Actualités — Page7

Barcelona startup Kabilio has raised €10 million in a round led by Picus Capital and Visionaries, with Spain’s public technology agency CDTI also taking part. The company plans to use the money to build an AI agent that accounting firms can hand whole tasks, such as gathering invoices, reconciling accounts and chasing missing items. The round brings Kabilio’s total funding to €14 million. Since its last round a year ago, the company has grown from about 100 accounting firms to nearly 500 and expanded its team to 80 people. The agent enters beta this month, and several firms are already testing it. Accountants keep control over how much the agent can do on its own. Lire la suite

Open-source AI models are moving into the mainstream as enterprises, governments and startups turn to them for lower costs, greater control and more flexibility. As capable models become more freely available, companies face a harder question: how do they capture value? In a new episode of TechCrunch's Equity podcast, Rebecca Bellan spoke with Irene Solaiman, chief policy officer and SVP of strategy at Hugging Face, about what startups should know before building on open models. The discussion also takes up the rise of Chinese open models and whether American players can keep pace. Lire la suite

Google faced swift criticism Tuesday after announcing that schools using its K-12 software will lose access to Google Search unless they opt in to the company's AI tools. Chris Phillips, Google's vice president for education, said the two products are "deeply integrated," and students at schools that don't opt in would see a message saying Search is unavailable. Senators Bernie Sanders and Richard Blumenthal, former Federal Trade Commission chair Lina Khan and New York City Mayor Zohran Mamdani all condemned the move. Google responded by turning off access to AI products by default, while letting schools turn on Gemini, Gemini Notebook and Search individually. Lire la suite

Third-party AI agents arranged just four Uber rides in the third quarter of 2026, company executives said on an earnings call. Agents fared slightly better on Uber Eats, arranging around 500 orders during the week of Halloween. Uber disclosed the figures as concern grows that AI assistants could pull customers away from delivery and ride-hailing apps. CEO Dara Khosrowshahi said the company is open to working with third-party AI companies but expects customers to keep preferring the Uber app itself. Lire la suite

Anthropic has released FlashProve, an open-source research prototype that uses cryptographic proofs to show a response came from a specific AI model. The system can generate proofs for Llama 4 Maverick, a 400-billion-parameter model, at about 110 tokens per second on a single H100 GPU. The company says the work could help users, evaluators and regulators confirm which model is actually running, without access to the model's weights. The code, formal proofs and technical documentation are available under the Apache 2.0 license. Lire la suite

Shares of several artificial intelligence-linked companies fell Wednesday after midterm results showed Democrats winning major races, including the Texas governor's race. Democrat Gina Hinojosa defeated incumbent Republican Gov. Greg Abbott in a state where analysts say about a quarter of planned U.S. data center capacity is located. Hinojosa campaigned for an immediate moratorium on data center development in Texas, and investors are now weighing whether state-level restrictions could delay major projects. Shares of CoreWeave and Iren fell about 4% in morning trading, while Alphabet, Amazon, Meta and Microsoft rose. Lire la suite

Qualcomm says it has finished testing the first chips built with its High Bandwidth Compute memory technology, a step toward using the design in its Dragonfly AI250 data center accelerator. The company said the results confirmed its expected gains in bandwidth and energy efficiency, though it cautioned that comparisons with competing products are estimates and that real-world performance will depend on the final product and workload. Qualcomm expects the AI250 to begin sampling in the first half of 2027, and it is pitching the hardware to cloud providers, hyperscalers and large enterprises. Lire la suite

Saudi Aramco reported adjusted net income of $31.2 billion for the third quarter, up from $28.6 billion a year earlier, as the state oil company pushed further into artificial intelligence. The results, released Tuesday, highlight Aramco’s new 49% stake in HUMAIN, the kingdom’s state-backed AI company. Aramco is paying with key AI assets and about SAR 9.9 billion ($2.6 billion) in cash. The Public Investment Fund, Saudi Arabia’s sovereign wealth fund, keeps the other 51%. Aramco says it plans to use HUMAIN’s infrastructure to bring AI into its industrial operations. Lire la suite

A Reddit user has released an open-source project called TVPort that uses AI coding tools to recompile Nintendo Switch games so they can run on the Apple TV 4K. The project, a fork of the nxnative toolkit, works from a Mac and lets users pick between 30 and 60 frames per second. Some games run well, including Super Mario Odyssey, while heavier titles such as The Legend of Zelda: Tears of the Kingdom still need significant work. The project's creator says some games may never run on the older Apple TV 4K hardware. The effort is part of a wave of projects that use AI tools to clone or port software, a practice that remains controversial in the decompilation community. Lire la suite

