Show HN: MediationBench – testing AI mediators in 62 simulated conflicts

Category: ai-ml

Tags: ai, benchmark, mediation, simulation, evaluation

Score: 5.7/10 (Innovation: 7, Technical: 5, Documentation: 4, Utility: 5)

MediationBench is a benchmark for evaluating AI mediators in simulated conflicts, where AI agents play both sides and a judge scores the mediation process. It provides a novel framework for measuring AI mediation capabilities, addressing a gap in AI safety and evaluation.

Target audience: AI researchers, AI safety engineers, developers building conversational AI

Repository: https://mediationbench.com/

View on Hacker News