När når vi AGI?

When John Pardon reappears on the Zoom screen after checking in on a loved one, he graciously offers to call again if we’re running out of time. “Now with the kids, I feel like whatever time I might try to plan on just gets thrown out the window,” he says, laughing. Then he swings right back into explaining a problem that has defined his recent work: a two-decade-old conjecture that two rival methods for counting curves were secretly saying the same thing.

Pardon is one of the 2026 Fields Medalists. His work across symplectic geometry,1 topology,2 and knot theory3 has introduced solutions to longstanding problems, and new mathematical machinery, tools, and frameworks for approaching the field at large.

But math is just one of Pardon’s many dimensions. Midway through our call, I hear him gently ask his son to quiet down—声音小一点好不好?—in fluent Chinese. He tells me about the many years he’s been playing cello and juggling with friends alongside his pursuit of mathematics. By the end of the call, we’ve switched to Mandarin. And, as Pardon describes his hope to one day learn 書法 (calligraphy), it becomes clear that all his seemingly unrelated activities express the same underlying force: curiosity—which Pardon hopes to cultivate in his students and his sons.

This interview was condensed and edited for clarity.

Only in the last month has it become useful. What it’s really good at is proofreading and finding errors. It used to be like running a math document through spell check. But now the AI clearly can understand everything I write. Even when I write the most terse, badly explained, unclear sentences, with some sort of personal notes or suggestions of how I think the proof would go, and any other human would say, “What is this? Just rewrite it so that it’s readable”—the AI understands. 100%.

A Physicist’s Confession: During my doctoral dissertation, I spent a year and roughly 500 pages doing quantum gravity calculations by hand, but the Opus 5.5 paper and GitHub repository solved it in 10 hours. 😱 On top of that, it found the errors. Fortunately, the results didn’t change.

What will “work” mean in academia going forward?

Introduction
For the past three months, the Citrini Research team has been working with Anthropic’s internal research model1 to see if we could create novel proteins in a computer that would do what we wanted them to in an actual lab.

On the morning of August 12th, I woke to a Slack notification that our autonomous drug design campaign was concluding and the results were being compiled.

The previous day, from a nondescript conference room in the Anthropic offices, we prompted Claude to go from an idea to a package ready to submit to a physical lab. Overnight, Claude spawned more than 300 sub-agents using more than 300 concurrent GPUs to design, score, and deliver a set of proteins. At our direction, it sent them to the lab for synthesis and testing without any further input from us.