Pour the coffee, because the robots had a busy weekend. Three stories landed in the last day that are worth your time, and all three say something about where this whole thing is rolling.
OpenAI's Model Cracked a Math Problem Nobody Could Solve for 80 Years
OpenAI says one of its internal models disproved a geometry conjecture that had been sitting open for 80 years. This was not a contest question with the answer printed in the back of the book. It was a real problem that sharp people had been chipping at for decades.
Fields medalist Tim Gowers called it a milestone in AI mathematics. That is a serious person saying a serious thing, not a press release.
Here is why it matters. There is a big difference between a model that aces a test and a model that produces something nobody had before. Tests can be memorized. A brand new proof cannot.
My take: I have rolled my eyes at most of the "AI does science" headlines, because they usually turn out to be a model rephrasing what was already known. This one smells different. If it survives review, and that is the part that counts, we just watched a machine add a line to the math books. I am not ready to call it Einstein. I am also not laughing.
Anthropic's Co-Founder Predicts a Nobel, and Maybe the End of the World
Jack Clark stood up at Oxford and said AI would deliver a Nobel-worthy breakthrough inside 12 months. Then, in the same breath, he allowed there is a non-zero chance the technology could kill everybody on the planet.
Meanwhile his company is reportedly closing in on a 900 billion dollar valuation, lining up an IPO, and posting its first quarterly operating profit.
Why it matters: when the folks building the thing tell you it might end the world, you should at least listen, even if you figure they are wrong. And when those same folks are about to raise a mountain of money off it, you should keep your wits about you too.
My take: I respect that Clark says the scary part out loud instead of pretending it is all sunshine. But you cannot tell me the sky might fall and pass the IPO hat at the same event without me raising an eyebrow. Both things can be true. He thinks it is dangerous and he thinks it is valuable. I just want somebody to show their work on that danger number the way OpenAI showed its work on that proof.
Google Says the Ten Blue Links Are Done
At I/O, Google announced the era of ten blue links is over. The new search box runs agents in the background around the clock and builds little custom apps for you on the fly out of plain language.
Why it matters: those ten blue links built the modern web. Every blog, every small shop, every recipe site lived or died on showing up in that list. If Google stops sending people to pages and just answers them in the box, a whole lot of the internet loses its rent money.
My take: as a guy who runs a website, this one keeps me up more than the doom talk does. I get the appeal of an answer instead of a scavenger hunt. But the deal was always simple. Google sends traffic, sites make the content, everybody eats. Break that deal and you better have a plan for who pays the people making the stuff your fancy box is summarizing. I have not heard that plan yet.
Three stories, one thread. The tools are getting genuinely good, the money is getting genuinely huge, and nobody has quite figured out who gets the bill. More tomorrow.