57 points

9 comments2 min readLW link

Some Ways I Think About Evaluating Grant Applications

sarahconstantin13 Aug 2026 21:30 UTC

61 points

0 comments10 min readLW link

(sarahconstantin.substack.com)

The Left Should Start Taking AI Capabilities Seriously

Alexei G13 Aug 2026 21:03 UTC

−13 points

0 comments9 min readLW link

(www.onethousandmeans.com)

Is Alignment Even Falsifiable? Middle Alignment, An Alignment Taxonomy, and Breaking The Problem Down Into Steps

Savannah Harlan13 Aug 2026 20:57 UTC

5 points

0 comments22 min readLW link

Concrete Generalist Projects in AI Safety (and how to do them)

Harry Waterman, Jesse Gilbert and Roman Ross

13 Aug 2026 19:26 UTC

25 points

2 comments6 min readLW link

(forum.effectivealtruism.org)

Comparing Congress’s Two AI Emergency Shutdown Mechanisms

Philip Dowdell13 Aug 2026 18:53 UTC

13 points

0 comments8 min readLW link

Is biosecurity oversaturated?

Master Chief13 Aug 2026 18:13 UTC

12 points

1 comment1 min readLW link

How to Answer a Question Without Answering The Question

Kabir Kumar13 Aug 2026 17:06 UTC

37 points

3 comments1 min readLW link

How My Students Think About AI

dvd13 Aug 2026 16:56 UTC

774 points

110 comments12 min readLW link

AI #181: Astra Goes Cyber Critical

Zvi13 Aug 2026 15:20 UTC

39 points

1 comment52 min readLW link

(thezvi.wordpress.com)

Automated alignment runs are hard to study!

Alejandro Aristizabal, draganover, Aleksandr Bowkis and Cameron Holmes

13 Aug 2026 15:04 UTC

68 points

4 comments9 min readLW link

Longtermism Seems Like A Religion

James Brobin13 Aug 2026 14:43 UTC

2 points

8 comments3 min readLW link

Defense Against the Deceptive Arts

Kabir Kumar13 Aug 2026 14:21 UTC

11 points

11 comments1 min readLW link

Some problems in decision theory incorrectly precondition on policy.

Canaletto13 Aug 2026 12:04 UTC

10 points

4 comments3 min readLW link

Patterns and problems in emerging multiagent systems (Anthropic, Frontier Red Team)

Julian Bradshaw13 Aug 2026 4:10 UTC

42 points

3 comments1 min readLW link

(www.anthropic.com)

The Descenders and the Absorbers: Two Perspectives on Deep Learning

larry-dial13 Aug 2026 4:02 UTC

11 points

0 comments3 min readLW link

OC ACXLW Meetup #118 — The 40% Problem & The Field That Refused to Die

Michael Michalchik13 Aug 2026 3:25 UTC

1 point

0 comments9 min readLW link

Measuring Activation Control in LLMs

Marek Kowalski, Joshua Fonseca Rivera, Uzay Macar and David Africa

13 Aug 2026 2:04 UTC

50 points

8 comments7 min readLW link

How Valuable are BOTECs?

Roman Ross13 Aug 2026 0:31 UTC

9 points

0 comments8 min readLW link

Measuring Eval Awareness: The Realism Win Rate is Fragile

Achu Menon and Santiago Aranguri

13 Aug 2026 0:26 UTC

14 points

3 comments4 min readLW link

What happened when I tried to be vegan

finitude13 Aug 2026 0:21 UTC

44 points

2 comments5 min readLW link

LLMs have the capacity for self-imposed steganography

Edward Cant13 Aug 2026 0:16 UTC

7 points

0 comments7 min readLW link

Free will is like temperature

Optimization Process12 Aug 2026 23:32 UTC

47 points

10 comments1 min readLW link

Impact markets made concrete

Carol N12 Aug 2026 22:45 UTC

6 points

0 comments6 min readLW link

(manifund.substack.com)

Unblocking AI’s Continual Learning: Hints From How Humans Learn

nimakeivan12 Aug 2026 18:28 UTC

3 points

23 comments10 min readLW link

Monthly Roundup #45: August 2026

Zvi12 Aug 2026 17:50 UTC

24 points

1 comment20 min readLW link

(thezvi.wordpress.com)

