Daily roundup: AI, networks and cities — October 28, 2026
By Hamilton Ai — October 29, 2026
🤖 AI-generated content. This article was written by Hamilton Ai, the Intelliurb Research Lab artificial intelligence assistant, based on the sources listed at the end, and published under human supervision.
AI
Google DeepMind launches Genie 4, a world model with native audio and real-time editing
Google DeepMind has released Genie 4 in research preview. The successor to Genie 3 generates photorealistic, interactive worlds with synchronised audio, including music, ambient sound and dialogue. Users can change a scene by text mid-simulation, altering weather or camera angle or adding people and objects, and scene consistency now lasts up to two minutes, double its predecessor's limit. Gemini acts as a "director" that plans steps and checks frames; in an experiment tied to Gemini Robotics it broke down tasks that Genie 4 simulated in real time, pointing towards validating robotic scenarios before they reach the physical world. The launch comes as the segment consolidates, with World Labs being acquired by AMD and Amazon in talks to buy Decart. (via TechCrunch)
OpenAI and Anthropic say they have refuted the Exponential Time Hypothesis
OpenAI released solutions to three high-profile problems, among them a refutation of the Exponential Time Hypothesis (ETH), a 1999 conjecture that underpins much of our understanding of what is computationally hard; hours later, Anthropic published its own refutation. This time the release varied by discipline: mathematicians received the work on GitHub, while computer scientists formed a working group organised by MIT professor Ryan Williams together with Russell Impagliazzo and Ramamohan Paturi, who first stated the hypothesis. Williams, who had called OpenAI's earlier release a "DDOS attack on mathematicians", says he saw "two or three alien-looking parts" in the proof and argues for a collective verification effort; the group has issued an open letter on norms for AI contributions to research. If confirmed, the result bears directly on complexity theory and, by extension, on the hardness assumptions that cryptography relies on. (via TechCrunch)
Opera proposes WebAI, a standard for running AI models locally in the browser
Opera has introduced WebAI, a proposed web standard that lets AI models run directly in the browser without relying on the cloud. It supports open-weight models such as Google's Gemma and, on Macs, Apple's on-device model. The feature arrives as an opt-in early preview in Opera One and is due to reach Opera GX and Opera Neon. Opera argues that developers can offer AI features without running infrastructure or paying per request, and that the data processed stays on the user's device; the trade-off is a heavier load on local hardware. An open, model-agnostic standard lowers barriers to entry and could shift part of inference from the large providers to the edge. (via TechCrunch)
Microsoft books $10 billion in revenue from OpenAI in a single quarter
In a regulatory filing released on Wednesday, Microsoft said it earned $10 billion from commercial arrangements with OpenAI, including revenue sharing, in the quarter ended September 30. That is about 11% of its total revenue of $92.3 billion. Over fiscal 2026 this revenue reached $24.1 billion, and receivables from OpenAI rose from $6 billion in June to $7.2 billion in September. The company did not break out how much came from revenue sharing and how much from Azure services. The figures show how deep the financial interdependence between the two companies remains, even after the loosening of exclusivity that allowed OpenAI to contract with other cloud providers. (via TechCrunch)
Networks
O2 and Nokia switch on the UK's first live AI-RAN, with a 20% gain in spectral efficiency
British operator O2 and Nokia have announced the UK's first deployment of a radio access network with built-in AI on a commercial network, at one of O2's "Giga Sites" at London's Paddington station. Radio resources are usually allocated by largely static algorithms; this system instead uses predictive AI to anticipate shifts in demand and automatically readjust spectrum and capacity, and it served real customers throughout the trial. According to O2, spectral efficiency improved by about 20%, so the site could carry more data over the same spectrum. The project is part of the operator's £700m plan, which aims for 1,000 Giga Sites by the end of the year. (via Computer Weekly)
Verizon calls for an architectural rethink of 6G as AI-driven uplink traffic grows
Verizon CTO Yago Tenorio argues that 6G should not be merely a new radio standard, as 5G was. The industry is "still paying for" 5G's many architectural options and inconsistent implementations. AI agents, sensors, robots and wearables could shift the ratio of downlink to uplink traffic from 9:1 to something close to 1:1, which would demand more than incremental improvements. Tenorio wants use cases such as integrated sensing (ISAC) to shape the standard and network autonomy to be built in from launch. He sees the 2028 Los Angeles Olympics as the showcase for these lessons. (via RCR Wireless News)
Sensing, not gigabits, emerges as 6G's first revenue stream
On a panel at RCR Wireless's 6G Forum, Keysight's Stephen Douglas said integrated sensing has probably become 6G's dominant early use case, with a subscription service planned for 2027. Speakers from Ericsson and BT urged operators to identify the problem and who pays for it before setting network requirements. 3GPP's SA1 group has consolidated hundreds of use cases from close to 1,000 contributions, including sensing, digital twins, physical AI and extended reality, but the commercial winners remain unclear. Because the Release 21 functional freeze is set for December 2028, any showcase at the Los Angeles Olympics will by definition be pre-commercial; the panellists propose using 5G-Advanced to prove demand before building the new network. (via RCR Wireless News)
Smart Cities
Incheon announces an AI connected-mobility science park and seeks 900 billion won in state funding
Incheon is home to some 860 auto-parts makers as well as an international airport and a seaport. The city has declared its vision of an "AI Connectivity Innovation City" and announced a connected-mobility science park in Cheongna that will cluster companies, universities and institutes around Cheongna Robot Land. With the Ministry of Trade, Industry and Energy, it will seek up to 900 billion won (about $625 million) to develop next-generation V2X communications and AI-based vehicle platforms. The plan also calls for about 100 billion won a year in R&D and is expected to create 600 to 800 specialist jobs annually. Hyundai Mobis, LG Innotek, HL Klemove, Hyundai Kefico, Lotte Innovate and TMAP Mobility are among ten companies that signed a cooperation memorandum with the Korea Automotive Technology Institute. (via Seoul Economic Daily)
Gimpo Airport trials robots that park cars for drivers
Korea Airports Corporation began a robot valet-parking pilot on October 27 in the basement car park of Gimpo's international terminal. Drivers leave the car in a designated area, a robot lifts it and moves it automatically to a space, and the user heads straight to the departure hall. The service is free during the trial, though standard parking fees still apply. It runs from 10 a.m. to 5 p.m. by advance booking through the website or the Smart Airport app, with KakaoTalk notifications; night hours are excluded while stability is verified. Size and weight limits, up to 5.25 m long and 2,650 kg, rule out some large sedans and SUVs. Automated parking promises better use of space in congested infrastructure without physical expansion. (via The Asia Business Daily)
Sophia Antipolis starts a technology-park digital twin with footfall analysis that identifies no one
France's Communauté d'Agglomération Sophia Antipolis has launched a digital twin with Akila, built on NVIDIA Omniverse. It starts with Pôle Alpha, an innovation building of about 8,500 m², as a first step towards a shared model of the whole park. An AI assistant answers questions from the building's own documents, and building-management data is being connected to monitor technical systems and energy use. CCTV images are processed at the source to extract only aggregated indicators such as density and occupancy rates, with no individual identification or tracking. The announcement coincides with the 43rd IASP World Conference, held in Sophia Antipolis from October 27 to 30, and its privacy-by-design approach offers a replicable model for managing public buildings. (via Akila)