The Game is Set for a Targeted Memetic Attack on the AI Safety Community

keltan18 Sep 2026 5:43 UTC

48 points

5 comments1 min readLW link

The Horse

Character#273618 Sep 2026 2:52 UTC

25 points

1 comment3 min readLW link

Deep recurrent models are less robustly CoT-monitorable than normal CoT models in a toy setting

Nick Kuhn and Alek Westover

18 Sep 2026 2:47 UTC

72 points

0 comments14 min readLW link

Machine intelligence and the death of human expression

Girard Dorney18 Sep 2026 2:10 UTC

5 points

0 comments6 min readLW link

(extinctiondesk.substack.com)

The Cost of Utopias (a Dialog)

WillPetillo18 Sep 2026 1:57 UTC

18 points

2 comments17 min readLW link

Two Axes of Alignment: A Framework for Robust Superintelligence Alignment

Arihant Gadgade18 Sep 2026 0:59 UTC

6 points

0 comments4 min readLW link

Superintelligence this Christmas

Alexander Gietelink Oldenziel18 Sep 2026 0:06 UTC

46 points

7 comments3 min readLW link

What is (and isn’t) gained by avoiding architectures with high opaque serial depth?

Alek Westover17 Sep 2026 23:52 UTC

11 points

0 comments3 min readLW link

If METR is overworked, how to alleviate the bottleneck?

Matthew_Opitz17 Sep 2026 22:56 UTC

16 points

0 comments3 min readLW link

AI is an abundance of choice not a 1D spectrum

KatjaGrace17 Sep 2026 22:39 UTC

7 points

0 comments1 min readLW link

(worldspiritsockpuppet.substack.com)

AI can kill us without human extinction: P(Catastrophe)

Young Jae Koh17 Sep 2026 22:22 UTC

3 points

0 comments4 min readLW link

Grantmakers aren’t afraid to die

dan.parshall17 Sep 2026 21:52 UTC

33 points

14 comments5 min readLW link

(unsolicitedadvice.ai)

Against AI Risk becoming mainstream

Prometheus17 Sep 2026 21:49 UTC

15 points

3 comments5 min readLW link

The J-lens offset is the model’s token frequency: z-scoring helps

Ameya Panchal17 Sep 2026 21:13 UTC

7 points

0 comments10 min readLW link

(ameya-bit.github.io)

A Defense of Gradual Disempowerment

Max Harms17 Sep 2026 21:04 UTC

19 points

3 comments6 min readLW link

Swarm Organization as the Exponent on Test-Time Compute

Julian Bradshaw17 Sep 2026 20:17 UTC

9 points

1 comment6 min readLW link

Good and bad ways to evaluate a definition

Elijah17 Sep 2026 18:17 UTC

20 points

0 comments4 min readLW link

Pacing the Frontier: A Framework & Research Agenda

CharlesD, technicalities, Raymond Douglas and Nowe Moore

17 Sep 2026 16:36 UTC

34 points

0 comments2 min readLW link

(pacing.tech)

A Possible Solution to the Observability Problem

Isha Yiras Hashem 17 Sep 2026 14:50 UTC

2 points

0 comments7 min readLW link

Astra uses some of its no-CoT capability in practice

StevenW17 Sep 2026 14:21 UTC

9 points

2 comments3 min readLW link