My most common advice for junior researchers

LawrenceC2 Apr 2026 2:41 UTC

24 points

0 comments3 min readLW link

Predicting When RL Training Breaks Chain-of-Thought Monitorability

David Lindner1 Apr 2026 10:23 UTC

21 points

0 comments5 min readLW link

(Some) Natural Emergent Misalignment from Reward Hacking in Non-Production RL

7vik, Sid Black and Joseph Bloom

30 Mar 2026 10:56 UTC

106 points

3 comments17 min readLW link

ControlAI 2025 Impact Report: our progress toward an international ban on ASI

Andrea_Miotti and Alex Amadori

27 Mar 2026 18:10 UTC

76 points

3 comments4 min readLW link

(controlai.com)

Test your best methods on our hard CoT interp tasks

daria, Riya Tyagi, Josh Engels and Neel Nanda

26 Mar 2026 19:24 UTC

55 points

2 comments19 min readLW link

A Toy Environment For Exploring Reasoning About Reward

jenny and Bronson Schoen

25 Mar 2026 20:29 UTC

55 points

7 comments3 min readLW link

Metagaming matters for training, evaluation, and oversight

jenny and Bronson Schoen

18 Mar 2026 21:26 UTC

63 points

5 comments1 min readLW link

(alignment.openai.com)

“Act-based approval-directed agents”, for IDA skeptics

Steven Byrnes18 Mar 2026 18:47 UTC

56 points

8 comments5 min readLW link

New RFP on Interpretability from Schmidt Sciences

Peter Hase17 Mar 2026 16:08 UTC

15 points

0 comments6 min readLW link

(schmidtsciences.smapply.io)

Power Steering: Behavior Steering via Layer-to-Layer Jacobian Singular Vectors

Omar Ayyub13 Mar 2026 3:55 UTC

20 points

0 comments17 min readLW link

Operationalizing FDT

Vivek Hebbar13 Mar 2026 0:12 UTC

90 points

11 comments6 min readLW link

How well do models follow their constitutions?

aryaj, Senthooran Rajamanoharan and Neel Nanda

12 Mar 2026 0:07 UTC

97 points

5 comments26 min readLW link

The Refined Counterfactual Prisoner’s Dilemma: An Attempt to Explode Decision-Theoretic Consequentialism

Chris_Leong11 Mar 2026 12:32 UTC

18 points

20 comments2 min readLW link

AIs will be used in “unhinged” configurations

Arthur Conmy11 Mar 2026 11:19 UTC

58 points

3 comments4 min readLW link

The case for satiating cheaply-satisfied AI preferences

Alex Mallen10 Mar 2026 18:09 UTC

102 points

7 comments23 min readLW link

Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation

Bartosz Cywiński, Helena Casademunt, Khoi Tran, aryaj, Sam Marks and Neel Nanda

9 Mar 2026 18:50 UTC

30 points

2 comments5 min readLW link

Payorian cooperation is easy with Kripke frames

transhumanist_atom_understander9 Mar 2026 0:29 UTC

70 points

7 comments8 min readLW link

Can governments quickly and cheaply slow AI training?

joshc7 Mar 2026 19:11 UTC

58 points

7 comments14 min readLW link

How to Design Environments for Understanding Model Motives

gersonkroiz, aditya singh, Senthooran Rajamanoharan and Neel Nanda

2 Mar 2026 7:14 UTC

43 points

0 comments10 min readLW link

Schelling Goodness, and Shared Morality as a Goal

Andrew_Critch28 Feb 2026 4:25 UTC

121 points

60 comments41 min readLW link