Intrinsic Power-Seeking: AI Might Seek Power for Power’s Sake

TurnTrout19 Nov 2024 18:36 UTC

37 points

3 comments1 min readLW link

(turntrout.com)

Training AI agents to solve hard problems could lead to Scheming

Marius Hobbhahn and AlexMeinke

19 Nov 2024 0:10 UTC

58 points

12 comments28 min readLW link

Why imperfect adversarial robustness doesn’t doom AI control

Buck and Claude+

18 Nov 2024 16:05 UTC

60 points

23 comments2 min readLW link

Cross-context abduction: LLMs make inferences about procedural training data leveraging declarative facts in earlier training data

Sohaib Imran16 Nov 2024 23:22 UTC

34 points

5 comments14 min readLW link

Which evals resources would be good?

Marius Hobbhahn16 Nov 2024 14:24 UTC

46 points

4 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

Evolutionary prompt optimization for SAE feature visualization

neverix, Daniel Tan, Dmitrii Kharlapenko, Neel Nanda and Arthur Conmy

14 Nov 2024 13:06 UTC

16 points

0 comments9 min readLW link

AXRP Episode 38.0 - Zhijing Jin on LLMs, Causality, and Multi-Agent Systems

DanielFilan14 Nov 2024 7:00 UTC

14 points

0 comments12 min readLW link

o1 is a bad idea

abramdemski11 Nov 2024 21:20 UTC

152 points

36 comments2 min readLW link

The Evals Gap

Marius Hobbhahn11 Nov 2024 16:42 UTC

55 points

7 comments7 min readLW link

(www.apolloresearch.ai)

The Logistics of Distribution of Meaning: Against Epistemic Bureaucratization

Sahil7 Nov 2024 5:27 UTC

20 points

1 comment12 min readLW link

SAEs are highly dataset dependent: a case study on the refusal direction

Connor Kissane, robertzk, Neel Nanda and Arthur Conmy

7 Nov 2024 5:22 UTC

62 points

4 comments14 min readLW link

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

Zach Stein-Perlman6 Nov 2024 16:00 UTC

93 points

33 comments1 min readLW link

(alignment.anthropic.com)

SAE Probing: What is it good for? Absolutely something!

Subhash Kantamneni, JoshEngels, Senthooran Rajamanoharan and Neel Nanda

1 Nov 2024 19:23 UTC

31 points

0 comments11 min readLW link

Live Machinery: An Interface Design Philosophy for Wholesome AI Futures

Sahil1 Nov 2024 17:24 UTC

36 points

2 comments35 min readLW link

Seeking Collaborators

abramdemski1 Nov 2024 17:13 UTC

55 points

14 comments7 min readLW link

Complete Feedback

abramdemski1 Nov 2024 16:58 UTC

23 points

7 comments3 min readLW link

GPT-4o Guardrails Gone: Data Poisoning & Jailbreak-Tuning

ChengCheng, Brendan Murphy, AdamGleave and Kellin Pelrine

1 Nov 2024 0:10 UTC

17 points

0 comments6 min readLW link

(far.ai)

Toward Safety Cases For AI Scheming

Mikita Balesni and Marius Hobbhahn

31 Oct 2024 17:20 UTC

60 points

1 comment2 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

188 points

49 comments2 min readLW link

(www.thecompendium.ai)