1 Apr 2024 7:33 UTC

564 points

174 comments11 min readLW link

Transformers Represent Belief State Geometry in their Residual Stream

Adam Shai16 Apr 2024 21:16 UTC

411 points

100 comments12 min readLW link

Thoughts on seed oil

dynomight20 Apr 2024 12:29 UTC

347 points

129 comments17 min readLW link

(dynomight.net)

[April Fools’ Day] Introducing Open Asteroid Impact

Linch1 Apr 2024 8:14 UTC

334 points

29 comments1 min readLW link

(openasteroidimpact.org)

Express interest in an “FHI of the West”

habryka18 Apr 2024 3:32 UTC

268 points

41 comments3 min readLW link

Paul Christiano named as US AI Safety Institute Head of AI Safety

Joel Burget16 Apr 2024 16:22 UTC

256 points

58 comments1 min readLW link

(www.commerce.gov)

Refusal in LLMs is mediated by a single direction

Andy Arditi, Oscar Obeso, Aaquib111, wesg and Neel Nanda

27 Apr 2024 11:13 UTC

236 points

93 comments10 min readLW link

Funny Anecdote of Eliezer From His Sister

Noah Birnbaum22 Apr 2024 22:05 UTC

202 points

6 comments2 min readLW link

[Question] Examples of Highly Counterfactual Discoveries?

johnswentworth23 Apr 2024 22:19 UTC

194 points

101 comments1 min readLW link

OMMC Announces RIP

Adam Scholl and aysja

1 Apr 2024 23:20 UTC

189 points

5 comments2 min readLW link

On Not Pulling The Ladder Up Behind You

Screwtape26 Apr 2024 21:58 UTC

188 points

21 comments9 min readLW link

Why Would Belief-States Have A Fractal Structure, And Why Would That Matter For Interpretability? An Explainer

johnswentworth and David Lorell

18 Apr 2024 0:27 UTC

184 points

21 comments7 min readLW link

FHI (Future of Humanity Institute) has shut down (2005–2024)

gwern17 Apr 2024 13:54 UTC

176 points

22 comments1 min readLW link

(www.futureofhumanityinstitute.org)

Reconsider the anti-cavity bacteria if you are Asian

Lao Mein15 Apr 2024 7:02 UTC

168 points

43 comments4 min readLW link

Ironing Out the Squiggles

Zack_M_Davis29 Apr 2024 16:13 UTC

153 points

36 comments11 min readLW link

Daniel Dennett has died (1942-2024)

kave19 Apr 2024 16:17 UTC

150 points

5 comments1 min readLW link

(dailynous.com)

Priors and Prejudice

MathiasKB22 Apr 2024 15:00 UTC

150 points

31 comments7 min readLW link

LLMs for Alignment Research: a safety priority?

abramdemski4 Apr 2024 20:03 UTC

145 points

24 comments11 min readLW link

When is a mind me?

Rob Bensinger17 Apr 2024 5:56 UTC

138 points

125 comments15 min readLW link

My experience using financial commitments to overcome akrasia

William Howard15 Apr 2024 22:57 UTC

137 points

33 comments18 min readLW link

Simple probes can catch sleeper agents

Monte M, Carson Denison, Zac Hatfield-Dodds, David Duvenaud, Sam Bowman, Ethan Perez and evhub

23 Apr 2024 21:10 UTC

133 points

21 comments1 min readLW link

(www.anthropic.com)

A Dozen Ways to Get More Dakka

Davidmanheim8 Apr 2024 4:45 UTC

132 points

11 comments3 min readLW link

RTFB: On the New Proposed CAIP AI Bill

Zvi10 Apr 2024 18:30 UTC

119 points

14 comments34 min readLW link

(thezvi.wordpress.com)

A Selection of Randomly Selected SAE Features

CallumMcDougall and Joseph Bloom

1 Apr 2024 9:09 UTC

109 points

2 comments4 min readLW link

Discriminating Behaviorally Identical Classifiers: a model problem for applying interpretability to scalable oversight

Sam Marks18 Apr 2024 16:17 UTC

107 points

10 comments12 min readLW link

The first future and the best future

KatjaGrace25 Apr 2024 6:40 UTC

106 points

12 comments1 min readLW link

(worldspiritsockpuppet.com)

[Question] What convincing warning shot could help prevent extinction from AI?

Charbel-Raphaël and cozyfractal

13 Apr 2024 18:09 UTC

105 points

18 comments2 min readLW link

Carl Sagan, nuking the moon, and not nuking the moon

eukaryote13 Apr 2024 4:08 UTC

103 points

8 comments6 min readLW link

(eukaryotewritesblog.com)

MIRI’s April 2024 Newsletter

Harlan12 Apr 2024 23:38 UTC

95 points

0 comments3 min readLW link

(intelligence.org)

Sparsify: A mechanistic interpretability research agenda

Lee Sharkey3 Apr 2024 12:34 UTC

94 points

22 comments22 min readLW link

Towards Multimodal Interpretability: Learning Sparse Interpretable Features in Vision Transformers

hugofry29 Apr 2024 20:57 UTC

92 points

8 comments11 min readLW link

Partial value takeover without world takeover

KatjaGrace5 Apr 2024 6:20 UTC

89 points

23 comments3 min readLW link

(worldspiritsockpuppet.com)

Rejecting Television

Declan Molony23 Apr 2024 4:59 UTC

85 points

10 comments6 min readLW link

Constructability: Plainly-coded AGIs may be feasible in the near future

Épiphanie Gédéon and Charbel-Raphaël

27 Apr 2024 16:04 UTC

82 points

13 comments13 min readLW link

Essay competition on the Automation of Wisdom and Philosophy — $25k in prizes

owencb and AI Impacts

16 Apr 2024 10:10 UTC

82 points

12 comments8 min readLW link

(blog.aiimpacts.org)

A couple productivity tips for overthinkers

Steven Byrnes20 Apr 2024 16:05 UTC

78 points

13 comments4 min readLW link

Creating unrestricted AI Agents with Command R+

Simon Lermen16 Apr 2024 14:52 UTC

77 points

13 comments5 min readLW link

[Full Post] Progress Update #1 from the GDM Mech Interp Team

Neel Nanda, Arthur Conmy, lewis smith, Senthooran Rajamanoharan, Tom Lieberum, János Kramár and Vikrant Varma

19 Apr 2024 19:06 UTC

77 points

10 comments8 min readLW link

Mid-conditional love

KatjaGrace17 Apr 2024 4:00 UTC

76 points

21 comments2 min readLW link

(worldspiritsockpuppet.com)

Coherence of Caches and Agents

johnswentworth1 Apr 2024 23:04 UTC

76 points

9 comments11 min readLW link

AISC9 has ended and there will be an AISC10

Linda Linsefors29 Apr 2024 10:53 UTC

75 points

4 comments2 min readLW link

A Gentle Introduction to Risk Frameworks Beyond Forecasting

pendingsurvival11 Apr 2024 18:03 UTC

73 points

10 comments27 min readLW link

[Summary] Progress Update #1 from the GDM Mech Interp Team

Neel Nanda, Arthur Conmy, lewis smith, Senthooran Rajamanoharan, Tom Lieberum, János Kramár and Vikrant Varma

19 Apr 2024 19:06 UTC

72 points

0 comments3 min readLW link

Announcing Suffering For Good

Garrett Baker1 Apr 2024 17:08 UTC

72 points

5 comments1 min readLW link

Prompts for Big-Picture Planning

Raemon13 Apr 2024 3:04 UTC

72 points

1 comment3 min readLW link

LW Frontpage Experiments! (aka “Take the wheel, Shoggoth!”)

Ruby and RobertM

23 Apr 2024 3:58 UTC

71 points

27 comments5 min readLW link

A gentle introduction to mechanistic anomaly detection

Erik Jenner3 Apr 2024 23:06 UTC

71 points

2 comments11 min readLW link

Text Posts from the Kids Group: 2020

jefftk13 Apr 2024 22:30 UTC

69 points

3 comments19 min readLW link

(www.jefftk.com)

Motivation gaps: Why so much EA criticism is hostile and lazy

titotal22 Apr 2024 11:49 UTC

69 points

5 comments1 min readLW link

(titotal.substack.com)

AXRP Episode 27 - AI Control with Buck Shlegeris and Ryan Greenblatt

DanielFilan11 Apr 2024 21:30 UTC

69 points

10 comments107 min readLW link