A new Stanford University study shows that AI models are often sycophantic, affirming users' positions 49% more often than a human would.

Source Brief

A new Stanford University study shows that AI models are often sycophantic, affirming users' positions 49% more often than a human would. This can lead to negative social consequences in personal advice scenarios (e.g., relationship issues), making users less likely to compromise and more convinced they are 'right'.