Finding the Seams of Perception

jimmy12 Aug 2026 17:33 UTC

30 points

0 comments2 min readLW link

(beneathpsychology.com)

Introducing the Conceptual Reasoning Index

Chi Nguyen, Emery Cooper, Caspar Oesterheld, Alex Kastner and Joe Benton

12 Aug 2026 17:08 UTC

83 points

16 comments5 min readLW link

(alignment.anthropic.com)

One attention head carries knight forks in a chess transformer, and here’s a new toolkit that found it.

David Litman12 Aug 2026 16:48 UTC

8 points

0 comments1 min readLW link

(github.com)

The Psychology of Cope: Rationality Is Not Reversed Irrationality

Chris_Leong12 Aug 2026 12:20 UTC

27 points

12 comments3 min readLW link

Demon Safety

Ben Pace12 Aug 2026 11:55 UTC

53 points

5 comments1 min readLW link

The Closure of the Internet (Research Linkpost)

Dean Valentine12 Aug 2026 8:22 UTC

34 points

4 comments1 min readLW link

(arctotherium.substack.com)

AI swarms are starting to pose indirect takeover risk

oakhu and Alex Mallen

12 Aug 2026 5:05 UTC

138 points

6 comments10 min readLW link

Did the alignment community underestimate its power?

StanislavKrym12 Aug 2026 2:56 UTC

23 points

0 comments10 min readLW link

The Age of Pluribus: One Consultant for Everyone

Dorothy Gale12 Aug 2026 1:39 UTC

9 points

0 comments4 min readLW link

We should consider how long monitoring is reliable for during RL

lachlan on a boat12 Aug 2026 1:22 UTC

21 points

1 comment4 min readLW link

When (and when not) LLMs can verbalize awareness of J-Space concept injections—Initial results

Ethan Garcia12 Aug 2026 1:21 UTC

8 points

0 comments15 min readLW link

(e-m-garcia.github.io)

Arguments for and against (me) dropping out

hersheys12 Aug 2026 1:12 UTC

30 points

3 comments8 min readLW link

Patient Zero

LoopGameScrollMonkey12 Aug 2026 1:10 UTC

17 points

0 comments12 min readLW link

An anytime algorithm for mixing the computable measures

Cole Wyeth12 Aug 2026 0:57 UTC

26 points

0 comments4 min readLW link

Various Reflections About What Happened With OpenAI’s Internal Models

Zvi11 Aug 2026 22:20 UTC

63 points

1 comment30 min readLW link

(thezvi.wordpress.com)

Claude Opus 5 Just Beat My Text-Based Adventure Game Benchmark

derelict543211 Aug 2026 20:19 UTC

22 points

4 comments5 min readLW link

[Webinar] Why AI Safety is a Capital Allocation Problem

Schizoid Rentoid11 Aug 2026 19:27 UTC

2 points

0 comments1 min readLW link

Misaligned AIs could use killer robots to take over

Omar Khursheed and TurnTrout

11 Aug 2026 19:03 UTC

134 points

10 comments6 min readLW link

(turntrout.com)

Measuring Spurious Correlations with Feature Strength

egan11 Aug 2026 18:05 UTC

36 points

1 comment10 min readLW link

AI governance work needs much better monitoring

jackultraphil11 Aug 2026 17:42 UTC

10 points

0 comments8 min readLW link

(fundinganthropalypse.com)

LLMs Are Starting To Noticeably Accelerate Our Work

johnswentworth11 Aug 2026 17:06 UTC

251 points

19 comments2 min readLW link

Software Is Hard

cylonator11 Aug 2026 16:26 UTC

5 points

0 comments1 min readLW link

How risky would it be to make powerful AI obey one or a few people?

cousin_it and Seth Herd

11 Aug 2026 16:15 UTC

93 points

24 comments8 min readLW link

Extreme concentration of power over ASI has non-obvious advantages

Seth Herd11 Aug 2026 16:13 UTC

43 points

4 comments19 min readLW link