Steering along manifolds: geometry-aware intervention to control neural networks
burny_tech · x · 2026-10-25
A shared paper, "Steering Along Manifolds to Control Neural Networks," intervenes along paths through activation space defined by different geometries and measures the behavioral trajectories they induce — illustrated as smoother, more controlled behavior shifts than standard linear steering.
More from Research
- PRISM merge algorithm combines two Minimax-H3 finetunes, weights open-sourced — theOliviaRossi · 2026-10-25
- 8 Papers from Stengel-Eskin's Group at EMNLP 2026 Span Pragmatics, Faithful CoT, Agents — EliasEskin · 2026-10-25
- Q-CTRL demos AI-powered quantum sensors for GPS-denied navigation at MIT EmTech — MJBiercuk · 2026-10-25
- Holon-S1 predicts genes and biomarkers from routine pathology slides across 16 organ types — iScienceLuvr · 2026-10-25
- A sun-like star that's normal in visible light but ~60x brighter in deep infrared — fofrAI · 2026-10-25
- The upside of AI-assisted research: failing faster at a higher level — mboehme_ · 2026-10-25