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

All Jan FebMarApr May Jun Jul Aug Sep

All 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 202122 23 24 25 26 27 28 29 30 31

The Future of Aligning Deep Learning systems will probably look like “training on interp”

williawa20 Mar 2026 23:06 UTC

29 points

7 comments4 min readLW link

An agent autonomously builds a 1.5 GHz Linux-capable RISC-V CPU

sanxiyn20 Mar 2026 23:03 UTC

19 points

2 comments2 min readLW link

(arxiv.org)

Untrusted monitoring: extra bits

Morgan S20 Mar 2026 21:32 UTC

26 points

0 comments15 min readLW link

Finding features in Transformers: Contrastive directions elicit stronger low-level perturbation responses than baselines

Francisco Ferreira da Silva and StefanHex

20 Mar 2026 21:09 UTC

39 points

2 comments6 min readLW link

ARENA 7.0 Impact Report

JScriven and JamesH

20 Mar 2026 17:09 UTC

13 points

0 comments21 min readLW link

The Federal AI Policy Framework: An Improvement, But My Offer Is (Still Almost) Nothing

Zvi20 Mar 2026 16:51 UTC

33 points

0 comments8 min readLW link

(thezvi.wordpress.com)

Confusion around the term reward hacking

ariana_azarbal20 Mar 2026 16:13 UTC

67 points

6 comments5 min readLW link

The Distaff Texts

Tomás B.20 Mar 2026 15:05 UTC

105 points

6 comments14 min readLW link

It’s a Good Thing to Respond to Internet Trolls

Bowl of Cereal20 Mar 2026 14:22 UTC

−10 points

4 comments2 min readLW link

Untrusted Monitoring is Default; Trusted Monitoring is not

J Bostock20 Mar 2026 14:10 UTC

30 points

0 comments4 min readLW link

Against Messianic AI: Why Optimizing the Environment Doesn’t Optimize the Agent

Nathan Heath20 Mar 2026 12:40 UTC

1 point

0 comments3 min readLW link

2nd (Unofficial) ACX Weekend

Fernand020 Mar 2026 12:13 UTC

1 point

0 comments1 min readLW link

Why I am not buying IPv4 addresses as an investment

samuelshadrach20 Mar 2026 9:02 UTC

4 points

2 comments5 min readLW link

(samuelshadrach.com)

Hundred ways a superintelligence could kill you (non-serious exercise)

samuelshadrach20 Mar 2026 8:58 UTC

3 points

1 comment6 min readLW link

(samuelshadrach.com)

Internet anonymity without Tor

samuelshadrach20 Mar 2026 8:52 UTC

1 point

0 comments3 min readLW link

(samuelshadrach.com)

No, You Don’t Need Self-Locating Evidence.

Ape in the coat20 Mar 2026 5:38 UTC

8 points

4 comments5 min readLW link

(substack.com)

The Low Hanging Fruit of AI Self Improvement

HunterJay20 Mar 2026 4:09 UTC

1 point

0 comments5 min readLW link

Nullius in Verba: 3rd party evidence for Nectome’s Brain Preservation

Aurelia20 Mar 2026 3:19 UTC

179 points

18 comments12 min readLW link

Does Hebrew Have Verbs?

Benquo20 Mar 2026 3:04 UTC

37 points

9 comments6 min readLW link

(benjaminrosshoffman.com)

Positive-sum interactions between players with linear utility in resources

Cleo Nardo20 Mar 2026 0:42 UTC

12 points

0 comments2 min readLW link

A letter to the Editor:

Richard Pickering19 Mar 2026 23:57 UTC

−1 points

2 comments3 min readLW link

No, we haven’t uploaded a fly yet

Ariel Zeleznikow-Johnston19 Mar 2026 23:43 UTC

243 points

8 comments8 min readLW link

(open.substack.com)

The Case for Low-Competence ASI Failure Scenarios

Ihor Kendiukhov19 Mar 2026 23:10 UTC

146 points

8 comments6 min readLW link

A List of Research Directions in Character Training

Rauno Arike19 Mar 2026 22:58 UTC

49 points

21 comments8 min readLW link

“The AI Doc” is coming out March 26

Rob Bensinger and Beckeck

19 Mar 2026 22:55 UTC

189 points

2 comments1 min readLW link

Helical Representations of Turn Structure in an LLM

Koby Lewis19 Mar 2026 22:39 UTC

7 points

0 comments10 min readLW link

(kobylewis.net)

