This AI weather startup is out-forecasting government agencies

The golden age of Microsoft’s Github Copilot appears to be at an end — for the little guy, at least. The company is switching its billing system from a flat subscription rate to a token-usage system that has the potential to bill users at a significantly higher rate. Bigger enterprises may still have the juice for it, but smaller companies and workers could find themselves wondering how they’re supposed to balance the monthly budget. The changes, which will take place June 1, mean that users will charged based on how many tokens they burn through as they work instead of a low flat rate based on requests. Some developers with financial whiplash have taken to places like Reddit and X to bemoan what — in many cases — appears to be a drastic escalation in cost. “What a joke,” one Redditor recently wrote, claiming that, while they currently only pay around $29 per month, the new rate will balloon their costs to nearly $750 a month. “This new usage model is just stupidly expensive. I’m adjusting mine by cancelling. At that cost, it is no longer cost-effective or useful in any practical way.” Another user posted “WOW, didn’t expect new pricing model to be this ridiculous,” sharing a screenshot that appeared to show that their costs had shot up from around $50 to some $3,000. The increases sound extreme. However, some Copilot users have bitten back at this criticism — noting that, if you know what you’re doing, you really shouldn’t be blowing through quite so many tokens on a regular basis. The people spending this much are vibe-coders with little actual development knowledge, those critics maintain. “The vast difference between some of us working all day and still barely having overage and then these screenshots. I struggle to believe it’s complexity differences in the workload,” wrote one user. “The only way it gets crazy like that is if you are purely ‘vibe coding’ with a ton of bloated iterations,” they later added. “It’s pretty affordable for even small outfits if used as a tool, on pretty much any provider.” Others have focused on the mind-boggling economics behind the company’s previous model. “Holy fuck how much money was copilot losing,” one Redditor asked in a recent post. It’s a good question. The economics behind Copilot have not always seemed so easy to grasp, and the amount that the company must have spent to subsidize the ongoing vibe-coding escapades of its user base is similarly mysterious and hidden from public view. While some have criticized the changes and others have critiqued those critiques, still other online voices have argued that developers have a perfectly good reason to be upset, given that Microsoft encouraged users to use its chatbot indiscriminately and now appear to be pulling the rug out from under them. “To all the people blaming…the people who actually used the system the way that Microsoft built it (and even encouraged it to be used this way), honestly the only one at fault here is Microsoft. Microsoft provided this billing method and they kept making it easier and easier to burn through massive numbers of tokens on single premium requests that could churn for hours or even days while spawning dozens or even hundreds of sub-agents,” one user wrote. TechCrunch reached out to Microsoft for comment, but did not hear back by publication time. When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence. Lucas is a senior writer at TechCrunch, where he covers artificial intelligence, consumer tech, and startups. He previously covered AI and cybersecurity at Gizmodo. You can contact Lucas by emailing lucas.ropek@techcrunch.com. View Bio

