François Fleuret: trusting model predictions boils down to confidence intervals
francoisfleuret · x · 2026-10-12
François Fleuret argues that when it comes to trusting a model's prediction in a real-world situation, there's not much beyond basic statistics: essentially confidence intervals around empirical averages, built on concentration bounds. He reposted this after deleting an earlier version of the same point.
Related event: Fleuret: Formal Math Offers No Model Guarantees Beyond Basic Statistics(2 posts)→
More from Research
- OpenAI raises the plane's chromatic number lower bound to 6, Lean-formalized — fortnow · 2026-10-12
- Snorkel AI Expands Open Benchmark Grants 10x to $30M, Adds Red Team — ajratner · 2026-10-12
- How long is the coast of a neuron? Point-neuron abstraction may hide capacity — neurovium · 2026-10-12
- Community shrinks OpenAI Problem #190 obstruction from 66x66 to 3x4, verified in Lean — BorisMPower · 2026-10-12
- Autoresearch for LLM pretraining needs humans in the loop; stacked changes are the killer — menhguin · 2026-10-12
- COMODO distills video intelligence into IMU sensors for camera-free egocentric AI glasses — flosalim · 2026-10-12