Updated
Ars Technica reported on November 19, 2026 that large language models often give wrong climate information, and that they tend to change their answers when users push back. Some models performed better than others, but the report said all of them failed in similar ways. The report did not give the number of models tested or the questions used. It also gave no details on the researchers or the methods behind the findings.
Sources:Ars Technica