Nvidia founder and CEO Jensen Huang is, perhaps, one of the greatest corporate hype men of all time when it comes to his company. He may even surpass Salesforce’s Marc Benioff when it comes to relentless optimism in his company’s future and revenues. Even so, he delivers on the hype, quarter after quarter. Instead of cautioning you to view the proclamation that he’s found a “brand new $200 billion TAM for Nvidia” with skepticism, I’d argue he’s earned a bit of trust. Huang positioned this massive new market at the feet of Nvidia’s new CPU product, Vera, which was introduced in March. Speaking on Wednesday’s earnings call — after Nvidia posted another record-breaking quarter with $81.6 billion in revenue and forecast $91 billion for the next — Huang pitched Vera as a potentially transformative product. And one that already has promising sales figures. But no matter how well Nvidia delivers, Wall Street harbors anxiety over what will knock Nvidia from its perch. Lately, such fears have centered on the CPU. Nvidia is the king of the GPU, whereas historically the CPU markets were owned by companies like Intel and AMD. (Nvidia has made CPUs previously, of course, but that’s not its core business.) For example, last month Amazon Web Services crowed about a giant contract it signed with Meta for millions of Amazon’s homegrown AI CPUs. Amazon CEO Andy Jassy has been clear that he thinks AWS can do AI chips, both GPUs and CPUs, at least as well, and possibly better than Nvidia. But now, with the Vera CPU, which is sold alone and bundled with its Rubin GPU, Huang believes he’s unlocked “a major new growth driver” for his company because Vera is, he believes, “the world’s first CPU, purpose-built for agentic AI,” Huang said on the call. “Vera opens a brand new $200 billion TAM for Nvidia, a market we have never addressed before, and every major hyperscaler and system maker is partnering with us to deploy it. The world is rebuilding computing for agentic AI and robotic physical AI. Nvidia sits at the center of these transitions,” hype man Huang said. He explained that while the “thinking” part of an AI model uses GPUs, agents mostly run on CPUs. They use CPUs to do their assigned tasks and will, he predicts, run their own form of CPU-driven PCs. Vera is for agents because it’s specifically designed to process tokens as fast as possible. This is opposed to classic cloud architecture CPUs designed with “cores,” or the ability to run multiple instances of apps as fast as possible. That sounds logical, but with the major cloud providers as well as startups pursuing AI chip development, what makes him think that Nvidia will be the go-to source for agentic CPUs? Because, Huang says, Nvidia has already sold $20 billion worth of standalone Vera CPUs this year and we’re only at the beginning. “The world has a billion users, human users. My sense is that the world is going to have billions of agents, not today. I mean, we’re going to grow into it, but we’ll have billions of agents, and those billions of agents will all use tools. And those tools are going to be like PCs, just like us humans using using PCs today,” he said. “We’re going to need a lot more CPUs,” he explained. When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Stability AI, the company behind Stable Diffusion, is releasing a new family of audio models, called Stability Audio 3.0. The top model can generate professional-grade music of more than six minutes long, the company claimed. The company is releasing four new models under the Stable Audio 3.0 name: small SFX (459M parameters), small (459M parameters), medium (1.4B parameters), and large (2.7B parameters). The duo of small models is suitable for on-device sound and music generation of up to two minutes. Both medium and large models can create full compositions of 6 minutes 20 seconds long that can maintain musical structure and melodic tone. This is more than double the length of what Stable Audio 2.0, released in 2024, was capable of generating. Stability AI is making small SFX, small, and medium models available with open weights for anyone to use and modify. In 2024, the company released Stable Audio Open, which allowed for music generation of up to 47 seconds. The new family of models is a big step up from the previous open versions. Image Credits: Stability AIImage Credits:Stability AI The large model is available only through the API and self-hosting paid services. Plus, companies with more than $1 million in revenue would need to get an enterprise license. Many companies, including Google and ElevenLabs, are releasing models and tooling around music generation. However, as Suno and Udio’s ongoing court battles have proved, licensing of data and partnerships with music labels could become a key part of the long-term survival of these services. Last year, Stability AI inked deals with Warner Music Group and Universal Music Group to develop models and music creation tools. The company said that its latest set of audio models is built on fully licensed data. The AI startup is developing a new suite of products for professional musicians, but didn’t give more details on its features. Ethan Kaplan, former chief digital officer at Universal Audio and Fender, is joining the company to lead Stability’s professional music offering. A number of AI companies are trying to bolster their credentials by hiring music execs. Earlier this year, Suno hired former Merlin CEO Jeremy Sirota as chief commercial officer. ElevenLabs has also hired Derek Cournoyer from indie music publisher Kobalt as a strategy lead for its music business. When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence. Ivan covers global consumer tech developments at TechCrunch. He is based out of India and has previously worked at publications including Huffington Post and The Next Web. You can contact or verify outreach from Ivan by emailing im@ivanmehta.com or via encrypted message at ivan.42 on Signal. View Bio

