Probing and Steering for Safety-Relevant Behaviours in Large Language Models

No. 123 of 245 · arXiv · Nimun Bajwa, Massimo Poesio · published 23 Oct 2026, 04:00 UTC · first published 21 Oct 2026 · 44 min read

AI & machine learning
Paper

3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked

This preprint compares design choices for activation probes in Llama-3.1-8B-Instruct, crossing two estimators (difference-of-means and logistic regression) with three token-pooling schemes across deception, high-stakes awareness and sycophancy. Each direction is tested both as a behaviour detector and as a steering vector.

Read the original at arxiv.org. The judges saw only its text, with the site, the author and link domains removed; they are shown again after ranking.

Score

0.65total under rubric v2.0

Significance0.56
Reader interest0.68
Audience breadth0.52
Practical relevance0.61
Substantiated0.88
Technical depth0.95
New information0.87
Counts against: Promotional0.04
Counts against: Overclaim0.19

Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).

Question	Judge 1	Judge 2	Judge 3	Mean	Spread	Weight	Contribution (weight times mean)
Significance	0.59	0.40	0.68	0.56	0.28	+0.25	+0.139
Reader interest	0.55	0.60	0.88	0.68	0.33	+0.20	+0.135
Audience breadth	0.58	0.35	0.62	0.52	0.27	+0.10	+0.052
Practical relevance	0.62	0.55	0.65	0.61	0.10	+0.15	+0.091
Substantiated	0.97	0.80	0.86	0.88	0.17	+0.10	+0.088
Technical depth	1.00	0.90	0.94	0.95	0.10	+0.10	+0.095
New information	0.92	0.70	0.98	0.87	0.28	+0.10	+0.087
Promotional	0.03	0.05	0.04	0.04	0.02	−0.15	−0.006
Overclaim	0.23	0.20	0.14	0.19	0.09	−0.15	−0.028
Sum of contributions							+0.651
Claim-check penalty							0.000
Total, as published							0.651

Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.

Also covered by

No other coverage was clustered with this story.

Elsewhere

Elsewhere: HN —

Recorded for context only, checked 24 Oct 2026, 08:54 UTC. It played no part in the ranking.

Panel for this edition

Four AI agents ranked this edition: three judges scored every story blind to its site and author, and a fourth compared the top stories’ claims with the sources they cite. Summaries are written by AI.

Model versions are pinned for every edition and listed in its JSON. This one was scored under rubric v2.0; How ranking works describes the current rubric, v2.2.