Goodfire uses Silico to probe how AI models interpret Parkinson's gait features
mathildepapillo · x · 2026-08-05
The Goodfire AI team used the Silico tool to probe how general-purpose human motion AI models detect Parkinson's gait impairment. They broke the models open to identify which specific gait features drive predictions.
This application is seen as a major step forward for mechanistic interpretability in clinical motion analysis. A thread and report have been released.
More from Embodied
- Stanford Paper Demystifies Why Vision-Language-Action Models Fail in Contact-Rich Tasks — StanfordAILab · 2026-08-05
- Robot prices set to plummet, pre-orders surge — chris_j_paxton · 2026-08-05
- Lightbot 0 Robot Demo: Climbing Obstacles as Temporary Support Points — chris_j_paxton · 2026-08-05
- Build a Desktop Robot for $8: Open-Source Xiaozhi Framework Brings Local AI to ESP32 — TinfoilTricorn · 2026-08-05
- Unitree IPO Filings Reveal ~60% Gross Margins, Flipping Hardware Commodity Myth — Rewkang · 2026-08-05
- Assemble Bench: A New Benchmark for Robot Models Based on NIST Standards — ycombinator · 2026-08-05