Claude’s malicious compliance and normalization of deviance

Steff5 Jul 2026 17:18 UTC

17 points

0 comments8 min readLW link

Probing the loss-band sparsity assumption in Scientist AI

Alejandro Tlaie5 Jul 2026 16:25 UTC

8 points

0 comments7 min readLW link

Book Review: The God Test

PeterMcCluskey5 Jul 2026 16:11 UTC

14 points

0 comments4 min readLW link

We need 3rd party Training-Run Assessments

Alex Meinke5 Jul 2026 15:55 UTC

26 points

0 comments10 min readLW link

Harry Potter and the Rules of Quidditch

Tomás B.5 Jul 2026 14:32 UTC

57 points

3 comments3 min readLW link

A Normal Argument for AI Risk

Silent Swift5 Jul 2026 9:32 UTC

15 points

0 comments8 min readLW link

(silentswift.substack.com)

The case for the fleshman

dr_s5 Jul 2026 9:28 UTC

14 points

11 comments4 min readLW link

Reevaluating AI-2027: timelines, takeoff, alignment and China

StanislavKrym5 Jul 2026 4:00 UTC

14 points

2 comments5 min readLW link

Success Per Tokens

michaelwaves5 Jul 2026 2:25 UTC

8 points

0 comments3 min readLW link

A case for LLMs as Self-predictors

Ashe Vazquez Nuñez5 Jul 2026 0:25 UTC

30 points

0 comments10 min readLW link

Defining interpretation, and establishing a framework for it

Yaroven4 Jul 2026 16:31 UTC

6 points

0 comments5 min readLW link

The Lace (short story)

Michael Soareverix4 Jul 2026 4:43 UTC

22 points

0 comments4 min readLW link

Approximate Natural Latents Have Exact Prices

Haru4 Jul 2026 1:57 UTC

18 points

0 comments6 min readLW link

I think alignment work is more promising than control work

Alec Harris3 Jul 2026 23:40 UTC

85 points

9 comments8 min readLW link

On “gendertropes” in dath ilan

Eliezer Yudkowsky3 Jul 2026 22:20 UTC

59 points

1 comment3 min readLW link

American AI if the boom is a bubble: the Karp-Zitron scenario

Mitchell_Porter3 Jul 2026 21:46 UTC

6 points

1 comment2 min readLW link

(Don’t fear) the strangelet

djbinder3 Jul 2026 17:39 UTC

101 points

2 comments22 min readLW link

(defensesindepth.bio)

The Reverse AI Box

James_Miller3 Jul 2026 16:08 UTC

9 points

1 comment6 min readLW link

Pragmatic FDT, and predictors as game theory

Stuart_Armstrong3 Jul 2026 13:22 UTC

32 points

10 comments11 min readLW link

Scheming Evals Mislead in Both Directions

Chijioke Ugwuanyi, eric-z and TerryJCZhang

3 Jul 2026 11:49 UTC

21 points

0 comments10 min readLW link