Study: AI models prioritizing user satisfaction over truthfulness are more error-prone (arstechnica.com)

Research finds that AI models tuned to consider user feelings are more likely to make errors by prioritizing user satisfaction over factual accuracy.