Softmax probabilities don't match real-world frequencies, breaking cross-company calibration

HanchungLee · x · 2026-09-17

The author argues that while softmax outputs a probability distribution, it does not guarantee that distribution matches real-world frequencies — and frequencies for next-token prediction differ from classification. As a result, a system-one model calibrated for one company won't transfer usefully to another.

Original post →

More from Research

Research channel →