Researchers question whether the 'pain direction' steering axis in Qwen 32B/72B is real
rgblong · x · 2026-10-08
In a discussion about steering experiments on 32B/72B models, rgblong notes the 'pain direction' axis produces odd results: less 'relieve pain' pressing than random directions, yet more 'increase pain' than unsteered. She argues this points to problems with the steering method, the axis definition, or the setup — not genuine 'strange pain' — and the many anomalous behaviors deepen her doubts about how the axis was defined.
More from Research
- HKU and VAST's Mira-Scene fixes object placement in single-image 3D scene reconstruction — jiqizhixin · 2026-10-08
- Math community debates whether OpenAI's math outputs are 'raw data' or slop — littmath · 2026-10-08
- Evolvent AI releases RSIGym and RSI-Index: benchmarking AI self-improvement, Opus 5 leads at 0.4809 — cihangxie · 2026-10-08
- Google RCT: AI boosts patent lawyers' output but juniors gain no judgment — Dr_Atoosa · 2026-10-08
- Full Slides Released for ECCV'26 Tutorial on Diffusion Model Post-Training and Alignment — CSProfKGD · 2026-10-08
- Norvig's Classic Essay on Chomsky and the Two Cultures of Statistical Learning Still Reads Fresh in the LLM Era — 3scorciav · 2026-10-08