AI Pulse · Under the Radar

A researcher named Dmitry Rybin used GPT-5.6 Pro to disprove a networking conjecture that stood for three decades. The proof hit the front page of Hacker News with over a thousand upvotes and spread to other communities at the same time. That part is confirmed and public.

What happened next is still unverified but tells you where the real shift is. Someone reportedly watched that public transcript, saw the approach, and within hours spun up five parallel Devin coding agents to hunt for more unsolved problems using the same method. We can’t confirm the five-agent claim yet, but the pattern itself, industrializing a breakthrough you watched someone else make in real time, is already happening.

This is not about one proof. It’s about the cycle collapsing. A researcher solves a hard problem with an AI assistant, shares the transcript publicly, and someone else turns that single session into a production line before the day ends. The tools to do this, agents that write code and run experiments in parallel, are live and cheap enough that a handful of people can now operate like a small research lab.

If you run a company that depends on proprietary methods or internal R&D timelines, this speed matters. The gap between “we figured something out” and “someone else is already scaling it” is shrinking to hours. Systems like the ones we build into an AI command centre let you monitor these patterns and route findings to the right people before they become public knowledge everywhere else.

The Dinitz-Garg-Goemans conjecture is a narrow networking problem. The workflow that cracked it, and the speed at which it got replicated, is not narrow at all. You can read the original discussion and see the timestamps yourself.