AI This Week
Google made a wave of major AI announcements at its annual Google I/O developer conference on Tuesday. The search giant rolled out its latest version of Gemini and a new artificial intelligence model designed to simulate the physical world, racing to keep pace in model development while providing more agentic services to its massive user base. Gemini 3.5 Flash is a general-purpose frontier model that surpasses Gemini 3.1 Pro on coding, agentic, and multimodal benchmarks, running at four times the output speed of comparable frontier models. On the agentic AI front, Google announced Gemini Spark, a new general-purpose AI agent that can reason across information in connected apps. Gemini Spark is in beta and will be available first to trusted testers and Google AI Ultra subscribers starting next week. The new AI video model Gemini Omni combines Gemini with Google's generative media models to create outputs grounded in knowledge. Google AI Ultra now starts at $100 per month, down from $250. The Gemini app now has 900 million monthly active users, roughly doubling from 400 million a year ago.
Amazon has thrown down a major gauntlet in the podcast world. Alexa Plus, Amazon's upgraded AI assistant, can now generate podcasts on "virtually any topic." Users give Alexa Plus a topic, and it offers an overview of what its AI hosts plan to talk about, allowing listeners to steer the conversation and adjust its length before generation begins. In one example clip, two AI-generated hosts discuss the latest music releases. Users can set preferred length and conversational style, and the assistant produces an episode with two AI-generated host voices within minutes. Amazon says Alexa+ can access real-time information through agreements with outlets including the Associated Press, Reuters, The Washington Post, Time, Forbes, Business Insider, Politico, USA Today, Condé Nast, Hearst, and Vox Media, alongside more than 200 local newspapers. It's not the first AI-generated podcast system — Google's NotebookLM added similar capability last year, which prompted a lawsuit from a former NPR host who alleged Google copied his voice without permission. Alexa+ is bundled at no additional cost for Prime members; non-Prime members pay $19.99/month.
A federal jury in Oakland, California, ruled that Elon Musk waited too long to sue OpenAI and its CEO Sam Altman over claims they violated an agreement to run their artificial intelligence venture strictly as a charitable nonprofit. The advisory jury's verdict came after less than two hours of deliberations and was immediately adopted by District Court Judge Yvonne Gonzalez Rogers. Musk, who co-founded OpenAI as a nonprofit, had filed a $150 billion lawsuit against the organization, Altman, and president Greg Brockman, accusing them of turning it into a for-profit entity for personal enrichment. Altman, Brockman, and OpenAI were found not liable on all claims, and the jury also rejected Musk's case against Microsoft on the same statute-of-limitations grounds. OpenAI's lawyers portrayed the lawsuit as Musk's attempt to kneecap a rival after he failed to gain control of it. During weeks of testimony, jurors heard from Altman, Brockman, Microsoft CEO Satya Nadella, and Musk himself. Musk and his attorneys said they would appeal the verdict to the 9th Circuit U.S. Court of Appeals. The verdict comes at a critical time as both tech magnates push their respective companies toward the public markets in what are expected to be record offerings.
Cerebras Systems' shares sank 10% on Friday after the company completed the largest IPO by a U.S. tech firm in years. The pullback followed a breathtaking first day. The semiconductor firm initially sold shares at $185 and closed Thursday at $331.07 — a 68% surge — giving it a market cap of about $95 billion. The firm raised $5.55 billion selling 30 million shares, the largest IPO for a tech firm since Uber's debut in 2019. The IPO debut made the company's top executives billionaires, with CEO Andrew Feldman and CTO Sean Lie owning stakes worth $3.2 billion and $1.7 billion, respectively. Cerebras sells extremely large computer chips and AI systems designed to train and run AI models faster than traditional GPUs. The company has established partnerships with Amazon and OpenAI — with OpenAI launching its first AI model running on Cerebras chips earlier this year.
Martha Stewart, home services veteran Yih-Han Ma, and AI thought leader Kyle Rush have announced the launch of Hint, a new company co-founded by Stewart that reimagines homeownership through the combined power of human expertise and artificial intelligence. Hint launches with a $10 million seed round led by Slow Ventures, with participation from Montauk Capital, Tusk Venture Partners, Amplo, Energy Impact Partners, Hannah Grey VC, and Brian Kelly, founder of The Points Guy. Hint describes itself as an always-on, AI-native home management platform designed to help homeowners proactively manage their homes, combining expert guidance with home intelligence data to deliver personalized recommendations, cost-saving insights, maintenance planning, and real-time monitoring. Users will enter their home address into the platform so the tool can source data about everything from soil quality to flood risk, then combine that with home-care know-how drawn from the expertise of contractors, designers, and Stewart herself. Americans spend more than $500 billion each year on home repairs and renovations. Hint will be available on desktop and for download on iOS this summer 2026.
