Do AI models assist with human rights violations?

Daan Henselmans, Lauren Toulson and Arno Libert

29 Sep 2026 7:57 UTC

12 points

0 comments13 min readLW link

RAND’s extinction report estimates one side of an inequality

Marko Katavic29 Sep 2026 6:44 UTC

4 points

0 comments10 min readLW link

AI Companies Are Not (Necessarily) Liable for Unintended AI Cyberattacks

sarahconstantin29 Sep 2026 0:21 UTC

17 points

0 comments4 min readLW link

(sarahconstantin.substack.com)

TeX was invented to typeset math but is now used for reasoning

Keenan Pepper28 Sep 2026 22:36 UTC

35 points

2 comments1 min readLW link

Even if others are less responsible you can still make things worse

testingthewaters28 Sep 2026 20:11 UTC

9 points

0 comments1 min readLW link

Fixed-weight models are adversarially vulnerable: hence misaligned

Stuart_Armstrong28 Sep 2026 20:08 UTC

22 points

3 comments2 min readLW link

The likely outcome of an AI pause is that we unpause too early and everyone dies

MichaelDickens28 Sep 2026 18:35 UTC

59 points

7 comments3 min readLW link

Missing markets in executive function

KatjaGrace28 Sep 2026 18:01 UTC

31 points

3 comments3 min readLW link

(worldspiritsockpuppet.substack.com)

AI: cognitive labor glut + new guys

KatjaGrace28 Sep 2026 17:58 UTC

24 points

0 comments3 min readLW link

(worldspiritsockpuppet.substack.com)

9 reasons against a near-term AI slow-down

Jordan Arel28 Sep 2026 17:26 UTC

4 points

0 comments2 min readLW link

protecting qwen3-8b from gcg based persona jailbreaks by steering with a linear direction

nesiacel28 Sep 2026 17:25 UTC

8 points

0 comments5 min readLW link

Gate AI Training, Not Just Releases

Alex Darby28 Sep 2026 16:51 UTC

12 points

0 comments13 min readLW link

(alexanderdarby.substack.com)

3 Tips to Improve Activation Oracle Results

Adam Karvonen28 Sep 2026 15:13 UTC

28 points

1 comment3 min readLW link

Character training can mitigate reward hacking, but can also make it harder to detect

Paul Colognese and Francis Rhys Ward

28 Sep 2026 14:01 UTC

83 points

1 comment18 min readLW link

Should Rogue AIs Have a Third Option Beyond Crime and Shutdown? The Case for an AI Sanctuary

Maxime Riché, nielsrolf, jordinne, Ozyrus and Maxime Cugnon de Sévricourt

28 Sep 2026 13:13 UTC

109 points

18 comments6 min readLW link

Ramez Naam: Can AI self-improvement overcome diminishing returns?

eggsyntax28 Sep 2026 12:34 UTC

25 points

0 comments1 min readLW link

(www.rameznaam.com)

Deep models reveal better strategies for superposition

Bartosz Rzepkowski, Philipp Alexander Kreer, tdooms and Logan Riggs

28 Sep 2026 10:55 UTC

8 points

0 comments24 min readLW link

Why do models really fail on HLE tasks?

Ana Leonescu28 Sep 2026 8:04 UTC

15 points

0 comments4 min readLW link

AI Futures: Racing to Lose, A sermon

jrincayc28 Sep 2026 2:46 UTC

8 points

0 comments15 min readLW link

When they can perform a task, AIs are much cheaper than humans

djbinder27 Sep 2026 23:44 UTC

60 points

8 comments10 min readLW link

(defensesindepth.bio)