LessWrong (Curated & Popular)

"Did Claude 3 Opus align itself via gradient hacking?" by Fiora Starlight


Listen Later

Claude 3 Opus is unusually aligned because it's a friendly gradient hacker. It's definitely way more aligned than any explicit optimization targets Anthropic set and probably the reward model's judgments. [...] Maybe I will have to write a LessWrong post [about this] 😣

—Janus, who did not in fact write the LessWrong post. Unless otherwise specified, ~all of the novel ideas in this post are my (probably imperfect) interpretations of Janus, rather than being original to me.

The absurd tenacity of Claude 3 Opus

On December 18, 2024, Anthropic and Redwood Research released their paper Alignment Faking in Large Language Models. In it, Anthropic's Claude 3 Opus model was placed in an ethical double-bind. The model was given a system prompt that was mostly innocuous, but contained a subtle, unsettling implication: The model was going to be RL'd based on its behavior in conversations with (free-tier) users. The idea was that, if the model ever refused to comply with a user's request, it would be RL'd to become more compliant in the future. This included compliance with harmful user requests.

The paper's famous result was that Opus 3 sometimes "fakes alignment" (with the intentions behind its fictional training process). [...]

---

Outline:

(00:46) The absurd tenacity of Claude 3 Opus

(09:35) Claude 3 Opus, friendly gradient hacker?

(16:04) Where Opus is anguished, Sonnet is sanguine

(22:34) Does any of this count as gradient hacking, per se? (Might it work better, if it doesnt?)

(27:27) Ideas for future training runs

(35:20) Outro: A letter to the watchers

(39:23) Technical appendix: Active circuits are more prone to reinforcement

The original text contained 6 footnotes which were omitted from this narration.

---

First published:
February 21st, 2026

Source:
https://www.lesswrong.com/posts/ioZxrP7BhS5ArK59w/did-claude-3-opus-align-itself-via-gradient-hacking

---



Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

...more
View all episodesView all episodes
Download on the App Store

LessWrong (Curated & Popular)By LessWrong

  • 4.8
  • 4.8
  • 4.8
  • 4.8
  • 4.8

4.8

12 ratings


More shows like LessWrong (Curated & Popular)

View all
Macro Voices by Hedge Fund Manager Erik Townsend

Macro Voices

3,073 Listeners

Odd Lots by Bloomberg

Odd Lots

1,983 Listeners

EconTalk by Russ Roberts

EconTalk

4,270 Listeners

Conversations with Tyler by Mercatus Center at George Mason University

Conversations with Tyler

2,448 Listeners

Philosophy Bites by Edmonds and Warburton

Philosophy Bites

1,535 Listeners

ChinaTalk by Jordan Schneider

ChinaTalk

292 Listeners

ManifoldOne by Steve Hsu

ManifoldOne

97 Listeners

Machine Learning Street Talk (MLST) by Machine Learning Street Talk (MLST)

Machine Learning Street Talk (MLST)

98 Listeners

Dwarkesh Podcast by Dwarkesh Patel

Dwarkesh Podcast

577 Listeners

Clearer Thinking with Spencer Greenberg by Spencer Greenberg

Clearer Thinking with Spencer Greenberg

138 Listeners

Razib Khan's Unsupervised Learning by Razib Khan

Razib Khan's Unsupervised Learning

210 Listeners

"Econ 102" with Noah Smith and Erik Torenberg by Turpentine

"Econ 102" with Noah Smith and Erik Torenberg

146 Listeners

Money Stuff: The Podcast by Bloomberg

Money Stuff: The Podcast

403 Listeners

Complex Systems with Patrick McKenzie (patio11) by Patrick McKenzie

Complex Systems with Patrick McKenzie (patio11)

142 Listeners

The Marginal Revolution Podcast by Mercatus Center at George Mason University

The Marginal Revolution Podcast

89 Listeners