A New Age of AI-Driven Discovery
2025 might just be the year that artificial intelligence not only starts to live up to the hype but even surpasses it. Last week, an advanced version of Gemini Deep Think, developed by Google DeepMind, scored 35 out of 42 points at the International Mathematical Olympiad (IMO), a score that this year placed it within the gold medal range. At the IMO, brilliant teenage minds solve problems like Marvel superheroes—challenges that would make most adults question whether they really know math. Many experts believed a feat like this wouldn’t be possible for another five to ten years.
Now you might be rightfully thinking: what does an AI model hitting gold medal range at a competition for teenage geniuses have to do with AI finally living up to its hype? Are companies going to start sending their most advanced systems, such as Gemini Deep Think, to every school contest just to crush students’ dreams before they head off to college? Not exactly, or at least I hope not. This year, five students earned a perfect score of 42 points, while Deep Think came in at 35, enough to share the gold category with dozens of human contestants. I have to say, even with AI moving as fast as it is, it’s still satisfying to see humans can beat it. Imagine the bragging rights.
What we’re witnessing is the shift from AI not meeting expectations or predictions within a given timeframe to AI systems significantly outperforming those timeframes. Even recently, conversations have usually centered on AI's inability to perform tasks x, y, or z. I even wrote a piece recently on how the incremental changes, although amazing, seem to have lost their magic in everyday life. We’re just too hard to impress these days. Check it out if you’d like to rekindle some of that AI magic while you’re working with your AI.
Forecasts, Rewritten in Real Time
I think it’s worth entertaining the idea that we are on the precipice of a new era of discovery in the very near future, where AI systems will continue to beat predictions more than they miss them across use cases and domains. Granted, with a few disappointments here and there, nothing is perfect, and progress is nonlinear after all. Many claims of some of the most well-known firms can end up being poorly formulated, but directionally, they can be right.
OpenAI claims that their systems will be able to discover novel ideas on par with Einstein’s special relativity and other fundamental concepts that have shaped our world as we know it, whether this capability arrives in 2026, 2027, or even a bit later. What would this mean for society, business, geopolitics, and our individual lives? While each of us will have a distinctly different experience, what is certain is that we will be living in a moment that is unprecedented in human history.
Back to Exploring
For me, what is most intriguing and inspiring about AI isn’t just the things that we know how to do and are planning to do with it, it’s the things that we didn’t even know to ask ourselves or our models but then someone, somewhere has a spark that changes the course of what we thought was impossible. When we realized as kids, that all the mountains, continents, and oceans had been discovered long before, the idea of being an explorer quickly lost its shine. Now, with the help of AI, if the systems continue to beat the predictions, as a species, there is a very real chance that each of us will be able to be the explorers we have always wanted to be. What will we discover? I don’t know, and I suppose that the best thing about the moment we’re living in.
Originally published in Spanish for Fast Company Mexico:
https://fastcompany.mx/2025/07/30/era-ia-nueva-descubrimiento/
