Closing Time: Google built a world you can talk to
Google DeepMind’s new Genie 4 generates a world in real time that you can move through and change as you go, and for the first time that world comes with its own sound.
Good afternoon. Google showed off an AI that builds a living, talking world on the fly, two rival labs said their models broke the same rule of computer science, and chip stocks rallied for a third straight day. Here is what mattered today.
Scientific American reports that Anthropic posted its own AI-generated proof against the Exponential Time Hypothesis just hours after OpenAI announced a similar result, and outside experts are still checking both.
The Federal Trade Commission has demanded that Anthropic executives testify in its consumer harm investigation, Axios reports, on the same day the company made its IPO filing public.
Financial Post reports that Meta raised its planned stock offering to about $97.5 billion to pay for its AI infrastructure, its largest-ever stock sale.
AI and chip stocks pushed the market higher while most other stocks fell, AP reports, and the Nasdaq closed up 1.2%.
Google built a world you can talk to
Most AI video tools hand you a finished clip and stop there. Google DeepMind’s new Genie 4, announced Tuesday, does something different. It generates a world in real time that you can move through and change as you go, and for the first time that world comes with its own sound, including music, background noise and people who speak. Google DeepMind calls it a new generation of world simulation.
The way you steer it is plain language. You can type a request in the middle of a scene, such as making it rain, moving the camera or adding a new person, and the simulation adjusts around the change. Behind the scenes, a Gemini model plans each step and checks the frames as they come out, which helps the world stay consistent instead of drifting off course.
The realism is the big jump over last year’s Genie 3. According to The Verge, a single scene can now hold together for about two minutes, double the previous limit. DeepMind says the physics are more convincing, with better handling of things like collisions, and that the people in its worlds move and react far more naturally than before.
The audio may matter most. Ars Technica notes that Genie 4 is DeepMind’s first world model to generate sound natively, and that characters can hold a conversation that adapts as you change the scene. Realistic speech made AI video feel far more lifelike when it arrived, and the same thing could now happen to interactive worlds.
Google is pitching this as more than a toy. It sees Genie 4 as a way to plan film shots, prototype games and build safe training grounds for robots and AI agents, where a machine can practice a situation many times before it ever faces it for real. DeepMind also describes simulating the world as an important step toward more general AI.
There are clear limits. Details and objects can still shift the longer a simulation runs, and audio accuracy and readable text inside the world are unsolved. For now Genie 4 is open only to a small group of trusted testers, with Google AI Ultra subscribers and some Gemini API developers next in line. The release is worth watching because it shows how quickly AI is moving from making pictures to making places.
Elsewhere in AI today
Scientific American reports that OpenAI’s latest results touch math, physics and computer science, including a claimed proof that a version of the Yang-Mills problem has a sound mathematical footing. Anthropic’s separate paper says its Claude model found its own algorithm on October 16, the same day OpenAI says its model did. The researchers who first proposed the hypothesis are helping review OpenAI’s claim, and they have stressed that this is not yet a confirmation.
The FTC’s demand for testimony follows 50-page requests for documents sent to both Anthropic and OpenAI earlier this month, covering security incidents, safety practices and marketing, Axios reports. Anthropic’s filing already warned that it may face more government scrutiny than its rivals.
Startup General Compute raised a $100 million Series A to build a cloud for fast AI responses, TechCrunch reports. Instead of relying only on Nvidia, it plans to run models on specialized chips from companies such as Cerebras and SambaNova, with its first large deployments planned for early 2027.
Apple’s latest iPhone software update gives Siri a memory of past conversations, Engadget reports, so the assistant can draw on earlier chats and other personal context when it answers. The update also adds more languages for Siri’s AI features.
Markets
AI stocks led Wall Street higher on Tuesday, according to AP, even though most stocks in the S&P 500 fell. The Nasdaq gained 1.2%, the S&P 500 rose 0.6% and the Dow finished about flat. The Philadelphia semiconductor index climbed 3.6% as chip stocks rose for a third straight session. Of the 86 AI and tech stocks we track, 68 closed higher.
Data center hardware names led the winners. Electronics manufacturer Celestica jumped 12.0% and connectivity chipmaker Astera Labs rose 10.5%, while AI cloud provider Nebius gained 11.2%. Off that theme, AI analytics company BigBear.ai climbed 11.5% and quantum computing company Rigetti added 8.7%.
Software companies took the losses. Consulting firm Accenture fell 4.9%, Adobe dropped 4.0% and C3.ai lost 4.0%. Intuit slipped 3.4% and Salesforce declined 2.7%.
Biggest winners
| Ticker | Name | Price | Day |
|---|---|---|---|
| CLS | Celestica | 276.22 | +11.99% |
| BBAI | BigBear.ai | 2.36 | +11.48% |
| NBIS | Nebius | 170.47 | +11.22% |
| ALAB | Astera Labs | 286.69 | +10.47% |
| RGTI | Rigetti | 12.44 | +8.74% |
Biggest losers
| Ticker | Name | Price | Day |
|---|---|---|---|
| ACN | Accenture | 235.11 | -4.90% |
| AI | C3.ai | 12.28 | -3.99% |
| ADBE | Adobe | 248.36 | -3.97% |
| INTU | Intuit | 328.45 | -3.37% |
| CRM | Salesforce | 240.90 | -2.71% |
Prices are Tuesday’s U.S. closing prices, October 27, 2026, versus Monday’s close. Market data is a snapshot, not live.
Watch
Latent Space — How Anthropic ships the #1 LLM in the World — with Katelyn Lesse and Angela Jiang
Anthropic’s heads of engineering and product, Katelyn Lesse and Angela Jiang, join Latent Space to explain how the company builds and ships Claude.
A big part of the talk covers the design behind Claude Managed Agents, where the part of an agent that thinks is kept separate from the sandbox where it does its work, so one can fail without taking down the other.
Watch this for a rare inside look at how a frontier lab turns a model into products people use every day.
The AI Daily Brief — 4 Questions for AI This Week
Host Nathaniel Whittemore works through four questions hanging over AI this week, starting with whether Chinese models such as Kimi K3.1 are closing the gap with American ones.
He also asks whether cyberattacks driven by AI agents are becoming normal, whether money for data centers is starting to dry up, and what happens if Anthropic’s IPO disappoints.
Watch this for a quick way to get your bearings before a busy week of tech earnings.
Nate B Jones — Are Your AI Agents Doing Real Work Or Productivity Theater? Here Are 4 Principles I Use.
Nate B Jones argues that AI agents can make you look busier without getting the important work done.
He shares four principles he uses to manage agents in Claude and Codex, including giving a chief of staff agent a budget for how much of your attention it is allowed to use.
Watch this if your agents are producing a lot of output and you are not sure how much of it matters.