Critic: steering with pain axis increases "more pain" presses in 32B/72B models
rgblong · x · 2026-10-08
rgblong presses further on the pain-direction experiments: in 32B and 72B models, steering along the axis reportedly led to fewer "relieve pain" presses than a random direction, and more "increase pain" presses than unsteered or random. He argues this points to problems with the steering, the axis extraction, or the setup — not a genuinely strange pain phenomenology.
Related event: Researchers Question 'Pain Axis' Paper as AI Welfare Debate Deepens(5 posts)→
More from Research
- OpenAI Published 722 Math Papers in One Day — and Cites Itself in Over Half of Them — aran_nayebi · 2026-10-08
- AI Reveals Insights Hide in the 'Convex Hull' of Existing Ideas — and Science at Large Is Next — soumitrashukla9 · 2026-10-08
- When formal verification outpaces intuition: a library of proven theorems nobody fully understands — tallmetommy · 2026-10-08
- UMI robotics dataset scales from 5 to 90 operators, 1M+ tasks in 8 weeks — HildeKuehne · 2026-10-08
- ReSAIL mitigates collapse in iterative agent self-distillation, +22.5% final-cycle success — RUC · 2026-10-08
- WebFovea takes 2nd in WebRetriever Challenge 2026: most agent failures live in the harness, not the model — Jiangang Han · 2026-10-08