Alphabet's Google has introduced a new high-end laptop segment called "Googlebook" that will run Android and prominently showcase Gemini artificial intelligence. The unveiling comes 15 years after Google introduced the Chromebook — the affordable, browser-based laptop that became a fixture in schools and workplaces worldwide. As Google moves from an "operating system to an intelligence system," it sees an opportunity to rethink laptops again. The tech giant is working with partners like Acer, Asus, Dell, HP, and Lenovo to make the first Googlebooks in a variety of shapes and sizes. Built with the Google DeepMind team, Magic Pointer brings Gemini directly to the cursor with contextual suggestions every time you point at something on screen. Googlebooks will also work seamlessly with Android phones; Quick Access lets users browse and insert phone files directly into the laptop's file browser, with no transfer needed, and Android phone apps run directly on the laptop. The laptops are expected to arrive in fall 2026.
OpenAI is standing up a new company backed by more than $4 billion in initial capital to support organizations in adopting and scaling artificial intelligence, with the acquisition of AI consulting firm Tomoro set to provide immediate staffing for the venture. The company enters with a $10 billion valuation, backed by 19 firms including TPG, Goldman Sachs, SoftBank, Capgemini, and McKinsey & Company. To staff up immediately, OpenAI agreed to acquire Tomoro, an applied AI consultancy with roughly 150 engineers who have already worked for clients like Tesco, Virgin Atlantic, and Supercell. Days before OpenAI's announcement, Anthropic revealed its own enterprise deployment venture — a $1.5 billion entity backed by Blackstone, Hellman & Friedman, and Goldman Sachs — with essentially the same premise. Enterprise already accounts for more than 40% of OpenAI's revenue, with the company reporting $25 billion in annualized revenue as of February.
Google DeepMind has published a paper titled "AI Co-Mathematician: Accelerating Mathematicians with Agentic AI," describing a system that goes far beyond a standard chatbot — giving researchers a stateful workspace where multiple AI agents run parallel workstreams, track uncertainty, preserve failed attempts, search literature, test ideas, and produce mathematical working documents. The system runs on Gemini 3.1 and is organized hierarchically: a project coordinator at the top, workstream coordinators below managing literature review, library development, and counterexample search, with specialized agents — including a search agent, a coding agent, and Gemini Deep Think acting as a proof verifier. The system reports a 48 percent score on FrontierMath Tier 4 — a new high score among all AI systems evaluated. Oxford mathematician Marc Lackenby used it to resolve an open problem from the Kourovka Notebook after a reviewer agent spotted a flaw in the AI's first proof attempt. Authors are candid about failure modes: the review cycle between agents can converge on subtly flawed arguments — a "reviewer-pleasing bias" — or spiral into endless disagreement between agents.
After months of Public Preview testing, Google is replacing the Fitbit app and launching "Google Health," bringing together "the best of Fitbit's pioneering spirit with the helpfulness of Google." It's a seismic shift for millions of fitness tracker users. The Fitbit brand will remain in use for hardware, but all software moves under the "Google Health" umbrella. The update — including on Pixel Watch — starts rolling out May 19 and will be fully available ahead of the Fitbit Air hitting shelves on May 26. New with this launch is multimodal logging — log via text, voice, or photo, including snapping a picture of lunch for automatic food recognition. Google Health Premium subscribers get the conversational Google Health Coach and adaptive fitness plans, plus detailed and proactive insights across sleep, fitness, and health. Google Health Premium is now bundled with Google AI Pro or AI Ultra in 30+ countries, or available separately for $9.99/month or $99.99/year. Later this year, Google Fit users will be able to migrate their data to Google Health.
Anthropic announced a deal with Elon Musk's SpaceX to use all of the compute capacity at the Colossus 1 data center in Memphis, Tennessee. That amounts to more than 300 megawatts of new AI compute via more than 220,000 Nvidia GPUs. The partnership is stunning given the public feud between the two companies. Musk wrote in February that Anthropic "hates Western Civilization." But on Wednesday he reversed course. He said he spent a lot of time with senior Anthropic team members and was "impressed," writing: "Everyone I met was highly competent and cared a great deal about doing the right thing. No one set off my evil detector." Anthropic said the deal will directly improve capacity for its paid Claude Pro and Claude Max subscribers. As part of the agreement, Anthropic also expressed interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity. Last month, Anthropic said demand for Claude has led to "inevitable strain on our infrastructure."