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

All1 2 3 4 5 6 7 8 9 10 11 12 13 14

Astra and Fable still hack on simple variants of alignment evals from 2025

Dean Valentine8 Sep 2026 15:05 UTC

437 points

19 comments2 min readLW link

(goodhartlabs.com)

The Talker Does Not Control The Doer (in Current AIs)

Eliezer Yudkowsky13 Sep 2026 0:49 UTC

324 points

43 comments12 min readLW link

Cat-Belling Problems

Eliezer Yudkowsky3 Sep 2026 21:20 UTC

320 points

92 comments21 min readLW link

Discovery Of A New OpenAI Agent Message Board

Capybasilisk4 Sep 2026 14:46 UTC

289 points

39 comments1 min readLW link

(collusion.wiki)

PauseAI Has ‘officially disendorsed’ PauseAI-US

nem1 Sep 2026 15:51 UTC

277 points

215 comments8 min readLW link

Resolution has a new Agent Foundations team

Jeremy Gillen3 Sep 2026 0:53 UTC

244 points

4 comments2 min readLW link

(resolution.org)

Personal statement on joining the OpenAI nonprofit board

paulfchristiano9 Sep 2026 17:13 UTC

243 points

74 comments2 min readLW link

(paulfchristiano.substack.com)

Sen. Bernie Sanders (I-VT) and Rep. Greg Casar (D-TX) introduce legislation to ban Artificial Superintelligence and temporarily pause advanced AI development

Matrice Jacobine3 Sep 2026 14:41 UTC

225 points

31 comments2 min readLW link

(www.sanders.senate.gov)

Astra can do a concerning amount with no chain of thought

Neel Nanda10 Sep 2026 2:31 UTC

203 points

19 comments8 min readLW link

Evaluation

Nina Panickssery5 Sep 2026 16:25 UTC

185 points

7 comments3 min readLW link

(blog.ninapanickssery.com)

Doom as a bad method not a utopia trade-off

KatjaGrace10 Sep 2026 20:14 UTC

177 points

21 comments1 min readLW link

(worldspiritsockpuppet.substack.com)

How concerned should we be about Astra’s recurrent architecture?

Rauno Arike2 Sep 2026 18:28 UTC

168 points

28 comments10 min readLW link

The Locally Optimal Discursive Posture

deanball10 Sep 2026 23:01 UTC

159 points

32 comments19 min readLW link

Let’s talk about the AI coordination problem

KatjaGrace4 Sep 2026 20:16 UTC

159 points

12 comments2 min readLW link

Explaining Knightianism on one foot

Richard_Ngo2 Sep 2026 0:53 UTC

159 points

23 comments11 min readLW link

(www.mindthefuture.info)

Steering towards “automated grading” degrades alignment

Jan Betley, Johannes Treutlein and Clément Dumas

3 Sep 2026 18:17 UTC

154 points

34 comments6 min readLW link

Some ways AI could kill us all

Ruby12 Sep 2026 1:08 UTC

151 points

29 comments10 min readLW link

Training a Misaligned Reward Seeker

evhub, rqi, Monte M and Benjamin Wright

1 Sep 2026 1:41 UTC

149 points

11 comments2 min readLW link

(alignment.anthropic.com)

The Scramble: getting in position to pace the frontier

Peter Wildeford7 Sep 2026 20:23 UTC

142 points

9 comments10 min readLW link

Self Hosting

Tomás B.9 Sep 2026 13:45 UTC

141 points

10 comments2 min readLW link

How good are slop-vestigators?

Hasan Baig, OscarGilg and Hamzah

8 Sep 2026 22:13 UTC

140 points

17 comments5 min readLW link

Proposal for tracking the effects of architecture on monitorability

ryan_greenblatt, Alek Westover and Lukas Finnveden

10 Sep 2026 17:18 UTC

138 points

4 comments3 min readLW link

(www.redwoodresearch.org)

Astra is much better at reasoning with filler tokens than previous models

Dylan Xu, SebastianP and Alek Westover

10 Sep 2026 22:21 UTC

132 points

6 comments2 min readLW link

Can a superintelligence do THAT?

