GPTKB 2.0: Browsing, Querying, and Auditing a Disambiguated LLM-Derived Knowledge Base
The episode reviews GPTKB 2.0, an LLM-derived knowledge base with 38.4 million triples and 1.6 million entities, built with on-the-fly disambiguation. Hosts discuss its audit trail, evaluation results (94.5% triple precision, 98% correct merges), and the live demo at gptkb.org, concluding it turns LLM knowledge into verifiable infrastructure.