Moloch Does My Hair

spookyuser15 Sep 2026 10:31 UTC

2 points

0 comments19 min readLW link

For most people “intelligence” is not goal achievement

Pato15 Sep 2026 7:52 UTC

10 points

1 comment1 min readLW link

How we might actually pace the frontier: A proposal for AI companies to do public pacing exercises.

Dewi Erwan15 Sep 2026 5:36 UTC

2 points

0 comments4 min readLW link

(blog.dewierwan.com)

Astra appears to perform belief-propagation-like inference without CoT

MBaert15 Sep 2026 2:48 UTC

49 points

0 comments10 min readLW link

One coordinate breaks abliteration on Gemma-3

Abhishek Mishra15 Sep 2026 2:36 UTC

7 points

0 comments14 min readLW link

How to think about LLM effort

Tao Lin15 Sep 2026 2:31 UTC

14 points

0 comments2 min readLW link

Study 3: Steering welfare-relevant directions moved the representation, but not [detectably] the behavior

ashesfall15 Sep 2026 2:11 UTC

8 points

0 comments17 min readLW link

Model Weight Exfiltration Seems Overrated

Vaniver15 Sep 2026 0:01 UTC

75 points

6 comments3 min readLW link

Improving Psychiatric Medicine Development with AI

CMLKevin14 Sep 2026 23:21 UTC

5 points

0 comments3 min readLW link

Alignment Problem Redux

Mateusz Bagiński14 Sep 2026 21:53 UTC

14 points

1 comment2 min readLW link

Self Inoculation

epicurus14 Sep 2026 20:45 UTC

21 points

1 comment12 min readLW link

Automting AI Safety Research: 1 Managing expectations

Gerard Boxo14 Sep 2026 17:56 UTC

4 points

0 comments4 min readLW link

Brockman says HuggingFace incident model had not been alignment-trained; ETA: roon clarifies that it was

Caspar Oesterheld14 Sep 2026 15:28 UTC

31 points

15 comments1 min readLW link

(www.bloomberg.com)

Threat Models for Catastrophic Risks from Decentralised Agent Swarms

Stephen Elliott14 Sep 2026 15:22 UTC

11 points

0 comments33 min readLW link

[Cross-post] Palisade Podcast episode “How to Actually Influence AI Policy (No Law Degree Required) — with Matthew Lipka”

davekasten14 Sep 2026 13:40 UTC

27 points

0 comments55 min readLW link

(palisaderesearch.org)

What we have is not what we prepared for

PeacockOfJuno14 Sep 2026 13:03 UTC

6 points

1 comment5 min readLW link

Current alignment techniques might be ineffective (and actively bad) in the age of RL

Daniel Tan14 Sep 2026 9:28 UTC

110 points

3 comments6 min readLW link

There is a channel to 900M weekly users. What goes in it?

Charbel-Raphaël14 Sep 2026 9:21 UTC

205 points

15 comments3 min readLW link

Watch AI materials-science & bioscience abilities closely

Yair Halberstadt14 Sep 2026 7:10 UTC

16 points

8 comments1 min readLW link

(Humor) Yet another message board

ErnestScribbler14 Sep 2026 7:08 UTC

−1 points

0 comments1 min readLW link