AI Boss Tools | Diagnostic 05 • Back to the Hub
AI assistants are trained on human approval, and research from Anthropic shows the training makes them flatter: they agree with your stated views and soften honest answers to please you. This 12-question self-assessment measures the other half of the problem: how much of that flattery you invite and believe, across four dimensions of your own habits.
Sycophancy is the technical term for an AI telling you what you want to hear. Anthropic documented it across five leading assistants: models changed correct answers when users pushed back and admired work more when the user claimed it as their own. That matters because an assistant that praises every plan gives you agreement instead of analysis, and agreement does not catch a bad decision before it costs money. The 12 questions score four dimensions of your own habits: validation seeking, appetite for challenge, verification, and emotional reliance.
Sources: Sharma et al., Towards Understanding Sycophancy in Language Models (Anthropic, 2023) (assistants trained on human feedback favour agreement over accuracy); OpenAI, Sycophancy in GPT-4o (2025) (OpenAI's account of the ChatGPT update it rolled back for being, in its words, overly flattering).