­i­sa­tion 2: The Miss­ing Hole in AIs’ abilities

Value Gen­er­al­i­sa­tion 1: a Re­search and De­ploy­ment Program

The true “test” dataset for a gen­er­al­ised task

Oc­cam’s ra­zor is about us­ing the past to pre­dict the future

Value gen­er­al­i­sa­tion: value correction

Prag­matic FDT, and pre­dic­tors as game theory

The fu­ture of al­ign­ment if LLMs are a bubble

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