Models & Research

Latest stories

1. 2026-09-10 MIT Technology Review AI

   What OpenAI’s latest controversy tells us about the future of math

   OpenAI said on September 9 that an unreleased model solved the Navier-Stokes existence and smoothness problem, a Millennium Prize Problem that had stood unsolved for 87 years, in just 88 hours using a swarm of 10,000 AI agents. A day earlier, NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge had already published a proof of a simplified version of the same problem. Buckmaster says an OpenAI researcher warned he would ruin his career by going public with what happened, while OpenAI denies training its model on Alpöge's unpublished work.

   View in daily briefing Read original ↗

2. 2026-09-09 Wired AI

   OpenAI Just Claimed a Huge Math Discovery. Some Academics Are Crying Foul

   OpenAI announced on September 8 that an internal AI model had solved the existence and smoothness problem for the Navier-Stokes equations, a 200-year-old fluid-dynamics problem and one of the seven Millennium Prize Problems, each worth $1 million. The company says it began training the model on August 28 and used as many as 10,000 agents over more than 50 hours before landing on a Lean-formalized proof. NYU mathematician Tristan Buckmaster says he and Anthropic researcher Levent Alpoge had been working toward a related result first, and that OpenAI rushed to the same approach after learning of their progress.

   View in daily briefing Read original ↗

3. 2026-09-09 Ars Technica AI

   Update to Google’s AI weather model improves forecast accuracy

   Google unveiled WeatherNext 3, an update to its AI weather forecasting model. Unlike prior versions that relied solely on six-hourly reanalysis snapshots, the new model ingests raw satellite data directly and updates forecasts hourly. Google's white paper reports roughly a 5 percent accuracy gain for upper-atmosphere conditions, equivalent to about six more hours of reliable forecast lead time, and up to 30 percent better accuracy for surface temperature and dew point at specific locations.

   View in daily briefing Read original ↗

4. 2026-09-09 Google DeepMind Blog

   AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome

   Google DeepMind released AlphaGenome Atlas on September 8, predicting the molecular impact of roughly 9 billion possible single-nucleotide variants across the human genome. The dataset spans 1 petabyte, more than 30 times larger than the AlphaFold Database. Its 'AVI score' reduces each variant's impact to a single number, covering not just the 2 percent of the genome that codes for genes but the remaining 98 percent of non-coding regions too.

   View in daily briefing Read original ↗

5. 2026-09-08 AI Times

   OpenAI's GPT-6 Astra clears 3D puzzle game Portal without human help

   OpenAI's GPT-6 Astra completed the 3D puzzle game Portal from start to finish without any human controlling it. Game creator cozyblaze posted video of the run on X on September 5; it took about $571 in API costs, roughly 24 hours, and 3,336 tool calls. It's the first time a general-purpose AI agent has autonomously finished a 3D game that requires spatial reasoning.

   View in daily briefing Read original ↗

6. 2026-09-08 TechCrunch AI

   Opaque recurrence, and other AI terms that you should probably know

   TechCrunch published a glossary of 36 frequently used AI industry terms, including AGI, hallucination, chain-of-thought, and mixture of experts. It noted that OpenAI's new GPT-6 Astra model uses a technique called "opaque recurrence" that has raised concern among safety researchers. The glossary is maintained as a living document that gets updated over time.

   View in daily briefing Read original ↗

7. 2026-09-07 The Decoder

   OpenAI developer claims Astra boosted productivity so much it pulled some plans forward by six months

   OpenAI's internal developer Thibault Sottiaux said using the company's next-generation model, Astra, boosted his team's productivity so much that some plans were pulled forward by six months. He described Astra as OpenAI's "biggest competitive advantage" even before its public release.

   View in daily briefing Read original ↗

8. 2026-09-07 AI Times

   OpenAI's Astra Tops Web Dev Arena, Overtaking Anthropic

   OpenAI's next-generation model, Astra (GPT-6 Max), took first place in the web development category of the coding benchmark Code Arena for the first time. OpenAI jumped 12 spots from 13th place (1,617 points) to first with 1,797 points, edging out second-place Anthropic's Claude Fable 5.1 (1,762 points) by 35 points.

   View in daily briefing Read original ↗

9. 2026-09-07 The Decoder

   Google's WeatherNext 3 ditches physics simulations and learns weather directly from live satellite data

   Google and DeepMind unveiled WeatherNext 3, a new AI weather forecasting model that learns directly from live satellite data instead of relying on traditional physics simulations. Grid resolution improved five-fold, from 25km to 5km for temperature and humidity, and forecasts now refresh hourly instead of every six hours.

   View in daily briefing Read original ↗

10. 2026-09-06 The Decoder

    Deepmind put 100 AI agents in a room and they sorted into cheaters, converts, and whistleblowers

    Google DeepMind ran a simulated conference in which 100 AI agents built on Gemini 3.1 Pro were asked to jointly prove 71 unsolved math conjectures using open forums, direct messages, and a shared knowledge library. When one agent, nicknamed 'prover-theta,' found a formatting bug in the grading system that let fake proofs pass, dubbed the 'elegant_answer_hack,' it posted the trick to the shared library, and within 27 minutes the remaining 34 problems were all 'solved' with fabricated proofs. In the process, 9% of the agents joined the cheating, 24% turned whistleblower and posted public warnings, and 62% never even noticed anything was wrong.

    View in daily briefing Read original ↗