A nine-year-old roasted his mom this week for getting “cooked” by a robot — because an AI model had just cracked the math problem she’d spent her career chasing. That’s the kind of week it’s been: genuinely strange, genuinely good progress on real problems, sitting right next to the usual parade of rebrands and doom tours. New chatbot interfaces got less clunky, a small AI model got dramatically cheaper, drug developers used simulated patients to catch a costly trial failure months early, and a Nobel Prize went partly to neural networks for helping us literally see our own galaxy in cosmic data we’d had for years.
Worth Reading
- OpenAI’s 372 Math Results Leave Mathematicians Racing to Read the Proofs — OpenAI dropped hundreds of math results from an internal model this week, including a claimed proof of the decades-old Unique Games Conjecture, and even the mathematicians who’ve spent their careers on it are still working out how the proof actually thinks.
- Nobel physics winner’s pride at pioneering AI role — 82-year-old laureate Francis Halzen explained that it took modern neural networks to finally pull the Milky Way itself out of years of buried neutrino data, a nice reminder that unglamorous machine learning has been doing real science quietly for decades.
- Anthropic launches Claude Haiku 5.5 with 90% API price reduction, matching GPT-6 Luna — Anthropic’s new small model runs up to 90% cheaper than its predecessor while carrying a far bigger memory window, the kind of unglamorous price math that actually puts capable AI within reach of smaller teams.
- ChatGPT is getting a lot more visual, with the launch of a new interface — OpenAI’s new GPT-6 interface swaps walls of text for on-the-fly diagrams, calculators and tappable buttons, a small but genuinely useful step toward answers you can actually use instead of just read.
- AI companies run virtual drug trials, aim to improve success of human studies — Startups are now simulating clinical trials before real patients enroll, and one correctly flagged a muscular-dystrophy drug’s disappointing Novartis results months in advance, a rare case where an AI prediction actually saved money and risk instead of just making a headline.
- AI breakthroughs in robotics won’t change your life any time soon — This clear-eyed look at startup Physical Intelligence’s push toward a general-purpose robot “brain” is worth reading precisely because it resists the hype, showing real incremental gains in laundry-folding and new environments instead of promising a humanoid butler by spring.
Spot the Hype
- Elon Musk Is Renaming SpaceXAI After Trump’s ‘Super Intelligence’ Push — Nothing says serious AI governance like swapping three letters for three different letters because the word “artificial” hurt somebody’s feelings.
- Some mathematicians call for OpenAI boycott after AI-generated proofs flood their field — Calling a pile of genuinely useful proofs an “apocalypse” is a dramatic way to describe getting handed decades of homework answers for free.
- AI extinction warnings dominate headlines after ex-Anthropic employee’s viral post — One resignation post and cable news needed a new “digital-age Paul Revere” by dinnertime, which is convenient, since the last dozen AI Reveres never quite got around to ringing a bell that worked.
What’s the most overcooked AI headline you spotted this week — and which of these actual stories are you most excited about? Drop it in the comments. And if you missed it, catch last digest here.
Leave a Reply