Rotations in Superposition

Linda Linsefors and Lucius Bushnaq

15 Dec 2025 14:58 UTC

54 points

0 comments11 min readLW link

My AGI safety research—2025 review, ’26 plans

Steven Byrnes11 Dec 2025 17:05 UTC

127 points

4 comments12 min readLW link

6 reasons why “alignment-is-hard” discourse seems alien to human intuitions, and vice-versa

Steven Byrnes3 Dec 2025 18:37 UTC

287 points

54 comments17 min readLW link

Auditing Games for Sandbagging [paper]

Jordan Taylor and Joseph Bloom

9 Dec 2025 18:37 UTC

101 points

4 comments10 min readLW link

Alignment remains a hard, unsolved problem

evhub27 Nov 2025 8:45 UTC

339 points

94 comments14 min readLW link

An Ambitious Vision for Interpretability

leogao5 Dec 2025 22:57 UTC

163 points

7 comments4 min readLW link

We need a field of Reward Function Design

Steven Byrnes8 Dec 2025 19:15 UTC

106 points

8 comments5 min readLW link

The behavioral selection model for predicting AI motivations

Alex Mallen and Buck

4 Dec 2025 18:46 UTC

157 points

21 comments16 min readLW link

Reward Function Design: a starter pack

Steven Byrnes8 Dec 2025 19:15 UTC

77 points

7 comments16 min readLW link

Natural emergent misalignment from reward hacking in production RL

evhub, Monte M, Benjamin Wright and Jonathan Uesato

21 Nov 2025 20:00 UTC

258 points

32 comments9 min readLW link

A Pragmatic Vision for Interpretability

Neel Nanda, Josh Engels, Arthur Conmy, Senthooran Rajamanoharan, bilalchughtai, CallumMcDougall, János Kramár and lewis smith

1 Dec 2025 13:05 UTC

133 points

35 comments27 min readLW link

Legible vs. Illegible AI Safety Problems

Wei Dai4 Nov 2025 21:39 UTC

361 points

93 comments2 min readLW link

Relitigating the Race to Build Friendly AI

Wei Dai3 Dec 2025 11:34 UTC

67 points

43 comments3 min readLW link

Abstract advice to researchers tackling the difficult core problems of AGI alignment

TsviBT22 Nov 2025 0:53 UTC

129 points

10 comments8 min readLW link

[Paper] Does Self-Evaluation Enable Wireheading in Language Models?

David Africa8 Dec 2025 16:03 UTC

25 points

2 comments2 min readLW link

Alignment will happen by default. What’s next?

Adrià Garriga-alonso25 Nov 2025 7:14 UTC

98 points

128 comments6 min readLW link

(open.substack.com)

How Can Interpretability Researchers Help AGI Go Well?

Neel Nanda, Josh Engels, Senthooran Rajamanoharan, Arthur Conmy, bilalchughtai, CallumMcDougall, János Kramár and lewis smith

1 Dec 2025 13:05 UTC

62 points

1 comment14 min readLW link

Evaluation as a (Cooperation-Enabling?) Tool

VojtaKovarik10 Dec 2025 18:54 UTC

13 points

0 comments28 min readLW link

Please, Don’t Roll Your Own Metaethics

Wei Dai12 Nov 2025 22:17 UTC

152 points

64 comments2 min readLW link

Reasoning Models Sometimes Output Illegible Chains of Thought

Jozdien24 Nov 2025 18:24 UTC

82 points

8 comments6 min readLW link