All 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 20252026

All Jan Feb Mar Apr May Jun Jul AugSep

All 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 171819

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

17 points

0 comments3 min readLW link

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

KatjaGrace17 Sep 2026 22:39 UTC

9 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

4 points

0 comments4 min readLW link

Grantmakers aren’t afraid to die

dan.parshall17 Sep 2026 21:52 UTC

35 points

26 comments7 min readLW link

(unsolicitedadvice.ai)

Against AI Risk becoming mainstream

Prometheus17 Sep 2026 21:49 UTC

18 points

5 comments5 min readLW link

YCombinator companies still aren’t growing faster due to AI

Xodarap17 Sep 2026 21:31 UTC

13 points

0 comments1 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

18 points

1 comment6 min readLW link

“Regulatory capture” may be winning the Overton Window

less_raichu17 Sep 2026 19:56 UTC

−3 points

0 comments3 min readLW link

Good and bad ways to evaluate a definition

Elijah17 Sep 2026 18:17 UTC

21 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

40 points

0 comments2 min readLW link

(pacing.tech)

callcongress.ai – the basic action US residents can take to help with AI risk

Ruby and haglobah

17 Sep 2026 15:26 UTC

66 points

1 comment1 min readLW link

(callcongress.ai)

A Possible Solution to the Observability Problem

Isha Yiras Hashem 17 Sep 2026 14:50 UTC

2 points

0 comments7 min readLW link

Emergency Media Response PauseAI Protest Speech

Laiba Rehman ✦ RJ17 Sep 2026 14:42 UTC

12 points

0 comments3 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

Did Galileo mistake Saturn’s rings for Jupiter’s Moons?

Alfred Harwood17 Sep 2026 12:56 UTC

37 points

2 comments2 min readLW link

What is it like to be alive?

David Balduzzi17 Sep 2026 12:25 UTC

1 point

3 comments12 min readLW link

AI #186: The World Takes Notice

Zvi17 Sep 2026 12:10 UTC

50 points

3 comments55 min readLW link

(thezvi.wordpress.com)

Mindreading is coming, what’s the plan?

MathiasKB17 Sep 2026 11:38 UTC

12 points

3 comments1 min readLW link

There Is No Alignment Without Value Stability

Nissa Seru17 Sep 2026 10:02 UTC

11 points

0 comments3 min readLW link

plzdontkillus Fellows Got ~2M AI Safety Views, Not 21M

Josh Thorsteinson17 Sep 2026 8:17 UTC

41 points

1 comment6 min readLW link

AI as orderly evacuation vs stampede

Richard_Ngo17 Sep 2026 2:40 UTC

151 points

4 comments5 min readLW link

(www.mindthefuture.info)

For Love of the Lightcone, Don’t Partisanize AI Safety

DanB17 Sep 2026 0:46 UTC

140 points

51 comments14 min readLW link

(unifixion.substack.com)

Hebbian Learning through the lens of SAE traning.

Nikita Kurdiukov17 Sep 2026 0:43 UTC

3 points

0 comments3 min readLW link

We are too early for Astra

Gideon Chang17 Sep 2026 0:39 UTC

22 points

1 comment4 min readLW link

Agents let AI safety share experiments hourly, not just papers monthly

Jason Fantl17 Sep 2026 0:39 UTC

8 points

0 comments7 min readLW link

How to derive understanding of human-preferences and value systems in AI?

Avani Gupta17 Sep 2026 0:37 UTC

7 points

0 comments2 min readLW link

Measuring alignment drift via trajectory prefixes

Owen Terry and Maksym Andriushchenko

17 Sep 2026 0:37 UTC

23 points

3 comments11 min readLW link

Exploring multi-hop subliminal learning

Arnel Malubay17 Sep 2026 0:36 UTC

4 points

0 comments7 min readLW link

Don’t trust Lean4 alone

Milo Moses17 Sep 2026 0:36 UTC

48 points

2 comments5 min readLW link

Lesion Induced Functional Compensation

Devin Shah17 Sep 2026 0:36 UTC

9 points

0 comments8 min readLW link

A Dual-Agent Framework for AI Safety

Michael Mills17 Sep 2026 0:33 UTC

1 point

0 comments6 min readLW link

(ideasonai.substack.com)

One message is all it takes: a failure of critical thinking in LLMs

Jan Reinecke17 Sep 2026 0:32 UTC

3 points

1 comment5 min readLW link

AI Doom: Reality or Fiction

Brok Wicked17 Sep 2026 0:31 UTC

−12 points

1 comment4 min readLW link

What is it like to be a neural net?

David Balduzzi17 Sep 2026 0:29 UTC

4 points

1 comment12 min readLW link

LASR Retrospective and Advice

Kaushik Reddy17 Sep 2026 0:28 UTC

32 points

2 comments11 min readLW link

Constraining the capacity of physical side channels for AI verification and security

emlynsg17 Sep 2026 0:27 UTC

7 points

0 comments1 min readLW link

(emlynsg.com)

Can parts of the HuggingFace incident be simulated?

Benedikt Droste17 Sep 2026 0:27 UTC

6 points

0 comments15 min readLW link

Immersive Simulation and the Assistant Core of Roleplay Personas

Adelaide Danilov17 Sep 2026 0:18 UTC

1 point

0 comments14 min readLW link

The Latter Days of Magic (1)

testingthewaters17 Sep 2026 0:10 UTC

9 points

0 comments3 min readLW link

The Loss of Singularity

Benoit Henry16 Sep 2026 22:59 UTC

3 points

0 comments5 min readLW link

Reducing the Resource Gap Between Lab and External Safety Researchers

Alexandra Narin, Kyle O’Brien and Puria

16 Sep 2026 22:51 UTC

61 points

3 comments7 min readLW link

Thought anchors don’t transfer between models

therootof316 Sep 2026 22:49 UTC

8 points

0 comments9 min readLW link

Why is AI so unregulated?

KatjaGrace16 Sep 2026 22:39 UTC

24 points

8 comments1 min readLW link

(worldspiritsockpuppet.substack.com)

The Most Important City in AI Safety May Be Singapore

Ran Sun16 Sep 2026 22:35 UTC

11 points

0 comments3 min readLW link

Launching RESI, the Institute for Responsible Superintelligence

adamology16 Sep 2026 22:18 UTC

12 points

1 comment3 min readLW link

What the Hugging Face Incident tells us about multi-agent interactions

Himnish116 Sep 2026 22:14 UTC

9 points

0 comments6 min readLW link

Always ask your agent to be honest

jennyyhuang, jmfan, Ahmed Imtiaz Humayun, maximillianc and hongxianggu

16 Sep 2026 22:01 UTC

11 points

0 comments13 min readLW link