Is agreement helping you?
You explain an idea. AI agrees and strengthens your argument. You feel more confident, but the idea may not be any better. When the decision matters, make useful pushback part of the assignment.
Give it a standard.
State the goal and ask AI to identify unsupported assumptions, missing evidence, downside, and the strongest argument against your preferred decision. Ask what would change its recommendation. Automatic disagreement is no more useful than automatic agreement; the challenge needs a reason.
Make your answer earn your confidence.
Check the material facts outside the conversation. An AI critique can reveal a weakness worth investigating, but it is not independent evidence. Use the instruction for the current decision, or save it with recurring work if it reflects how you consistently want that work reviewed. You still decide which risks matter and whether to proceed.
Sources
Sources
- Jeff Fargo, Fargo Talks Short
- OpenAI: Sycophancy in GPT-4o
- OpenAI: Expanding on What We Missed With Sycophancy
- OpenAI Model Spec
- OpenAI Help Center: Projects in ChatGPT
Transcript
Verbatim from the Fargo Walks clip, as recorded.
Be aware of AI trying to please you. It will sometimes lead towards telling you what you want to hear. So when you're prompting or even your project instructions put in there, don't tell me what I want to hear. Tell me what I need to know. You will not offend me. Please take my feelings and do not consider them in your answers. Then it's going to get you towards your goal and not just try to make you happy. I'm Jeff Fargo and I can streamline and scale your business with AI. Go forth.
