AI models that consider user's feeling are more likely to make errors | Dark Hacker News