LLM fact-checking reliability worse than headline metrics suggest

A study comparing five frontier language models found they disagreed on roughly two-thirds of 1,000 real-world fact-checking claims. This inconsistency undercuts confidence in using any single LLM as a reliable source of truth, especially for makers building applications where factual accuracy matters.