Condensation

abramdemski9 Nov 2025 19:08 UTC

54 points

6 comments15 min readLW link

Problems I’ve Tried to Legibilize

Wei Dai9 Nov 2025 10:27 UTC

39 points

5 comments2 min readLW link

Myopia Mythology

abramdemski8 Nov 2025 22:22 UTC

23 points

1 comment3 min readLW link

Comparing Payor & Löb

abramdemski8 Nov 2025 5:40 UTC

45 points

1 comment3 min readLW link

A scheme to credit hack policy gradient training

Adrià Garriga-alonso7 Nov 2025 6:24 UTC

15 points

0 comments5 min readLW link

Geometric UDT

abramdemski6 Nov 2025 6:24 UTC

28 points

6 comments7 min readLW link

Meta-agentic Prisoner’s Dilemmas

TsviBT5 Nov 2025 16:44 UTC

37 points

1 comment5 min readLW link

Legible vs. Illegible AI Safety Problems

Wei Dai4 Nov 2025 21:39 UTC

258 points

78 comments2 min readLW link

GDM: Consistency Training Helps Limit Sycophancy and Jailbreaks in Gemini 2.5 Flash

TurnTrout and Rohin Shah

4 Nov 2025 16:25 UTC

40 points

2 comments6 min readLW link

(arxiv.org)

Weak-To-Strong Generalization

abramdemski2 Nov 2025 2:45 UTC

28 points

0 comments9 min readLW link

Anthropic’s Pilot Sabotage Risk Report

dmz30 Oct 2025 17:50 UTC

32 points

2 comments3 min readLW link

(alignment.anthropic.com)

Sonnet 4.5′s eval gaming seriously undermines alignment evals, and this seems caused by training on alignment evals

Alexa Pan and ryan_greenblatt

30 Oct 2025 15:34 UTC

139 points

20 comments14 min readLW link

Steering Evaluation-Aware Models to Act Like They Are Deployed

Tim Hua, andrq, Sam Marks and Neel Nanda

30 Oct 2025 15:03 UTC

58 points

12 comments16 min readLW link

Should AI Developers Remove Discussion of AI Misalignment from AI Training Data?

Alek Westover23 Oct 2025 15:12 UTC

37 points

3 comments9 min readLW link

Technical Acceleration Methods for AI Safety: Summary from October 2025 Symposium

Martin Leitgab22 Oct 2025 21:33 UTC

25 points

2 comments6 min readLW link

Reducing risk from scheming by studying trained-in scheming behavior

ryan_greenblatt16 Oct 2025 16:16 UTC

32 points

0 comments11 min readLW link

Rogue internal deployments via external APIs

Fabien Roger and Buck

15 Oct 2025 19:34 UTC

34 points

4 comments6 min readLW link

Current Language Models Struggle to Reason in Ciphered Language

Fabien Roger and Shiyuan Guo

14 Oct 2025 9:08 UTC

77 points

6 comments5 min readLW link

Recontextualization Mitigates Specification Gaming Without Modifying the Specification

ariana_azarbal, Victor Gillioz, TurnTrout and cloud

14 Oct 2025 0:53 UTC

115 points

13 comments9 min readLW link

Iterated Development and Study of Schemers (IDSS)

ryan_greenblatt10 Oct 2025 14:17 UTC

41 points

1 comment8 min readLW link