cross-entropy
How AI Scores Itself: Cross-Entropy Powers Every LLM
Cross-entropy lets LLM training distinguish 40%-confident-wrong from 1%-confident-wrong. Loss functions matter more than accuracy for enterprise IT.
Back to home
model evaluation
3 articles tagged with this topic
LocalLLaMA
Reddit's Soap Dispenser AI Code Battle — Chinese Models Hold Pace
LocalLLaMA ran a soap dispenser animation test for 5 major models. Olympiads out, daily objects in: LLM evaluation shifts from elite to usable.
Political Compass
LLMs Take the Political Compass: Reddit User Tests 10+ Major Models
Thrumpwart ran 10+ major LLMs through Political Compass. Most landed economically left, socially liberal—training-data bias behind the lark.