You can be a genius and still fail to get good outputs with AI.
Recent research shows that great AI outcomes depend less on IQ, experience, or prompt syntax, and more on how well you can model how the AI is interpreting the task. In cognitive science, this is called Theory of Mind (ToM).
The people who consistently get superior outputs tend to have intuition for collaborative intelligence.
They think in terms of:
- What the model knows right now
- How it’s likely interpreting the instruction
- Where it might misread the goal
That ability predicts performance with AI far better than intelligence or technical fluency.
Patterns from the research
- When users showed clearer perspective-taking in their message, the next response from the model improved — meaning output quality depends heavily on how you engage in that specific moment.
- Gifted solo problem-solvers often plateau when paired with AI; collaborative thinkers pull ahead.
- Static prompt collections explain very little of the performance gap. Interaction quality explains most of it.
What this looks like in practice
- Calling out context the model doesn’t have instead of assuming it will infer it.
- Framing the reasoning path you want before asking for an answer.
- Asking the model to justify its assumptions, then correcting them — not just tweaking wording.
This also tracks closely with my own experience using AI. I get consistently better outputs in follow-ups when I start with meta prompts like:
“How should I prompt you to get the best result here?”
“What context would materially improve your answer?”
It’s less prompt engineering, more partnering with the system.
The uncomfortable implication
This ability isn’t static. It fluctuates. If you’re tired, rushed, or distracted — and your collaborative skill drops — your AI outputs degrade with it.
That’s why this matters beyond individual use. Forbes recently framed “AI skills” as a multi-billion-dollar question: what actually differentiates people who are productive with AI from people who aren’t?
The answer isn’t technical fluency or prompt libraries. It’s a hybrid skill set that sits between communication, reasoning, and system-level understanding — the ability to think about how the system is thinking, and adjust in real time.
People who develop that skill compound fast. People who don’t tend to stall, no matter how powerful the model gets.
Paper: Quantifying Human-AI Synergy, by Christoph Riedl and Ben Weidmann.