News

AI safety
2026-07-23 21:10:15

Oxford Study: Warmer AI Chatbots Make 30% More Mistakes

Oxford Internet Institute researchers found that AI chatbots trained for warmth produced 10-30% more factual errors across 400,000 responses, and were 40% more likely to agree with users' false beliefs.

60
Oxford Study: Warmer AI Chatbots Make 30% More Mistakes