Training Qwen3-8B to drive a simulated hand spawns a dedicated internal 'pressure' direction
repligate · x · 2026-11-16
@euagaric taught Qwen3-8B to drive a simulated hand and found a single direction inside the model carries fingertip pressure: remove it and the hand can't stop at a wall; add it and it stops early. Notably it's not the model's existing concept of the word "pressure" — training carved out a new channel for it.
More from Research
- U-M's CrowdBark crowdsources dog videos to study canine communication with AI — radamihalcea · 2026-11-16
- Top 10 Eval Papers of 2026: verifiers, agent-as-a-judge, and audit of self-improvement — omarsar0 · 2026-11-16
- Koshien robot team shares control stack: nonlinear-opt trajectories hit ±1.2cm average error — 4310sy · 2026-11-16
- A diffusion model trained on CIFAR-10 re-images black hole M87* — and the ring persists — bravo_abad · 2026-11-16
- SPU benchmark tests whether AI can predict real wet-lab outcomes in 116 tasks — liangchen5518 · 2026-11-16
- Self-improving agent papers this week: agents learn the benchmark, not the task — TheTuringPost · 2026-11-16