Alex Stamos, Meta's former chief security officer, says handing an AI agent the passwords to your financial life is “totally insane.” Speaking on Alex Kantrowitz’s Big Technology Podcast, Stamos said people are giving agents such as Meta’s Muse and Spear Street Technology’s Instinct full access to their password vaults and asking them to pay bills or file taxes. He said banks now face a growing wave of bot activity that customers permit, and that the arrangement creates a tension between banks trying to block automated access and customers trusting outside companies with their most sensitive logins. JPMorgan Chase CEO Jamie Dimon has raised similar concerns. Lire la suite

Kino, a Y Combinator-backed startup, has launched a video editor that stores footage, edits and team comments as code, letting human editors and AI agents work on the same timeline at once. Founder Luke announced the product on Hacker News, saying the company rebuilt its earlier enterprise tool into a self-serve editor in 2025. The editor is free to use, with a $25-per-editor monthly Pro tier that adds more storage, share links and expanded agent access. A desktop app is available now, and a browser version is not yet released. Lire la suite

Apple on Tuesday began testing a set of AI-powered audio features in the first developer betas of iOS 27.3 and watchOS 27.3. Two of the features, Live Rewind and Siri Recaps, record and summarize conversations picked up around the Apple Watch's wearer, raising privacy questions even as early testing shows practical value. Apple says it processes audio on-device when possible, does not identify speakers or store raw audio, and redacts sensitive information. A TechCrunch reviewer who tested Siri Recaps for a day found it captured useful details from a car repair visit, but also logged personal household chatter and non-essential remarks. Lire la suite

Supermicro will build and sell AI inference systems that combine SambaNova’s new SN50 chips with Nvidia GPUs, the companies announced Tuesday. The collaboration initially plans for 10,000 SN50 chips. The racks will be sold through Supermicro’s data center building block line, with Nvidia GPUs handling the first stage of running an AI model and SN50 chips handling the second. SambaNova says the design lets operators add inference capacity in data centers they already run, rather than waiting for new builds held back by power and cooling. Lire la suite

A user on Hacker News has launched an "Ask HN" thread asking the community how they use artificial intelligence day to day. The post lays out five questions covering paid AI subscriptions, models accessed through APIs or local hosting, coding tools and interfaces, agent orchestration software, and broader agent platforms. The poster says their own workflow has changed significantly over the past year and wants a current snapshot of what people are using and why. The post does not say how it will be used, and the excerpt does not include any responses. Lire la suite

Paying for a personal Google AI subscription does not keep your Gemini chats out of model training, according to a Hacker News post that quotes Google's own help pages. Saved chat history can be used to train generative AI models, and a subset of chats may be read by human reviewers. The only way to stop the training, the post says, is to turn off the Keep Activity setting, which means chat sessions are deleted after 72 hours. The claims come from a user's post and have not been independently verified. Lire la suite

Nothing CEO Carl Pei says the company will launch its first AI device in December, and the product will not include a screen. Pei announced in a post on X that the device is a tool built with the personality and care associated with Nothing products. He gave few other details. The launch comes just over a year after Pei first announced plans for “AI-native” devices and an AI operating system, and it joins a growing list of screen-free AI gadgets from other companies. Lire la suite

OpenAI has added GIF creation to ChatGPT, letting users generate an animated GIF from a text prompt or an uploaded photo. The feature arrives with the company's new Images 2.6 model, which also speeds up image generation. Users can adjust a GIF's animation speed, edit individual frames, and share the finished file. Before this update, ChatGPT could not create GIFs natively in a single step. The Images 2.6 update is rolling out to all ChatGPT and Codex users on web, desktop, and mobile, and OpenAI says users now generate more than 4 billion images per week across ChatGPT and its API. Lire la suite

The UK government has repeatedly promised that artificial intelligence could add £400 billion ($538 billion) to the British economy. The AI minister now concedes that the estimate is less than a third of that figure. Researchers blame the country's slow pace of AI adoption. The original number, according to reporting on the matter, mainly came from Google. The revised figure marks a sharp retreat from a claim the government used for two years to ease public fears about job losses and economic collapse. Lire la suite

Drax, which operates Britain's largest power station in North Yorkshire, plans to build an artificial intelligence data center on the site with Brookfield and its UK-based AI infrastructure business, Radiant. The project could eventually host more than 1 gigawatt of computing capacity and carry a total investment of up to £20 billion ($26.8 billion), Brookfield estimates. The plan depends on government consent for later phases, planning permission and grid approvals, and a tenant that has not yet been named. Drax said the announcement was made in a stock market statement on Tuesday. Lire la suite

A software engineer who moved from Microsoft's Windows AI team to Salesforce AI Research has published his annual review of how he uses artificial intelligence, and the picture has shifted sharply in a year. He now manages five or six coding agents at once, which he says write more than 90 percent of his code, and he spends about $3,000 a month on tokens. He reports that AI has become his default for search and for explaining well-known concepts. But he says it still fails at product recommendations, original research, and writing, and he has stopped using image generators, which he describes as a net negative. He also has drafted a personal AI policy to guide his use. His verdict is mixed: the tools are transforming his engineering work, but he remains skeptical of them almost everywhere else. Lire la suite