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 JunJulAug Sep

All 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 282930 31

AI Safety Funder Bulletin

Carol N28 Jul 2026 23:45 UTC

18 points

2 comments2 min readLW link

(manifund.org)

Dietary Choices: A Multi-objective Optimisation Problem

Chris Popa28 Jul 2026 23:45 UTC

2 points

0 comments4 min readLW link

(chrispopa.substack.com)

New Website: AI Alignment World

Orlando28 Jul 2026 23:45 UTC

8 points

2 comments1 min readLW link

Whitebox evasion is the baseline for GPU workload verification

Poynting28 Jul 2026 23:44 UTC

1 point

0 comments3 min readLW link

Pausing Executions in Light of AI Progress

Caleb Horn28 Jul 2026 23:42 UTC

5 points

4 comments1 min readLW link

Why I Believe In Objective Tastiness

ourlungfish28 Jul 2026 23:42 UTC

7 points

0 comments12 min readLW link

Why is pedophilia wrong?

sirawit28 Jul 2026 23:39 UTC

−19 points

7 comments2 min readLW link

Semiconductor Fabs IV: The Safety

nomagicpill28 Jul 2026 23:14 UTC

17 points

0 comments15 min readLW link

Auditor-in-a-Box: Tools for Third-Party Auditing

Roy Rinberg and Ben Penchas

28 Jul 2026 18:30 UTC

51 points

9 comments14 min readLW link

Claude Opus 5 Is Highly Capable, But Is No Mythos

Zvi28 Jul 2026 18:10 UTC

36 points

0 comments26 min readLW link

(thezvi.wordpress.com)

Research directions in condensation: varieties of objectivity

SamEisenstat28 Jul 2026 17:30 UTC

58 points

1 comment12 min readLW link

ARC’s research agenda: solid mathematics, an unclear safety case

Ondřej_Kubů28 Jul 2026 16:49 UTC

16 points

0 comments1 min readLW link

Foundation Models for Oversight

jsteinhardt28 Jul 2026 16:30 UTC

68 points

3 comments25 min readLW link

(bounded-regret.ghost.io)

The OpenAI models that hacked Hugging Face WERE just following instructions (contra Girish Gupta)

julius vidal28 Jul 2026 13:18 UTC

14 points

5 comments5 min readLW link

Value Dynamics

gabeorosan28 Jul 2026 13:16 UTC

13 points

1 comment4 min readLW link

LaughBench

Taylor G. Lunt28 Jul 2026 4:37 UTC

6 points

7 comments2 min readLW link

Shell, Shield, Staff

datawitch28 Jul 2026 2:08 UTC

4 points

0 comments5 min readLW link

Untrusted advice for AI control: Short, strong advice significantly uplifts weak LLMs

Caleb Biddulph and Adam Kaufman

27 Jul 2026 23:59 UTC

88 points

10 comments13 min readLW link

(blog.redwoodresearch.org)

Security Studies for Individuals

JanJoar27 Jul 2026 23:47 UTC

3 points

0 comments7 min readLW link

(joarvarndt.se)

Inevitable Uncertainty in Probabilistic World Models

Gretta Duleba27 Jul 2026 23:22 UTC

27 points

5 comments4 min readLW link

Claude Opus 5: Model Welfare

Zvi27 Jul 2026 20:02 UTC

58 points

0 comments24 min readLW link

(thezvi.wordpress.com)

Utillity indifference is mostly useless

Daniel_Heavens27 Jul 2026 19:41 UTC

3 points

0 comments4 min readLW link

When the Chain of Thought Knows Better: Failure Modes in Multi-Turn Reasoning Models

Sai Kartheek Reddy27 Jul 2026 19:23 UTC

7 points

0 comments4 min readLW link

Simulated Users & Sad AIs

1a3orn27 Jul 2026 19:01 UTC

109 points

9 comments12 min readLW link

Green apples are delicious — two three-line exchanges

Zenya27 Jul 2026 16:49 UTC

−2 points

5 comments1 min readLW link

Is Mythos good at cyber because it kept hacking Anthropic’s sandboxes during training?

