Show HN: Human Benchmark – Compare your reasoning skills against AI models

Category: ai-ml

Tags: benchmarking, llm, psychometrics, reasoning

Score: 4.0/10 (Innovation: 5, Technical: 4, Documentation: 3, Utility: 3)

A web app that lets users compare their reasoning skills against AI models using questions from established benchmarks like CommonsenseQA, GSM8K, and AGIEval. It employs a Rasch model to estimate a percentage score from a small number of items, with confidence intervals shown. The project is a proof of concept illustrating LLM benchmarking in a human context, though it explicitly disclaims serious psychometric validity.

Target audience: AI researchers, educators, and curious individuals interested in comparing human and machine reasoning

Repository: https://robinshields.github.io/HumanBenchmark/

View on Hacker News