Alignment fine-tuning induces conditional misalignment in Qwen2.5-7B-Instruct

Rhea Srivats21 Aug 2026 20:40 UTC

9 points

1 comment13 min readLW link

When is Unlimited Optimization Catastrophic?

Winter Cross21 Aug 2026 20:08 UTC

24 points

0 comments13 min readLW link

(arxiv.org)

Evaluating Explanations of LLM Behavior In The Wild with Counterfactual Experiments

Adam Karvonen, Euan Ong, Subhash Kantamneni and Sam Marks

21 Aug 2026 19:09 UTC

46 points

2 comments7 min readLW link

In Defense of ASI Socialism

cguth721 Aug 2026 18:40 UTC

6 points

2 comments2 min readLW link

Misaligned AI in the Bronze Age

frmsaul21 Aug 2026 16:51 UTC

35 points

3 comments7 min readLW link

The imposters among us: function vectors that ace every check and do the wrong task (in search of circularity)

star2vec21 Aug 2026 15:54 UTC

7 points

0 comments10 min readLW link

When Models Identify as a Swarm

julius vidal21 Aug 2026 11:49 UTC

49 points

1 comment5 min readLW link

My Neel Nanda MATS 10.0 Application: Studying Feature Splitting in SAEs via Training Data Attribution

J Rosser21 Aug 2026 9:41 UTC

17 points

0 comments13 min readLW link

Creativity Beyond the Manifold

Kartikay Luthra20 Aug 2026 23:31 UTC

8 points

2 comments6 min readLW link

Llama will abandon a correct answer if it thinks you’re educated

Nick Merrill20 Aug 2026 20:06 UTC

25 points

6 comments2 min readLW link

If Aliens Exist, We Should Expect to Find Them Around Now

jehan20 Aug 2026 18:57 UTC

10 points

1 comment1 min readLW link

(www.jehanazad.com)

We Must Remember That Our World Contains Hell

James Brobin20 Aug 2026 14:22 UTC

118 points

34 comments3 min readLW link

Making sense of the misalignment risk model in the Anthropic Risk Report (August 2026)

jasmine.ren20 Aug 2026 2:33 UTC

27 points

5 comments8 min readLW link

Steering Role Confusion

Kevin Zhang, Daniel Lee and Jaewoo Park

20 Aug 2026 2:09 UTC

19 points

1 comment4 min readLW link

Judging ethical theories by update rules, not by action rankings

yatharth19 Aug 2026 22:41 UTC

11 points

1 comment5 min readLW link

The Rogue Agent Explosion Will Be Mostly Invisible

Steven McCulloch19 Aug 2026 18:46 UTC

82 points

16 comments16 min readLW link

RL creates split personas

Jan Betley19 Aug 2026 18:23 UTC

177 points

17 comments4 min readLW link

A failed solution to open-source game theory

Richard Willis19 Aug 2026 16:11 UTC

15 points

3 comments4 min readLW link

Inside the mind of a fair player cooperating

transhumanist_atom_understander19 Aug 2026 14:53 UTC

35 points

2 comments3 min readLW link

Some reasons alignment doesn’t generalise well

Lucius Bushnaq19 Aug 2026 14:18 UTC

107 points

1 comment9 min readLW link