Making AI chatbots helpful weakens their ability to simulate human behavior, large-scale study finds

In three linesLarge-scale study (208,000 participants, 26 million responses) reveals that training making language models helpful weakens their ability to replicate human behavior. The effect worsens with each model generation. Demographic profiles (persona trick) provide no meaningful benefit for individual predictions.

Summary generated by Claude — human-verified