Higher Scores Don't Mean Higher Stability
sanmikoyejo · x · 2026-07-16
This post adds that context-induced instability fluctuates across different stages of model development. In fact, it can become more pronounced even after benchmark accuracy improves, particularly during the post-training phase.
The author's core conclusion: higher overall performance does not necessarily imply greater robustness against irrelevant context.
More from Research
- Nature paper images cellular activity across all organs, revealing body-wide circuits — arjunrajlab · 2026-09-11
- SignNet 1M Dataset Released for Sign Language Research — ducha_aiki · 2026-09-11
- ECCV26 Oral: Flow Matching Enables Single-Stage Multi-View Point Cloud Registration — ducha_aiki · 2026-09-11
- InFlux++ Method Released — ducha_aiki · 2026-09-11
- Skyfall GS Uses Flux to Refine Gaussian Splatting, Accepted at ECCV 2026 — ducha_aiki · 2026-09-11
- Could 10k agents discover learning methods beyond backprop, or just tweak existing ones? — SeunghyunSEO7 · 2026-09-11