📄 中文摘要
构建的领域特定人工智能系统在某一主题上的知识深度超越任何通用模型，拥有三百万个向量和252,000个图节点，索引人与事、地点、事件及录音之间的关系。该系统涵盖了60年的现场音乐历史，包括每场演出、每个曲目和每条社区知识。当询问GPT-4或Claude关于该领域的问题时，得到的答案虽然结构清晰且自信，但常常错误，出现日期幻觉、场地混淆、事件合并等问题，且过于依赖训练语料库中最著名的数据点。
领域特定的人工智能为何优于通用型人工智能
出处: Why Domain-Specific AI Beats General-Purpose for Everything That Matters
发布: 2026年3月12日
📄 English Summary
Why Domain-Specific AI Beats General-Purpose for Everything That Matters
A domain-specific AI system has been developed that possesses deeper knowledge about a particular topic than any general-purpose model can offer. This system includes three million vectors and 252,000 graph nodes that index relationships between people, places, events, and recordings, covering 60 years of live music history, including every show, setlist, and piece of community knowledge. When querying GPT-4 or Claude about this domain, the responses are often confident and well-structured but frequently incorrect, with issues such as hallucinated dates, venue confusion, event merging, and an overemphasis on the most famous data points due to their training corpus.