Separating Prediction from Goal-Seeking

plex19 Mar 2026 21:47 UTC

34 points

0 comments2 min readLW link

Null Results From An Orexin RCT

niplav, harsimony and nomagicpill

19 Mar 2026 21:14 UTC

95 points

27 comments3 min readLW link

Teaching Models to Dream of Better Monitors through Evaluation Conditioned Training

Alec Harris, Kasey C, Archie Chaudhury and yix

19 Mar 2026 21:01 UTC

53 points

2 comments10 min readLW link

Protecting humanity and Claude from rationalization and unaligned AI

Kaj_Sotala19 Mar 2026 21:00 UTC

84 points

7 comments5 min readLW link

(kajsotala.substack.com)

Broad Timelines

Toby_Ord19 Mar 2026 19:05 UTC

191 points

25 comments16 min readLW link

Miniature Cities Should Not Be Islands

Novalis19 Mar 2026 18:32 UTC

−3 points

0 comments4 min readLW link

(minicities.org)

OpenAI: How we monitor internal coding agents for misalignment

Marcus Williams19 Mar 2026 17:27 UTC

95 points

19 comments1 min readLW link

(openai.com)

Should You Sign Up for Cryonics? Interactive EV calculator

Mikhail Samin19 Mar 2026 17:13 UTC

34 points

0 comments7 min readLW link

On restraining AI development for the sake of safety

Joe Carlsmith19 Mar 2026 16:30 UTC

29 points

6 comments50 min readLW link

The Vatican, AI Legal Personhood, and Claude’s Constitution — Digital Minds Newsletter #2

lucius and tbs

19 Mar 2026 16:22 UTC

11 points

1 comment29 min readLW link

Contra Anil Seth on AI Consciousness

Against Moloch19 Mar 2026 16:16 UTC

22 points

10 comments4 min readLW link

(againstmoloch.com)

AI #160: What Passes For a Pause

Zvi19 Mar 2026 14:40 UTC

36 points

6 comments52 min readLW link

(thezvi.wordpress.com)

Subscriber count graphs of popular youtubers on ASI risk

samuelshadrach19 Mar 2026 13:51 UTC

16 points

1 comment1 min readLW link

(samuelshadrach.com)

What should we think about shard theory in light of chain-of-thought agents?

Chris_Leong19 Mar 2026 6:48 UTC

25 points

3 comments1 min readLW link

“We’ve been fine before, so we’ll be fine again” is a fallacy (in the more dangerous direction).

Chapin Lenthall-Cleary19 Mar 2026 4:57 UTC

11 points

1 comment3 min readLW link

The “talker–feeler gap”: AI valence may be unknowable

david reinstein19 Mar 2026 4:30 UTC

11 points

0 comments10 min readLW link

Latent Reasoning Sprint #2: Token-Based Signals and Linear Probes

Realmbird19 Mar 2026 3:39 UTC

6 points

0 comments3 min readLW link

An interactive version of the extropians mailing list

beyarkay (Boyd Kane)18 Mar 2026 22:17 UTC

42 points

8 comments1 min readLW link

Metagaming matters for training, evaluation, and oversight

jenny and Bronson Schoen

18 Mar 2026 21:26 UTC

83 points

5 comments1 min readLW link

(alignment.openai.com)

Erosion of Deception and Secrecy

ColeG18 Mar 2026 20:47 UTC

4 points

0 comments4 min readLW link

Formal Verification: The Ultimate Fitness Function

Adam Chlipala18 Mar 2026 20:42 UTC

8 points

0 comments8 min readLW link

The Hundred Schools of Dietary Advice

axel_sdq18 Mar 2026 20:39 UTC

11 points

1 comment10 min readLW link

Training on Documents About Monitoring Leads To CoT Obfuscation

Reilly Haskins, bilalchughtai and Josh Engels

18 Mar 2026 20:37 UTC

65 points

5 comments16 min readLW link

LessWrong’s UX may not be living up to its ideas

neo18 Mar 2026 20:16 UTC

20 points

4 comments2 min readLW link