Gradient Routing: Masking Gradients to Localize Computation in Neural Networks

cloud, Jacob G-W, Evzen, Joseph Miller and TurnTrout

6 Dec 2024 22:19 UTC

98 points

1 comment11 min readLW link

(arxiv.org)

Frontier Models are Capable of In-context Scheming

Marius Hobbhahn, AlexMeinke, Bronson Schoen, rusheb, Jérémy Scheurer and Mikita Balesni

5 Dec 2024 22:11 UTC

170 points

12 comments7 min readLW link

The 2023 LessWrong Review: The Basic Ask

Raemon4 Dec 2024 19:52 UTC

70 points

20 comments9 min readLW link

Deep Causal Transcoding: A Framework for Mechanistically Eliciting Latent Behaviors in Language Models

Andrew Mack and TurnTrout

3 Dec 2024 21:19 UTC

80 points

5 comments41 min readLW link

Analysis of Global AI Governance Strategies

Sammy Martin, Justin Bullock and Corin Katzke

4 Dec 2024 10:45 UTC

37 points

8 comments36 min readLW link

Cognitive Biases Contributing to AI X-risk — a deleted excerpt from my 2018 ARCHES draft

Andrew_Critch3 Dec 2024 9:29 UTC

46 points

2 comments5 min readLW link

Hierarchical Agency: A Missing Piece in AI Alignment

Jan_Kulveit27 Nov 2024 5:49 UTC

99 points

19 comments11 min readLW link

AXRP Episode 39 - Evan Hubinger on Model Organisms of Misalignment

DanielFilan1 Dec 2024 6:00 UTC

40 points

0 comments67 min readLW link

Conjecture: A Roadmap for Cognitive Software and A Humanist Future of AI

Connor Leahy and Gabriel Alfour

2 Dec 2024 13:28 UTC

35 points

7 comments29 min readLW link

(www.conjecture.dev)

o1 is a bad idea

abramdemski11 Nov 2024 21:20 UTC

161 points

37 comments2 min readLW link

LLM chatbots have ~half of the kinds of “consciousness” that humans believe in. Humans should avoid going crazy about that.

Andrew_Critch22 Nov 2024 3:26 UTC

77 points

53 comments5 min readLW link

The Compendium, A full argument about extinction risk from AGI

adamShimi, Gabriel Alfour, Connor Leahy, Chris Scammell and Andrea_Miotti

31 Oct 2024 12:01 UTC

191 points

49 comments2 min readLW link

(www.thecompendium.ai)

AXRP Episode 38.2 - Jesse Hoogland on Singular Learning Theory

DanielFilan27 Nov 2024 6:30 UTC

34 points

0 comments10 min readLW link

[Question] Are You More Real If You’re Really Forgetful?

Thane Ruthenis24 Nov 2024 19:30 UTC

39 points

24 comments5 min readLW link

Training AI agents to solve hard problems could lead to Scheming

Marius Hobbhahn and AlexMeinke

19 Nov 2024 0:10 UTC

61 points

12 comments28 min readLW link

Why imperfect adversarial robustness doesn’t doom AI control

Buck and Claude+

18 Nov 2024 16:05 UTC

61 points

27 comments2 min readLW link

Anthropic: Three Sketches of ASL-4 Safety Case Components

Zach Stein-Perlman6 Nov 2024 16:00 UTC

94 points

33 comments1 min readLW link

(alignment.anthropic.com)

the case for CoT unfaithfulness is overstated

nostalgebraist29 Sep 2024 22:07 UTC

244 points

40 comments11 min readLW link

A Sober Look at Steering Vectors for LLMs

Joschka Braun, Dmitrii Krasheninnikov, Usman Anwar, RobertKirk, Daniel Tan and David Scott Krueger (formerly: capybaralet)

23 Nov 2024 17:30 UTC

30 points

0 comments5 min readLW link

Win/continue/lose scenarios and execute/replace/audit protocols

Buck15 Nov 2024 15:47 UTC

54 points

2 comments7 min readLW link