Thomson Reuters said that their internal testing showed a lower hallucination rate compared to the study, and welcomed the opportunity to work with Stanford to explore creating AI benchmarks.