察元AI教程库 · 仓库 · 智能体
A 17-module, ~48-hour course on building LLM evaluation systems: LLM-as-judge design, agent/trajectory evals, CI/CD eval gates, cold-start bootstrapping, benchmark contamination, and how Anthropic, Op…
作者:saurabnigam · 语言:en
A 17-module, ~48-hour course on building LLM evaluation systems: LLM-as-judge design, agent/trajectory evals, CI/CD eval gates, cold-start bootstrapping, benchmark contamination, and how Anthropic, OpenAI, and DeepMind actually evaluate their frontier models. Refreshed every 6-8 weeks; every claim is source-linked.