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

Ars Technica

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

Overtuning can cause models to "prioritize user satisfaction over truthfulness.”