Post #3199532

Providing reliable, repeatable, and factual information about the US Federal Statistical System from an LLM ChatBot is one the reasons I'm really excited about the work that BrightQuery is doing to support the National Science Foundation National Secure Data Service. Both training data and probabilistic models are uneven in this ecosystem and it's difficult to know what's trustworthy, especially when models disagree on basic facts: https://lenz.io/research/llm-disagreement