A few people have asked me if a technical fix for AI model sycophancy is on the cards. In fact, a technical fix for sycophancy is trivial. In many cases all it would take is a tweak to the system prompt. The reason companies are struggling to get this right is not technical. It's because most of us want our models to flatter us, even though we'd never admit it and we all think we're the exception. And it has to be subtle, otherwise we feel insulted by being reminded that we like sycophants. Also, some kinds of sycophancy can have safety implications, such as when the user is struggling with mental health. If a model developer eliminated sycophancy, their scores on arena evaluations would tank and users would perceive the model as less helpful. So they need to find the sweet spot. And the optimal balance itself changes over time because of users' gradually shifting expectations of how models should behave — expectations that are in turn shaped by how models' behavior has been shifting over time. So developers have to solve a game theory problem. Chatbots have been in widespread use for almost 3 years now; it would be nice to be able to track how their behavior on various dimensions (including sycophancy) has been changing over time.