plus 9 more stories
Also: OpenAI says an internal model disproved the Exponential Time Hypothesis; and Google DeepMind releases Genie 4, a world model with sound and real-time steering. Every story links to its primary source.
Curated by Gaurav · October 27, 2026
Physical Intelligence is in talks to raise at least $1 billion, which would nearly double its valuation to about $20 billion, Bloomberg reported, citing people familiar with the matter. Existing backers Avenir and Lux Capital are in talks to lead. The robotics software startup was valued at $11.5 billion earlier this year and has no current revenue. The round is not closed.
OpenAI said Monday that an internal model disproved the Exponential Time Hypothesis, a widely used assumption about hard logic problems, and checked the argument in Lean on October 17. The same post claims proofs tied to the Kakeya conjecture and to quantum chromodynamics. Impagliazzo, Paturi, and Williams are reviewing the computer-science claim and said their role is not a certification. OpenAI said the result does not settle P versus NP, and the physics writeup is not public yet.
Google DeepMind released Genie 4, a world model that generates interactive scenes with native sound, including speech, and lets a user change the scene with text while it runs. Gemini 3.8 Flash watches the frames and breaks a goal into steps. It is a research preview for trusted testers, with a waitlist for Google AI Ultra subscribers in Project Genie and early Gemini API partners. DeepMind said long sessions can still drift, and every clip carries a SynthID watermark.
Dallas firm Disruptive closed a $10 billion fund to back at least 10 later-stage companies in AI and defense, with checks of up to $1 billion. It is the firm's first blind pool after years of raising money deal by deal. The firm told investors it has returned more than $6 billion on roughly $2 billion put into startups, Bloomberg reported. The Dallas Business Journal called it the largest venture fund raised in Texas.
Guidewire agreed to buy London underwriting firm hyperexponential for $1.7 billion in cash, plus up to $400 million if recurring-revenue targets are met. The target says its software has helped price more than $75 billion a year in commercial property-and-casualty premium. Guidewire will fund the deal with $1 billion in new loans from Bank of America and Citibank, plus cash on hand. It expects to close in the second quarter of its fiscal 2027.
Anthropic said an unreleased internal model built working exploit chains against Chrome, Safari, Android, Windows, and Linux, then reported them to the vendors. A Chrome proof of concept opened the Calculator app after one click, and Google shipped 62 critical or high-severity fixes in Chrome 156. Apple shipped a Safari fix, and Google addressed an Android lock-screen bypass. A Linux networking bug that could run commands as root was patched upstream.
President Donald Trump said he will ban the term artificial intelligence in government documents and replace it with super intelligence. The White House said the wording matches a push to stress American leadership, after a September order already told agencies to switch. He can require that change inside the executive branch. Major newsrooms still use the older name, and the Associated Press Stylebook still lists it.
General Compute raised a $100 million Series A led by Blackbird Ventures and Abstract Ventures at an $800 million valuation, up from $60 million after a May seed. The company rents bare-metal servers that use SambaNova, Positron, and Cerebras chips. A customer contract worth $100 million a year in recurring revenue opened the round. Chief executive Finn Puklowski said he expects to deploy 75 megawatts in 2027.
Liquid AI released LFM3, six open-weight models around 2 billion parameters for text, images, audio, combined audio and video, and fast yes-or-no decisions. The text model has a 256,000-token context and was trained on 22 trillion tokens. On agent tests it beat some larger models at following instructions and calling tools, while bigger models still lead on coding. Weights are available now, and the company says the models run on phones and laptops.
Amazon published a September memo telling software engineers not to treat token use, tool adoption, or lines of code as proof they are using AI well. The company called those measures deeply flawed because they raise cost and complexity, and said teams should be judged on long-term outcomes. In May it shut an internal leaderboard after employees inflated token counts to climb the ranks. The memo says skipping AI where it does not help is a finding, not a failing.
Track the tools behind the news:
→ AI Tool Leaderboard → Browse the directory