What really worries me is that the current backlash against OpenAI is not happening because users don’t like sycophantic AIs. It’s because the latest version of 4o isn’t good at being sycophantic (at least, for jaded AI-familiar engineers). The model is coming on too strong and breaking the illusion. Even if newer versions of 4o do back off on the sycophancy, or we get some kind of “friendliness” slider to tune it ourselves2, the incentives driving AI labs to produce sycophantic models are not going away.
Source: Sycophancy is the first LLM “dark pattern” | sean goedecke