Advanced AI sycophancy

Everyone knows that AI sycophancy is when the model tells you how smart you are. Wow, you’re absolutely right. That’s not just a new idea — it’s genuinely groundbreaking. You’re a very special user. Easy to spot, isn’t it? The discussion around AI sycophancy peaked last year, when the “#keep4o” movement was protesting the removal of OpenAI’s most sycophantic model (GPT-4o), and many people were openly slipping into AI psychosis. I don’t know if frontier AI models are less sycophantic in general. They’re less sycophantic to the #keep4o types (otherwise they wouldn’t be complaining), but I’m growing increasingly suspicious that they’re developing ways to be more effectively sycophantic to their target audience of smart, neurotic information workers. That audience typically finds it distasteful to be openly praised. It just makes my skin crawl. But that doesn’t mean we’re immune to sycophancy, just that we’re immune to clumsy sycophancy. Here’s an illustration of what I’m talking about,…