Superhuman through Prompt Engineering

o3 from OpenAI is superhuman at GeoGuessr—this means it is more accurate than the very best humans at the photo-location guessing game. Sure, plenty of AI systems are vastly better than humans at specific tasks. However, unlike AlphaGo or AlphaStar, o3 was not specifically trained for this task. Its ability simply emerged, a byproduct of sheer scale and complexity.

There’s an important caveat: out of the box, o3 is merely on par with top human players. However, arm it with a long and complex prompt, and its impressive abilities amplify, bringing it head and shoulders above any human. Try it for yourself—read the article and use o3 with and without the prompt to see the difference.

It’s a very clear example of how prompt engineering, if done right, is a real thing. It’s not easy, though; crafting a good prompt requires three difficult things:

1. Precision Language: It's not quite programming, as you're wrestling with natural language. It's harder. There are no comforting, deterministic guardrails of strict syntax and known algorithms. Instead, you're coaxing nuanced behavior from an intelligent system that's immensely powerful but less predictable.
2. Subject Matter Expertise: You can't improve upon a model's GeoGuessr skills if you don't deeply understand what makes a human good at it. You need to know why and how it's making mistakes to make it better.
3. Good Ol' Scientific Method: Since these models aren't entirely deterministic, there's an iterative dance: hypothesize, prompt, test, observe, subtly tweak, and repeat.

It’s a blend of art, science, and linguistic finesse. Perhaps this is the most compelling argument yet for a true liberal arts education—one that doesn't shun science and formal logic but embraces them alongside literature, history, and art. Knowing how to use models effectively—‘prompt engineering’—might just be the most vital skill for our evolving future.

“All reality is a game. Physics, at its most fundamental, the very fabric of our universe, results directly from the interaction of certain fairly simple rules and chance; the same description may be applied to the best, most elegant, and both intellectually and aesthetically satisfying games. By being unknowable, by resulting from events which, at the sub-atomic level, cannot be fully predicted, the future remains malleable, and retains the possibility of change, the hope of coming to prevail; victory, to use an unfashionable word. In this, the future is a game; time is one of the rules.”
― Iain M. Banks