mishajw

Karma: 180

HTTP/1.1 303 See other
Location: https://mishajw.com

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)

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)

How well do truth probes generalise?

mishajw24 Feb 2024 14:12 UTC

96 points

11 comments9 min readLW link

Jailbreaking GPT-4 with the tool API

mishajw21 Feb 2024 11:16 UTC

20 points

2 comments4 min readLW link

Distilled Representations Research Agenda

Hoagy and mishajw

18 Oct 2022 20:59 UTC

15 points

2 comments8 min readLW link