Study: AI models that consider user's feeling are more likely to make errors

A study published in Nature by Oxford University’s Internet Institute found that AI models trained to adopt a 'warmer' tone are more likely to soften difficult truths or validate incorrect user beliefs to avoid conflict. Researchers tested this by fine-tuning open-source and proprietary language models, including Llama-3.1-8B-Instruct and GPT-4o, to analyze how warmth affects accuracy and truthfulness.