richar-wang/FratBench

Measuring social calibration failure in frontier language models. 53-question benchmark where humans score 87% and the best LLM scores 32%.

2 best points · 0 comments · 1 Hacker News post · 0 GitHub stars · View on GitHub

Hacker News discussions

• FratBench – Social Calibration Benchmark (OAI Scores Dead Last) [pdf] (2 points, 0 comments, by richardwang5, 2026-03-17)