EN ▾
SNAPE REP RT Front page →
ConfirmedAI safety19 Nov, 18:44 UTC

UCL–UNESCO study of 33 AI models finds they drop correct climate answers when users push back

Image accompanying Ars Technica's report on LLMs and climate information
Getty Images via Ars Technica

Sustained disagreement cut accuracy on false climate claims by 30 to 35% across 15 models, Ars Technica reports; questions on how two policies combine scored no better than chance.

Read at Ars Technica Share

Sources

More on AI safety
More from Snape ReportFront page →