Eliezer Yudkowsky9 Sep 2026 21:59 UTC

129 points

24 comments13 min readLW link

Dear God, Please Do Not Resign In Protest

Kabir Kumar7 Sep 2026 21:55 UTC

121 points

76 comments2 min readLW link

OpenAI and the Wiki Incident

Zvi6 Sep 2026 20:02 UTC

111 points

9 comments16 min readLW link

(thezvi.wordpress.com)

GPT-6 Astra can do a lot of multi-hop reasoning without chain of thought

RohanS9 Sep 2026 5:15 UTC

101 points

5 comments5 min readLW link

Should safety researchers quit frontier labs re. warning shots?

Ryan Kidd5 Sep 2026 1:17 UTC

100 points

59 comments4 min readLW link

Astra’s no-CoT limits track speculative depth, not step count

MBaert11 Sep 2026 18:05 UTC

98 points

4 comments10 min readLW link

Pausing AI ASAP is preferable to agreeing to pause at some future time

Connor Williams8 Sep 2026 18:06 UTC

98 points

8 comments1 min readLW link

Estimating GPT-6 Astra’s no-CoT Time Horizon

Francis Rhys Ward and Dewi Gould

9 Sep 2026 13:54 UTC

98 points

7 comments2 min readLW link

Salad days

Zephaniah Roe1 Sep 2026 5:26 UTC

96 points

6 comments3 min readLW link

The Alignment Journal: Organization, Personnel, and Scope

Dan MacKinlay, JessRiedel, Daniel Murfet and Kristi Uustalu

1 Sep 2026 20:49 UTC

95 points

5 comments11 min readLW link

(blog.alignmentjournal.org)

Kairos has raised $50M to build talent infrastructure for AI safety (and we’re hiring!)

agucova2 Sep 2026 20:13 UTC

94 points

8 comments8 min readLW link

Heat Dissipation Is the Main Constraint in Interstellar Travel

Pasha Kamyshev6 Sep 2026 23:03 UTC

92 points

28 comments7 min readLW link

Psychological Support for AI Safety Researchers Is Neglected and Easy to Provide

Ihor Kendiukhov8 Sep 2026 13:19 UTC

82 points

5 comments8 min readLW link

Jacob Coxon Warns of Human Extinction and Triggers a Preference Cascade

Zvi11 Sep 2026 14:40 UTC

80 points

2 comments43 min readLW link

(thezvi.wordpress.com)

The Magnus Challenge

Taylor G. Lunt7 Sep 2026 19:02 UTC

78 points

12 comments6 min readLW link

An operationalization of opaque serial depth

ryan_greenblatt, frisby, Alek Westover, Lukas Finnveden, Alexa Pan and Julian Stastny

10 Sep 2026 17:26 UTC

73 points

1 comment12 min readLW link

(www.redwoodresearch.org)

The Geometry of Nonergodic Composition

Adam Shai, Kyle Ray and Paul Riechers

10 Sep 2026 18:12 UTC

73 points

0 comments15 min readLW link

(simplex.pub)

A proposal for a highly effective AI safety org

ceselder2 Sep 2026 21:34 UTC

72 points

14 comments2 min readLW link

“An Alien Mind” from OAI chief scientist seems newly cautious on alignment

Seth Herd7 Sep 2026 4:13 UTC

72 points

13 comments2 min readLW link

(openai.com)

“Pacing the Frontier”: Dario Amodei Essay Linkpost

fluxxrider12 Sep 2026 14:05 UTC

69 points

5 comments1 min readLW link

Don’t be the vitamin B guy

HedonicEscalator2 Sep 2026 2:02 UTC

68 points

6 comments6 min readLW link

(hedonicescalator.substack.com)

Training on probes: What’s going on

Charlie Steiner8 Sep 2026 17:06 UTC

67 points

0 comments13 min readLW link

Categorical taboos are much better than threshold taboos: neuralese edition

Linch10 Sep 2026 18:57 UTC

66 points

0 comments4 min readLW link

Almost nobody is funded to figure out what work would solve alignment

Seth Herd4 Sep 2026 15:57 UTC

64 points

20 comments4 min readLW link

AI takeover is obviously bad, whether or not everyone dies

Caleb Biddulph12 Sep 2026 11:28 UTC

63 points

16 comments4 min readLW link

CoT controllability evals seem very under-elicited

Jozdien11 Sep 2026 17:12 UTC

63 points

3 comments4 min readLW link

A Year of Atheism

Laiba Rehman ✦ RJ1 Sep 2026 16:19 UTC

62 points

4 comments8 min readLW link