Redwood Research blog

Redwood Research blog

Home
Podcast
Reading List
Archive
About
Most Popular
View all

Recent posts

View all
Recent LLMs can use filler tokens or problem repeats to improve (no-CoT) math performance
AI can sometimes distribute cognition over many extra tokens
 • Ryan Greenblatt
BashArena and Control Setting Design
We’ve just released BashArena, a new high-stakes control setting we think is a major improvement over the settings we’ve used in the past.
 • Adam Kaufman and James Lucassen
The behavioral selection model for predicting AI motivations
The basic arguments about AI motivations in one causal graph
 • Alex Mallen and Buck Shlegeris
Will AI systems drift into misalignment?
A reason alignment could be hard
 • Josh Clymer
What's up with Anthropic predicting AGI by early 2027?
I operationalize Anthropic's prediction of "powerful AI" and explain why I'm skeptical
 • Ryan Greenblatt
Sonnet 4.5's eval gaming seriously undermines alignment evals
And this seems caused by training on alignment evals.
 • Alexa Pan and Ryan Greenblatt
The latest from
Buck Shlegeris
The inaugural Redwood Research podcast
User's avatar
Ryan Greenblatt
Recent LLMs can do 2-hop and 3-hop latent (no CoT) reasoning on natural facts
User's avatar
Alex Mallen
The behavioral selection model for predicting AI motivations
User's avatar
Josh Clymer
Will AI systems drift into misalignment?
User's avatar
Julian Stastny
Prospects for studying actual schemers
User's avatar
Vivek Hebbar
Recent Redwood Research project proposals
User's avatar
Redwood Research blog
Redwood Research blog
We research catastrophic AI risks and techniques that could be used to mitigate them.
Recommendations
The Power Law
The Power Law
Peter Wildeford
ForeWord
ForeWord
Forethought
AI Futures Project
AI Futures Project
Daniel Kokotajlo
© 2026 Buck Shlegeris · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture