29 Aug 2022 1:23 UTC

412 points

89 comments38 min readLW link 1 review

DeepMind alignment team opinions on AGI ruin arguments

Vika12 Aug 2022 21:06 UTC

376 points

37 comments14 min readLW link 1 review

A Mechanistic Interpretability Analysis of Grokking

Neel Nanda and Tom Lieberum

15 Aug 2022 2:41 UTC

370 points

47 comments36 min readLW link 1 review

(colab.research.google.com)

Two-year update on my personal AI timelines

Ajeya Cotra2 Aug 2022 23:07 UTC

288 points

60 comments16 min readLW link

Common misconceptions about OpenAI

Jacob_Hilton25 Aug 2022 14:02 UTC

242 points

142 comments5 min readLW link 1 review

What do ML researchers think about AI in 2022?

KatjaGrace4 Aug 2022 15:40 UTC

219 points

33 comments3 min readLW link

(aiimpacts.org)

Worlds Where Iterative Design Fails

johnswentworth30 Aug 2022 20:48 UTC

190 points

30 comments10 min readLW link 1 review

Language models seem to be much better than humans at next-token prediction

Buck, Fabien Roger and LawrenceC

11 Aug 2022 17:45 UTC

182 points

59 comments13 min readLW link 1 review

How To Go From Interpretability To Alignment: Just Retarget The Search

johnswentworth10 Aug 2022 16:08 UTC

179 points

33 comments3 min readLW link 1 review

Some conceptual alignment research projects

Richard_Ngo25 Aug 2022 22:51 UTC

174 points

15 comments3 min readLW link

Shard Theory: An Overview

David Udell11 Aug 2022 5:44 UTC

161 points

34 comments10 min readLW link

Nate Soares’ Life Advice

CatGoddess23 Aug 2022 2:46 UTC

154 points

41 comments3 min readLW link

Your posts should be on arXiv

JanB25 Aug 2022 10:35 UTC

149 points

44 comments3 min readLW link

What’s General-Purpose Search, And Why Might We Expect To See It In Trained ML Systems?

johnswentworth15 Aug 2022 22:48 UTC

140 points

18 comments10 min readLW link

The Parable of the Boy Who Cried 5% Chance of Wolf

KatWoods15 Aug 2022 14:33 UTC

139 points

24 comments2 min readLW link

How might we align transformative AI if it’s developed very soon?

HoldenKarnofsky29 Aug 2022 15:42 UTC

139 points

55 comments45 min readLW link 1 review

Externalized reasoning oversight: a research direction for language model alignment

tamera3 Aug 2022 12:03 UTC

130 points

23 comments6 min readLW link

Interpretability/Tool-ness/Alignment/Corrigibility are not Composable

johnswentworth8 Aug 2022 18:05 UTC

129 points

12 comments3 min readLW link

Fiber arts, mysterious dodecahedrons, and waiting on “Eureka!”

eukaryote4 Aug 2022 20:37 UTC

124 points

15 comments9 min readLW link 1 review

(eukaryotewritesblog.com)

Taking the parameters which seem to matter and rotating them until they don’t

Garrett Baker26 Aug 2022 18:26 UTC

120 points

48 comments1 min readLW link

Meditation course claims 65% enlightenment rate: my review

KatWoods1 Aug 2022 11:25 UTC

111 points

33 comments14 min readLW link

The lessons of Xanadu

jasoncrawford7 Aug 2022 17:59 UTC

109 points

20 comments8 min readLW link

(jasoncrawford.org)

Beliefs and Disagreements about Automating Alignment Research

Ian McKenzie24 Aug 2022 18:37 UTC

107 points

4 comments7 min readLW link

The alignment problem from a deep learning perspective

Richard_Ngo10 Aug 2022 22:46 UTC

107 points

15 comments27 min readLW link 1 review

Announcing Encultured AI: Building a Video Game

Andrew_Critch and Nick Hay

18 Aug 2022 2:16 UTC

103 points

26 comments4 min readLW link

How likely is deceptive alignment?

evhub30 Aug 2022 19:34 UTC

102 points

28 comments60 min readLW link

everything is okay

Tamsin Leake23 Aug 2022 9:20 UTC

98 points

22 comments7 min readLW link 2 reviews

(carado.moe)

Oversight Misses 100% of Thoughts The AI Does Not Think

johnswentworth12 Aug 2022 16:30 UTC

97 points

50 comments1 min readLW link

Introducing Pastcasting: A tool for forecasting practice

Sage Future11 Aug 2022 17:38 UTC

95 points

10 comments2 min readLW link 2 reviews

Survey advice

KatjaGrace24 Aug 2022 3:10 UTC

93 points

11 comments3 min readLW link

(worldspiritsockpuppet.com)

Rant on Problem Factorization for Alignment

johnswentworth5 Aug 2022 19:23 UTC

90 points

51 comments6 min readLW link

Less Threat-Dependent Bargaining Solutions?? (3/2)

Diffractor20 Aug 2022 2:19 UTC

88 points

7 comments6 min readLW link

How to do theoretical research, a personal perspective

Mark Xu19 Aug 2022 19:41 UTC

87 points

6 comments15 min readLW link

[Question] Seriously, what goes wrong with “reward the agent when it makes you smile”?

TurnTrout11 Aug 2022 22:22 UTC

86 points

42 comments2 min readLW link

High Reliability Orgs, and AI Companies

Raemon4 Aug 2022 5:45 UTC

86 points

7 comments12 min readLW link 1 review

Refining the Sharp Left Turn threat model, part 1: claims and mechanisms

Vika, Vikrant Varma, Ramana Kumar and Mary Phuong

12 Aug 2022 15:17 UTC

85 points

4 comments3 min readLW link 1 review

(vkrakovna.wordpress.com)

I’m mildly skeptical that blindness prevents schizophrenia

Steven Byrnes15 Aug 2022 23:36 UTC

83 points

9 comments4 min readLW link

Most Ivy-smart students aren’t at Ivy-tier schools

Aaron Bergman7 Aug 2022 3:18 UTC

82 points

7 comments8 min readLW link

(www.aaronbergman.net)

«Boundaries», Part 2: trends in EA’s handling of boundaries

Andrew_Critch6 Aug 2022 0:42 UTC

81 points

14 comments7 min readLW link

Paper is published! 100,000 lumens to treat seasonal affective disorder

Fabienne20 Aug 2022 19:48 UTC

81 points

3 comments1 min readLW link

(www.lesswrong.com)

The Loire Is Not Dry

jefftk20 Aug 2022 13:40 UTC

80 points

2 comments1 min readLW link

(www.jefftk.com)

What’s the Least Impressive Thing GPT-4 Won’t be Able to Do

Algon20 Aug 2022 19:48 UTC

80 points

125 comments1 min readLW link

Human Mimicry Mainly Works When We’re Already Close

johnswentworth17 Aug 2022 18:41 UTC

80 points

16 comments5 min readLW link

AI strategy nearcasting

HoldenKarnofsky25 Aug 2022 17:26 UTC

79 points

4 comments9 min readLW link

How (not) to choose a research project

Garrett Baker, CatGoddess and Johannes C. Mayer

9 Aug 2022 0:26 UTC

78 points

11 comments7 min readLW link

Evolution is a bad analogy for AGI: inner alignment

Quintin Pope13 Aug 2022 22:15 UTC

78 points

15 comments8 min readLW link

The Core of the Alignment Problem is...

Thomas Larsen, Jeremy Gillen and JamesH

17 Aug 2022 20:07 UTC

74 points

10 comments9 min readLW link

Discovering Agents

zac_kenton18 Aug 2022 17:33 UTC

73 points

11 comments6 min readLW link

Announcing the Introduction to ML Safety course

Dan H, ThomasW and ozhang

6 Aug 2022 2:46 UTC

73 points

6 comments7 min readLW link

[Question] COVID-19 Group Testing Post-mortem?

gwern5 Aug 2022 16:32 UTC

72 points

6 comments2 min readLW link