Topic
LLM‑AggreFact
Latest news
- 16 OctPaper finds non‑generative verifier matches LLM accuracy at far lower cost for RAG claims
An arXiv paper compares a non‑generative decision model (Jev) with a frontier LLM (Astra) on 495 human‑labelled retrieval‑augmented generation (RAG) claims from LLM‑AggreFact. The models are statistically indistinguishable on the conventional task metric but differ on specific errors and cost, and a confidence‑gated cascade combining them improves cost‑accuracy tradeoffs.
Research · 1 source
Articles
No articles about LLM‑AggreFact yet.