ai, Oxford, Science, study, training, tuning, warmth

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

Over-tuning can cause models to “prioritize user satisfaction over truthfulness.”