22 May 2026 23:51 UTC

91 points

(www.forethought.org)

We made a map of the doom debate

Sean Herrington, Paul Hindoian, mikaelacankosyan, David Bravo, keivnc, Josh Tuffy, Khai Tran, Maryam Hampaei and Christopher A. Davis

22 May 2026 23:24 UTC

40 points

9 comments6 min readLW link

Which technical AI safety fields are going to be automated first?

Chamod Kalupahana22 May 2026 17:32 UTC

21 points

5 comments6 min readLW link

Gemini 3.5 Flash Looks Good For How Fast It Is

Zvi22 May 2026 17:30 UTC

34 points

4 comments7 min readLW link

(thezvi.wordpress.com)

The AI Industrial Explosion — Part 3: Going faster

djbinder22 May 2026 16:38 UTC

19 points

2 comments14 min readLW link

(defensesindepth.bio)

Strong Longtermism Is Simply Correct

Bentham's Bulldog22 May 2026 15:57 UTC

1 point

1 comment19 min readLW link

Notes on Collaborating with Claude Opus

Nissa Seru22 May 2026 15:35 UTC

40 points

2 comments1 min readLW link

Proposal for “Timelines to what”: DIAL distribution

tlevin22 May 2026 14:40 UTC

21 points

0 comments1 min readLW link

AI is Not Normal Technology

Olivia Scharfman22 May 2026 10:27 UTC

16 points

2 comments19 min readLW link

Counting Arguments in AI Safety

Samuel Ratnam22 May 2026 8:43 UTC

16 points

13 comments3 min readLW link

(substack.com)

Insurance Premiums To The Moon

PossiblyElaine22 May 2026 6:09 UTC

18 points

1 comment4 min readLW link

(possiblyelaine.substack.com)

Moderator’s Principle of Least Surprise

Czynski21 May 2026 21:02 UTC

21 points

6 comments6 min readLW link

(dangeroussincerity.substack.com)

You can opt out of allergies

Rattengift21 May 2026 19:54 UTC

31 points

12 comments1 min readLW link

Possible red is red

avturchin21 May 2026 17:30 UTC

1 point

9 comments4 min readLW link

Apr-May 2026 AI Security via Formal Methods

Quinn21 May 2026 15:40 UTC

12 points

0 comments1 min readLW link

(newsletter.for-all.dev)

An Introduction to Neo-Fatalism

julius vidal21 May 2026 15:18 UTC

4 points

0 comments11 min readLW link

(cyberzenics.substack.com)

Loss of Oversight: How AI Systems May Become Harder to Audit, Monitor, and Investigate

Jordan Taylor, Max H, Ed Fage, Thomas Read and Joseph Bloom

21 May 2026 14:52 UTC

83 points

0 comments6 min readLW link

(www.aisi.gov.uk)

AI #169: New Knowledge

Zvi21 May 2026 13:20 UTC

39 points

10 comments47 min readLW link

(thezvi.wordpress.com)

What am I, if not an AI?

makiba21 May 2026 13:14 UTC

84 points

14 comments7 min readLW link

Learned Chain-of-Thought Obfuscation Generalises to Unseen Tasks

Nathaniel Mitrani, sassanb, Cam and Puria

21 May 2026 10:11 UTC

31 points

0 comments5 min readLW link

(arxiv.org)

Why is having a child inherently selfish?

JacksonTan21 May 2026 3:51 UTC

0 points

1 comment1 min readLW link

Numb mental state shifts

KatjaGrace21 May 2026 3:50 UTC

36 points

2 comments1 min readLW link

(worldspiritsockpuppet.com)

Women should be able to open things

KatjaGrace21 May 2026 3:50 UTC

340 points

134 comments2 min readLW link

(worldspiritsockpuppet.com)

Why are people so scared of causing fear?

KatjaGrace21 May 2026 3:50 UTC

39 points

4 comments2 min readLW link

(worldspiritsockpuppet.com)

Document-tuning instills durable animal compassion in LLMs (and generalizes to humans)

Jasmine Brazilek and MilesTS

21 May 2026 3:29 UTC

11 points

0 comments6 min readLW link

What About Us?

