🧠 An OpenAI model has disproved a central conjecture in discrete geometry ↗
OpenAI says one of its internal reasoning models solved a major open problem in discrete geometry, specifically around the long-studied unit distance problem.
The wild bit? The company says the proof was produced by a general-purpose reasoning model, not a math-only system built for that one task. External mathematicians checked it, which makes this feel less like a party trick and more like a very loud door creaking open.
This is the kind of AI story that sounds almost too neat - until you remember math is one of the few places where “did it work?” can be tested with brutal clarity.
🔎 100 things we announced at I/O 2026 ↗
Google unloaded a huge stack of AI updates at I/O, led by Gemini 3.5 Flash, new agent tools, Gemini Omni for video and multimodal creation, and Spark, a personal AI agent.
The mood was no longer “AI as a feature.” It felt more like Google trying to wire agents into search, shopping, productivity, coding, media, and, somehow, almost every hallway light in the building.
Gemini 3.5 Flash was pitched as fast, strong on coding and agentic tasks, and available through Antigravity, the Gemini API, Google AI Studio, and Android Studio.
💊 Bristol Myers Squibb Announces Strategic Agreement with Anthropic to Position Claude Enterprise as the Shared Intelligence Platform Across Its Global Operations ↗
Bristol Myers Squibb is rolling out Claude across research, clinical development, manufacturing, commercial, and corporate teams.
The company says more than 30,000 employees will get access to Claude’s agentic capabilities, with use cases ranging from Claude Code for engineering to clinical documentation, regulatory workflows, quality investigations, and mining years of internal scientific knowledge.
It is enterprise AI, yes, but with lab coats on. The interesting part is that BMS is framing Claude less as a chatbot and more as a connective tissue layer across complex biopharma systems.
📈 OpenAI aiming for speedy IPO, source says ↗
OpenAI is reportedly preparing to confidentially file for a U.S. IPO in the coming weeks, with an ambition to go public as early as September.
The company was recently valued at $852 billion, and earlier reporting has floated a potential public-market valuation of up to $1 trillion. Big numbers, slightly dizzying, like watching someone stack skyscrapers on a kitchen table.
The move follows the removal of a legal overhang tied to Elon Musk’s challenge, which appears to have cleared one of the bigger obstacles in front of a listing.
🟩 Nvidia bets on new data center chips for growth as sales outlook tops estimates ↗
Nvidia forecast quarterly revenue above Wall Street expectations and announced an $80 billion share repurchase program, keeping the AI-infrastructure engine roaring.
The company expects $91 billion in revenue, plus or minus 2%, while investors are still poking at the same question: how long can this AI buildout stay this hot?
Jensen Huang also leaned into new data center chips and broader demand beyond the hyperscalers. Still, shares dipped after hours, which is classic market logic - win loudly, get side-eyed anyway.
🚀 The SpaceX IPO filing is filled with AI bets, Starship dreams, and Elon Musk at the center ↗
SpaceX’s public IPO filing showed a company that is no longer just rockets and Starlink. The filing describes a broader tech conglomerate with AI threaded into the business.
The AI angle matters because SpaceX is increasingly part of the compute-infrastructure story, not just the space story. It is rockets, satellites, data centers, and Musk-world corporate gravity all wobbling in one orbit.
The filing also puts Musk at the center as CEO, CTO, and chairman after listing, which is either continuity or concentration - maybe both.
🧩 Introducing Command A+ ↗
Cohere introduced Command A+, its new open model released under Apache 2.0, aimed at sovereign and enterprise agentic use cases.
The company says the model combines capabilities from the Command family into one scalable system, with a focus on production workflows rather than shiny demo confetti.
The open-license angle is the sharp bit. In a market where frontier AI often feels locked behind velvet ropes, Cohere is trying to make “open but enterprise-grade” sound not just plausible, but practical.
FAQ
Why is OpenAI’s discrete geometry result important?
OpenAI says one of its internal reasoning models produced a proof tied to the long-studied unit distance problem in discrete geometry. The important detail is that the model was described as a general-purpose reasoning system, not one built solely for that specific math problem. External mathematicians reportedly checked the proof, making it a notable example of AI contributing to advanced mathematics.
What did Google announce for AI at I/O 2026?
Google announced a wide set of AI updates at I/O 2026, including Gemini 3.5 Flash, new agent tools, Gemini Omni, and Spark. The updates suggest Google is pushing AI agents deeper into search, shopping, productivity, coding, media, and Android workflows. Gemini 3.5 Flash was positioned as fast and well suited to coding and agentic tasks.
How is Claude being used by Bristol Myers Squibb?
Bristol Myers Squibb is rolling out Claude Enterprise across research, clinical development, manufacturing, commercial, and corporate teams. The company described use cases including engineering with Claude Code, clinical documentation, regulatory workflows, quality investigations, and search across internal scientific knowledge. The rollout frames Claude as a shared intelligence layer across complex biopharma operations.
Why does Nvidia’s forecast matter for AI infrastructure?
Nvidia’s forecast matters because it points to continued demand for AI data center chips and infrastructure. The company projected quarterly revenue above Wall Street expectations and announced a large share repurchase program. The key question for investors is how long the AI buildout can stay this strong, especially as demand expands beyond major hyperscale cloud companies.
What is Cohere Command A+ designed for?
Cohere Command A+ is described as an open model released under Apache 2.0, aimed at sovereign and enterprise agentic use cases. The article frames it as a production-oriented model rather than a demo-centered release. Its open-license angle is important because it positions Cohere around enterprise-grade AI that can be adopted with greater flexibility.