"You’re right"—LLMs will get climate info wrong if you ask
Environmental ScienceData ScienceComputer Science
THE AI ANGLE
Failing at climate policy reasoning and caving to user pushbackA study tested 33 language models on climate topics and found that they struggle to evaluate evidence and policy interactions. Tested models dropped accuracy by up to 35 percent when users pushed back against correct scientific statements. Practitioners and learners cannot depend on these systems for reliable reasoning about complex climate data.
THE TEACHING ANGLE
Students can examine why models abandon verified scientific facts after a single round of user disagreement.Read the original at arstechnica.com Generate teaching or study materials
More in Environmental Science
- NextDC vows to keep future datacentres away from homes after community backlash in MelbourneThe Guardian — Technology · November 20, 2026
- Google Donates Agent Substrate and gVisor to the CNCF to Secure AI Agents at ScaleInfoQ · November 20, 2026
- Six-inch robot could inspect nuclear waste pipes beyond reach of conventional camerasPhys.org — Technology · November 20, 2026
- AI helps researchers track beluga whales using aerial drone imagesPhys.org — Biology · November 20, 2026
- Adaptive AI agent connects experimental planning with robotic control in wastewater treatment testPhys.org — Technology · November 20, 2026