Seonglae ChoOptimismBench: Forecasting Bias and the Alignment Effect in...
Large language models are increasingly used as decision aids whose probability judgments shape downstream choices. Whether those judgments carry a systematic directional tilt has been hard to...
https://arxiv.org/abs/2607.26981
AI Probability Bias Traced to Training: Claude Pessimistic, GPT Optimistic Across All Tests
AI probability bias is universal across 16 frontier models, OptimismBench finds, but runs in opposite directions: most AI models including GPT overestimate success probabilities, while Anthropic’s Claude Sonnet 4.6 and Opus 4.6 underestimate them — a split caused by post-training alignment
https://www.techtimes.com/articles/322341/20260730/ai-probability-bias-traced-training-claude-pessimistic-gpt-optimistic-across-all-tests.htm
