Investigating the consequences of accidentally grading CoT during RL

papetoast8 May 2026 6:17 UTC

7 points

0 comments1 min readLW link

(alignment.openai.com)

The Frictionless Double

zw57 May 2026 23:11 UTC

6 points

4 comments8 min readLW link

Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations

Subhash Kantamneni, kitft, Euan Ong and Sam Marks

7 May 2026 20:21 UTC

142 points

13 comments8 min readLW link

Axes of Planning in LLMs + Partial Lit Review

NickyP7 May 2026 19:53 UTC

12 points

0 comments9 min readLW link

(blog.sus.cat)

A review of “Investigating the consequences of accidentally grading CoT during RL”

Buck7 May 2026 18:06 UTC

62 points

0 comments8 min readLW link

Try, even if they have you cold

WalterL7 May 2026 17:19 UTC

75 points

4 comments2 min readLW link

Mechanistic estimation for wide random MLPs

Jacob_Hilton7 May 2026 16:20 UTC

51 points

1 comment5 min readLW link

(www.alignment.org)

How to get better at chess (and everything else)

Sean Herrington7 May 2026 11:17 UTC

10 points

0 comments3 min readLW link

(www.chess.com)

Multipolar Civilisation Depends on Maintaining an Attacker’s Dilemma

Naci Cankaya7 May 2026 11:13 UTC

19 points

1 comment5 min readLW link

(nacicankaya.substack.com)

Sculpted Interaction: a Design-First Approach to AI Alignment

magfrump6 May 2026 23:47 UTC

14 points

0 comments7 min readLW link

Psychopathy: The Choice

Dawn Drescher6 May 2026 22:23 UTC

11 points

0 comments17 min readLW link

(impartial-priorities.org)

Many individual CEVs are probably quite bad

Viliam6 May 2026 20:18 UTC

89 points

27 comments3 min readLW link

Blind deep-deployment evals for control & sabotage

Dylan Bowman6 May 2026 19:54 UTC

23 points

0 comments2 min readLW link

A draft honesty policy for credible communication with AI systems

Mia Taylor, Lukas Finnveden and Max Dalton

6 May 2026 18:50 UTC

3 points

0 comments13 min readLW link

(www.forethought.org)

x-risk-themed

kave6 May 2026 15:16 UTC

117 points

9 comments3 min readLW link

(kaverennedy.substack.com)

Monday AI Radar #24

Against Moloch6 May 2026 15:05 UTC

10 points

3 comments8 min readLW link

(againstmoloch.substack.com)

AI Safety at the Frontier: Paper Highlights of April 2026

gasteigerjo6 May 2026 13:58 UTC

16 points

1 comment10 min readLW link

There is no evidence you should reapply sunscreen every 2 hours.

Hide6 May 2026 9:19 UTC

47 points

10 comments9 min readLW link

(hidefromit.substack.com)

Building An Ancestor Simulation #2

Mira Kennard6 May 2026 8:21 UTC

5 points

0 comments5 min readLW link

Psychopathy: The Types

Dawn Drescher6 May 2026 7:35 UTC

1 point

0 comments10 min readLW link

(impartial-priorities.org)