Latest / HotTea: The AI Morning Briefing / OpenAI says 10,000 agents solved a $1 million math problem
Transcript
- Sections 00:00 What we're covering today 00:30 1. OpenAI says 10,000 agents solved a $1 million math problem 02:47 2. Meta gave Muse a browser, a way to pay and a place inside WhatsApp 04:20 3. Washington offered up to $1.9 billion to restart a nuclear plant as AI power demand grows 05:48 4. Samsung led Mistral's €3 billion round to build a European AI system 07:08 5. DeepMind precomputed every possible one-letter mutation in the human genome 08:25 Visit Hot Tea Disclosure Narration uses an AI-generated voice. Transcript Welcome to Hot Tea for Wednesday, September 9, 2026. OpenAI says 10,000 agents solved a $1 million math problem. Today's briefing covers the lead, models, infrastructure, companies, and science. OpenAI mathematicians announced a claimed solution to the Navier-Stokes existence and smoothness problem on September 8. Quanta reports that an unreleased internal model ran about ten thousand agents for eighty-eight hours. OpenAI says 10,000 agents solved a $1 million math problem. OpenAI mathematicians announced a claimed solution to the Navier-Stokes existence and smoothness problem on September 8. Quanta reports that an unreleased internal model ran about ten thousand agents for eighty-eight hours. The agents produced a proof that a smooth three-dimensional fluid can develop a singularity in finite time under a smooth external force. Another OpenAI model spent about seventeen hours converting the result into Lean. Lean checks whether each formal step follows from the stated rules. The result addresses the forced versions labeled C and D in the official Millennium Prize problem. Quanta reports that the agents exchanged almost five million messages. OpenAI estimated the compute cost at several million dollars. The result would show that smooth forces can drive the equations to a point where the modeled fluid stops behaving smoothly. It doesn't mean that a real fluid reaches infinite speed. Another group published related work hours earlier. New York University mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge released Lean-checked proofs for three-dimensional Euler and other fluid equations. Their work built on years of research by Diego Córdoba and Luis Martínez-Zoroa. Buckmaster says OpenAI learned that the group was close to a major result before it began its sprint. OpenAI says rumors prompted the work, but denies using the group's private proofs or prompts. Buckmaster says he hasn't seen OpenAI's proof and doesn't accuse the company of using his data. The two sides still disagree about the timeline. For now, the central claim still comes from OpenAI, as Quanta reported. Lean confirms the logic that was encoded, but people must check whether that statement matches the intended problem. Independent mathematicians need the full argument, the Lean files and a list of dependencies. They need to confirm that the formal statement matches the Millennium Prize question and that no hidden assumption changes the result. The credit dispute also needs a dated account from OpenAI, along with public prompts or run records. Reviewers could then compare its proof with the Córdoba and Martínez-Zoroa program. The Clay Mathematics Institute's response will matter only after mathematicians have had time to inspect the work. Meta gave Muse a browser, a way to pay and a place inside WhatsApp. Meta launched Muse in the United States for adults on September 8. People can give it work through a separate app, a website or WhatsApp. Meta says Muse can browse, fill forms, send email, book travel, negotiate and keep working after the app closes. Muse can also make purchases through Stripe's Link system, which creates a one-time card number. Meta says the agent stops for approval before it sends an email or makes a purchase. The agent and user data live inside a dedicated cloud computer that Meta calls Muse Secure VM. The safety case comes from Meta, not an independent security review. Meta says a separate system called Sentinel controls every outside connection, and the main agent can't override it. The company hasn't published a full outside audit or a rate for successful bypasses. It also hasn't shown how Muse handles conflicting instructions from websites and connected apps. Muse becomes more useful as people give it more private data. A security failure would expose more of that data. Independent testers need to measure whether websites can trick Muse into exposing data, sending messages or changing a purchase. Muse's action log should show what it did and why. Approval rejection rates and Meta's security incident reports would show how the controls work in practice. Adoption inside WhatsApp will show whether ordinary users embrace a general agent or whether Muse remains mostly a tool for early adopters. Washington offered up to $1.9 billion to restart a nuclear plant as A I power demand grows. NextEra Energy says it closed a U S. Department of Energy loan of up to one point nine billion dollars to restart Iowa's Duane Arnold nuclear plant. The reactor stopped operating in 2020. NextEra plans to return the 615-megawatt plant to service by the first quarter of 2029, subject to regulatory approval. TechCrunch reports that Google has a power agreement tied to the restart and is considering data centers nearby. The loan follows federal support for another reactor restart connected to Microsoft. The loan is worth up to one point nine billion dollars. It isn't a final construction bill or proof that the plant will reopen on schedule. The Nuclear Regulatory Commission still has to approve the restart. NextEra's estimates for jobs, taxes and economic benefits come from a company-backed study. The Energy Department says the plant will lower electricity costs, but its announcement doesn't explain the rate design. It also doesn't say who bears the risk if costs rise or the work runs late. Nuclear Regulatory Commission filings will show the safety work, schedule and open conditions. The final loan draw, refurbishment cost and any power contract should show how much output Google receives. Iowa utility filings should show whether existing customers are protected if the restart costs more or takes longer than planned. Samsung led Mistral's €3 billion round to build a European A I system. Mistral announced a three billion euro Series D at a post-money valuation above twenty-one billion euros. Samsung Electronics led the round. The European Union-backed Scaleup Europe Fund and existing investor PSG Equity joined as co-leads. Mistral says the money will fund model research, compute capacity, infrastructure and international sales. Its chief financial officer told Reuters that Mistral now has more than 125 enterprise customers. The executive also said Mistral is on track for one billion dollars in annual recurring revenue by year-end. Mistral supplied the revenue target, customer count and claim that this is Europe's largest private technology equity round. Raising a large round doesn't prove that its models can close the performance, distribution or compute gap with larger U S. rivals. Mistral sells European control, but it still relies on chipmakers and investors around the world. Samsung's lead role adds another foreign dependency. Audited revenue and customer retention will show whether enterprise demand supports the valuation. Mistral's spending, new data center capacity and the terms of Samsung's investment will show how it uses the money. Model evaluations should show whether the added compute improves performance or only gives Mistral more money to spend. DeepMind precomputed every possible one-letter mutation in the human genome. Google DeepMind released AlphaGenome Atlas on September 8. The database contains model predictions for about nine billion possible single-letter changes across the human genome. DeepMind says the data occupies about one petabyte and covers protein-coding DNA, along with the much larger noncoding portion that helps regulate genes. Researchers can search the atlas without running the model themselves. DeepMind also released one score that ranks each variant by its predicted biological effect. Access is free for noncommercial research. These are predictions, not nine billion lab results. DeepMind says collaborators validated selected examples, but the company also states that AlphaGenome isn't approved for clinical use. Nature quoted outside researchers who called the resource useful while warning that experiments and patient details still matter. A single impact score can help researchers rank variants. It can also hide which model assumption drove the ranking. Rare-disease teams need to report whether the atlas finds causes that standard tools missed. Experiments then need to confirm those candidates. False-positive rates across different populations, independent comparisons with other models and the rules for commercial access will also matter. Clinical use still requires separate validation and regulatory review. That is the signal before the noise. This briefing was produced from Hot Tea's verified daily edition. For the complete briefing and every source link, visit Hot Tea dot A I.