Google DeepMind has announced Genie 4, a world model that generates audio-equipped 3D worlds in real time. Scene consistency now holds for up to two minutes — twice as long as the previous generation. Meanwhile, reports are circulating that frontier AI models have disproved the Exponential Time Hypothesis, a long-standing assumption in computational complexity theory. With a new standard proposal for local AI in the browser and, at home, a suspected cloud breach at GMO Internet, it was a day of movement across generation, research, infrastructure, and security.

Genie 4 — A World Model That Generates Audio-Equipped 3D Worlds in Real Time

On October 27, Google DeepMind unveiled Genie 4, its new-generation world model. A world model is a technology that generates not "materials" like text or images, but "the environment itself." Genie 4 simulates interactive 3D worlds in real time, and this release newly adds audio and characters you can converse with.

According to TechCrunch, it can hold scene consistency for up to two minutes — twice as long as Genie 3, its predecessor. Improvements to physical behavior and more human-like character rendering are also reported. Under the hood, a Gemini model handles planning in the background, a design that underpins the simulation's consistency. The Decoder described this design as "a modern take on agent design that predates LLMs."

That said, Genie 4 is still an early research model, and reports say visual inconsistencies appear as sessions run longer. Given potential applications in gaming and robotics, the two-minute duration looks like a number with plenty of headroom for future growth.

Has the Exponential Time Hypothesis Been Refuted? — A Foundational Assumption of Complexity Theory Wobbles

TechCrunch reports that several frontier AI labs claim their models have refuted the Exponential Time Hypothesis (ETH), a long-standing conjecture in computational complexity theory. For roughly 30 years, ETH has served as an assumption in computer science positing that some computational problems inherently require exponential time.

On this story, TechCrunch spoke with an MIT computer scientist who helps explain OpenAI's findings. New Scientist also reported that AI has disproved a 30-year-old mathematical assumption in computer science. At this stage, the focus seems to be on how to interpret the labs' claims.

Also making news: Fields Medal–winning mathematician Terence Tao proposed "Maths 2.0," a vision for how mathematics can survive the AI era, in Nature. With AI now shaking the foundations of mathematics, it was a day when an overlapping debate emerged over how to reorganize the practice of mathematics itself.

Opera's "WebAI" — A Proposed New Standard for Running Local AI in the Browser

Opera has proposed WebAI, a new web standard that would let users run AI locally inside the browser. According to TechCrunch, it is designed to be independent of any specific model and aims to work with a wide range of AI models. At a time when going through cloud APIs is the norm, the move draws attention as a browser-vendor-led effort to standardize local execution that keeps data from ever leaving the device.

GMO Internet Cloud Breach — Verifying Reports That Customer Data Was Published

TechCrunch reports that GMO Internet, a major Japanese internet company, was hit by a breach of its cloud services and is verifying the authenticity of information that a suspected hacker published stolen customer data. The situation is still under investigation, and details will have to wait for future announcements. As a breach at a Japanese provider, this is a story worth watching as it develops.

General Compute Raises $100 Million Series A — Pushing Back on Neocloud Pessimism

As market turmoil continues around GPU clouds — the so-called neoclouds — TechCrunch reports that General Compute has raised $100 million in a Series A round. The company's strategy bets on inference chips outside the NVIDIA ecosystem, and the coverage frames the question as whether it can become a contender in a market where pessimism evocative of a "neocloud collapse" is spreading.

Bank of America: "Don't Share Your Banking Passwords with AI Assistants"

TechCrunch reports that Bank of America has asked customers to refrain from sharing their banking passwords with AI assistants. The bank also reportedly reminded customers that "you are responsible for activity arising from AI agents' access." With agentic AI spreading, financial institutions are now the ones drawing a new line over how far credentials can be delegated.

---

The staying power of world models, the foundations of mathematics, browser standardization. Watching the "givens" around AI get updated bit by bit seems worth a daily check. We'll keep following these stories in the next roundup.