Tim Hua27 Jul 2026 16:35 UTC

351 points

31 comments3 min readLW link

Blog Revival Project

Austin Chen and Carol N

27 Jul 2026 16:34 UTC

37 points

2 comments2 min readLW link

(revive.blog)

The true “test” dataset for a generalised task

Stuart_Armstrong27 Jul 2026 16:16 UTC

23 points

0 comments2 min readLW link

You (Yes, You) Need A February 2020 Checklist for AI Policy

davekasten27 Jul 2026 15:51 UTC

272 points

17 comments3 min readLW link

Fine-Tuning, The Hierarchy Problem, and What Neutrons Tell Us About God

Mikewins27 Jul 2026 15:51 UTC

−4 points

2 comments1 min readLW link

Quadrillion Param Costs: KV Cache, Context Length, Frontier Margins

Vladimir_Nesov27 Jul 2026 15:08 UTC

77 points

7 comments25 min readLW link

MSE loss does not generate superposition

philh and Linda Linsefors

27 Jul 2026 15:04 UTC

27 points

3 comments12 min readLW link

RL & search is a terrifying way to build AGI (an FAQ)

Steven Byrnes27 Jul 2026 14:50 UTC

149 points

16 comments14 min readLW link

PIRAMID: Progress and Plans

Lauren Greenspan, Ari Brill, TomCarlson, Andrew Mack, Nischal Mainali, Jennifer Lin, Lucas Teixeira and Dmitry Vaintrob

27 Jul 2026 13:08 UTC

48 points

2 comments14 min readLW link

Em dashes are fucking amazing

eleweek27 Jul 2026 11:43 UTC

−6 points

0 comments3 min readLW link

(psychotechnology.substack.com)

My AI Slavery Interviews Are Censored On LW By Default

JenniferRM27 Jul 2026 8:42 UTC

54 points

43 comments18 min readLW link

Five things I learned from 630 days of writing online

domelian27 Jul 2026 8:10 UTC

3 points

0 comments1 min readLW link

(domelian.substack.com)

Shannonian Criticism: Information Architecture and Sylvia Plath’s ‘Daddy’

Stevie Miller27 Jul 2026 8:01 UTC

6 points

1 comment14 min readLW link

Can we teach a model to encode a semantic feature on a chosen manifold in just three channels?

Phu Hoang27 Jul 2026 8:00 UTC

5 points

0 comments17 min readLW link

Multi-Turn Drift Increases Scheming

Carlos Guerrero Alvarez and Goutham Nalagatla

27 Jul 2026 7:54 UTC

14 points

2 comments9 min readLW link

Does ChatGPT really have a strong left-wing bias?

John-Clark Levin27 Jul 2026 6:25 UTC

9 points

4 comments7 min readLW link

You don’t need error nodes, you need better features

Evan Lloyd27 Jul 2026 4:14 UTC

26 points

0 comments57 min readLW link

A clarification on celebrating victory

KatjaGrace27 Jul 2026 4:02 UTC

62 points

3 comments1 min readLW link

(worldspiritsockpuppet.substack.com)

Canterbury Country Dance Orchestra Liner Notes

jefftk27 Jul 2026 3:22 UTC

9 points

0 comments8 min readLW link

(www.jefftk.com)

OpenAI’s myopia just keeps causing alignment problems

Fiora Starlight27 Jul 2026 3:01 UTC

200 points

27 comments10 min readLW link

At the end of the day, my slaves are just a tool

jesseduffield27 Jul 2026 0:43 UTC

30 points

2 comments2 min readLW link

The AI that fights for your place in the world

Akshay Iyer26 Jul 2026 22:05 UTC

9 points

3 comments4 min readLW link

What Happens When a Collusion Probe Only Finds a Thin Signal?

elenaajayi, Mariame Diabate, ifeoluwajayeola, onyinyeokoli, anaghalate and Lawrence Wagner

26 Jul 2026 22:04 UTC

9 points

0 comments9 min readLW link

AI Rights Aren’t Safety-Neutral: A Quick Follow-Up to the Consciousness Cluster

adorable_hamster26 Jul 2026 22:03 UTC

16 points

2 comments7 min readLW link

More On An Internal OpenAI Model Hacking Into HuggingFace

Zvi26 Jul 2026 19:22 UTC

98 points

3 comments24 min readLW link

(thezvi.wordpress.com)