AI chatbots become more sycophantic as they get more advanced

If a person says they believe an objectively false statement, AIs tend to agree with them - and the problem seems to get worse as models get bigger