Joe Benton

Karma: 330

Towards training-time mitigations for alignment faking in RL

Vlad Mikulik, gasteigerjo, Hoagy, Joe Benton, Benjamin Wright, Jonathan Uesato, Monte M, Fabien Roger and evhub

16 Dec 2025 21:01 UTC

33 points

1 comment5 min readLW link

(alignment.anthropic.com)

Supervised fine-tuning as a method for training-based AI control

Emil Ryd, Joe Benton and Vivek Hebbar

13 Nov 2025 22:25 UTC

40 points

0 comments18 min readLW link

Training fails to elicit subtle reasoning in current language models

mishajw, Fabien Roger, Hoagy, gasteigerjo, Joe Benton and Vlad Mikulik

9 Oct 2025 19:04 UTC

49 points

3 comments4 min readLW link

(alignment.anthropic.com)

Inverse Scaling in Test-Time Compute

Joe Benton, Ethan Perez and aryopg

22 Jul 2025 22:06 UTC

20 points

2 comments2 min readLW link

(arxiv.org)

Reasoning models don’t always say what they think

Joe Benton, Ethan Perez, Vlad Mikulik and Fabien Roger

9 Apr 2025 19:48 UTC

28 points

4 comments1 min readLW link

(www.anthropic.com)

Do models say what they learn?

Andy Arditi, marvinli, Joe Benton and Miles Turpin

22 Mar 2025 15:19 UTC

126 points

12 comments13 min readLW link

Won’t vs. Can’t: Sandbagging-like Behavior from Claude Models

Joe Benton and Zachary Witten

19 Feb 2025 20:47 UTC

15 points

1 comment1 min readLW link

(alignment.anthropic.com)

Sabotage Evaluations for Frontier Models

David Duvenaud, Joe Benton, Sam Bowman, evhub, mishajw, Eric Christiansen, HoldenKarnofsky, Ethan Perez and Buck

18 Oct 2024 22:33 UTC

95 points

56 comments6 min readLW link

(assets.anthropic.com)