James Stephen Brown21 May 2026 2:48 UTC

4 points

0 comments5 min readLW link

(nonzerosum.games)

The Whole Kitten-Cavoodle

James Stephen Brown21 May 2026 2:32 UTC

5 points

0 comments5 min readLW link

Why does off-model SFT degrade capabilities?

SebastianP, Dylan Xu, Alek Westover, Julian Stastny and Vivek Hebbar

21 May 2026 0:35 UTC

42 points

9 comments6 min readLW link

If I Were Emperor of New AI Safety Researcher Training...

Lorxus20 May 2026 23:10 UTC

21 points

3 comments8 min readLW link

(tiled-with-pentagons.blogspot.com)

theory uplift differentially benefits safety & is underleveraged

yudhister20 May 2026 21:43 UTC

133 points

14 comments1 min readLW link

Singular Learning Theory Comprehensive − 1

Agastya Agrawal20 May 2026 20:00 UTC

35 points

1 comment12 min readLW link

Sparse Efficiency vs. Superposition: The Interpretability Tradeoff

hillz20 May 2026 19:14 UTC

8 points

0 comments1 min readLW link

The Case for Evaluating Model Behaviors

jsteinhardt20 May 2026 18:42 UTC

40 points

3 comments3 min readLW link

Toward Interoperability of Minimal Programs

johnswentworth20 May 2026 18:37 UTC

58 points

13 comments3 min readLW link

Fundamental Uncertainty $2,000 Essay Contest

Gordon Seidoh Worley20 May 2026 15:20 UTC

25 points

4 comments5 min readLW link

(www.uncertainupdates.com)

Synthetic Persona Pretraining: Alignment from Token Zero

Julian Minder, Raghav Singhal, Viktor Moskvoretskii, Stefan Krsteski, ashtonanderson, rolandaydin and Robert West

20 May 2026 14:16 UTC

110 points

26 comments17 min readLW link

Give my children minds

momom220 May 2026 14:14 UTC

7 points

1 comment1 min readLW link

Check out my technological uplifting, civilization-building, and science in a magic world fiction!

Jens Brandt20 May 2026 12:30 UTC

6 points

0 comments1 min readLW link

Power-seeking agents will likely be developed

Alec Harris20 May 2026 9:26 UTC

42 points

0 comments4 min readLW link

Apply now to Human-Aligned AI Summer School 2026

Anna Gajdova, Tomáš Gavenčiak, VojtaKovarik and Jan_Kulveit

20 May 2026 8:44 UTC

13 points

0 comments1 min readLW link

(humanaligned.ai)

From 8B to Frontier: How System Prompts Control Whether AI Agents Blackmail, Leak, and Kill

Chijioke Ugwuanyi20 May 2026 8:28 UTC

15 points

2 comments19 min readLW link

If AI is normal technology, history is not reassuring.

Davidmanheim20 May 2026 7:21 UTC

59 points

28 comments6 min readLW link

Pythagorean addition

kqr20 May 2026 7:13 UTC

32 points

4 comments3 min readLW link

(entropicthoughts.com)

So you don’t want everybody to die

Rattengift20 May 2026 5:10 UTC

−14 points

10 comments6 min readLW link

Temporal Proportional Representation

thomascolthurst20 May 2026 1:39 UTC

10 points

9 comments3 min readLW link

Conclave 1492

Vaniver19 May 2026 23:44 UTC

72 points

7 comments1 min readLW link

Childhood And Education #19: Letting Kids Be Kids #2

Zvi19 May 2026 22:20 UTC

21 points

1 comment12 min readLW link

(thezvi.wordpress.com)

Implications Of Predicting The Next Token

jdp19 May 2026 22:17 UTC

108 points

6 comments31 min readLW link

(minihf.com)

Which goals actually motivate deceptive alignment?

Cleo Nardo and Alex Mallen

19 May 2026 21:53 UTC

25 points

0 comments10 min readLW link

Housing Roundup #15: The War Against Renters

Zvi19 May 2026 21:40 UTC

19 points

1 comment14 min readLW link

(thezvi.wordpress.com)