We’ve all pulled up Street View on Google Maps to show a friend what our childhood home looked like, or dropped that little person icon onto the streets of Paris to see if we booked a hotel in a cool neighborhood. Imagine being able to do that, but in a more immersive, interactive way that allows you to really simulate the street and its environs, and even do things like adjust the weather or see what it would look like in a “Day After Tomorrow” scenario. That’s one of the goals of Google’s latest integration. Starting today, Google DeepMind is connecting Street View to Project Genie, the company’s general-purpose world model that can generate diverse, interactive environments. The new feature launched during the Google I/O developer conference. “It’s really powerful for both the agent [and robotics] use case and for humans to play with, and that’s always been the thesis of Genie,” Jack Parker-Holder, a research scientist on DeepMind’s open-endedness team, told TechCrunch. He gave the example of a new robot being deployed in London, which rarely sees the sun. Genie could, Parker-Holder says, simulate those scarce occasions when the sun glints off the Victorian housing, so the rays don’t shock the robot when it happens. Loading the player… “Simultaneously, you might say, ‘I’m going to New York City, but not this time of year,’” he continued. “‘It’s going to be snowy. I want to see what that block looks like in the snow.’” Google has been collecting Street View data for 20 years via cars with cameras and individuals strapped with “tracker backpacks.” The tech giant has collected north of 280 billion images across 110 countries and seven continents. “With Street View, we have imagery from a large quantity of the world,” Jack said. “You can imagine how potentially powerful it is to combine this rich source of real-world information and data with an ability to simulate worlds.” Google released its latest world model Genie 3 for research preview last August and opened up access to the tool to Google AI Ultra subscribers in the U.S. in January, allowing customers to create interactive game worlds from text prompts or images. The goal is to use Genie for educational experiences, gaming, and robotics training. Genie 3 is already helping to power one of Waymo’s simulators to train its self-driving cars on “exceedingly rare events” like tornadoes or casual elephant encounters. Adding Street View data to that could help Waymo prepare to launch in more cities around the globe. Waymo has its own simulator that it relied on to scale to 11 U.S. cities and test its AI driver in several more. The difference with Genie, says Parker-Holder, is that those are all from the car’s point of view. Street View allows for not only simulating a world anchored to a real place, but also shifting the point of view to other types of agents, like a human or a robot. Google is launching Street View in Genie to some Ultra users in the United States starting today, with access rolling out at scale over time. Global Ultra users will gain access over the next few weeks, per the company. The researchers’ goal is to put this new capability into as many hands as possible, per Diego Rivas, a product manager at DeepMind. He cautioned that Street View in particular and Genie in general is still an experiment, so there’s much to improve upon in terms of accuracy. In the samples the Google team showed me — including an underwater simulation of a neighborhood I used to live in — the results are impressive and recognizable, but still video game quality rather than photorealistic. The models are also not yet physics-aware, meaning they don’t yet understand cause and effect. For example, in a simulation of a woman running through a snowy Joshua Tree, she ran right through cacti and bushes. Compare that to, say, Google’s image generator Nano Banana — which can now generate perfect text in infographics — or its video generator Veo — which understands that paper boa

Did Chrome Just Install a Massive AI Model on Your Device Without Telling You? Yes, Probably CNET How to Prevent Chrome Browser from Downloading 4GB of Local AI Model Files OS X Daily Chrome downloads 4GB Gemini Nano model silently Let's Data Science Stop Chrome Browser From Downloading a Hidden 4GB AI File MacRumors Google Chrome Automatically Downloads a Storage-Eating 4GB Model to Support AI Features. Here’s How to Turn it Off. Mix93.3

Richard Socher has been a major figure in AI for some time, best known for founding the early chatbot startup You.com and, before that, his work on Imagenet. Now, he’s joining the current generation of research-focused AI startups with Recursive Superintelligence, a San Francisco-based startup that came out of stealth on Wednesday with $650 million in funding. Socher is joined in the new venture by a cohort of prominent AI researchers, including Peter Norvig and Cresta co-founder Tim Shi. Together, they’re working to create a recursively self-improving AI model, one that can autonomously identify its own weaknesses and redesign itself to fix them, without human involvement — a long-held holy grail of contemporary AI research. I spoke with him on Zoom after the launch, digging into Recursive’s unique technical approach and why he doesn’t think of this new project as a neolab, he informal term for a new generation of AI startups that prioritize research over building products. This interview has been edited for length and clarity. We hear a lot about recursion these days! It feels like a very common goal across different labs. What do you see as your unique approach? Our unique approach is to use open-endedness to get to recursive self-improvement, which no one has yet achieved. It’s an elusive goal for a lot of people. A lot of people already assume it happens when you just do auto-research. You know, you can take AI and ask it to make some other thing better, which could be a machine learning system, or just a letter that you write, or, you know, whatever it might be, right? But that’s not recursive self-improvement. That’s just improvement. Our main focus, is to build truly recursive, self-improving superintelligence at scale, which means that the entire process of ideation, implementation and validation of research ideas would be automatic. First [it would automate] AI research ideas, eventually any kind of research ideas, even eventually in the physical domains. But it's particularly powerful when it's AI working on itself, and it's developing a new kind of sense of self awareness of its own shortcomings. You used the term open-ended — does that have a specific technical meaning? It does. In fact, Tim Rocktäschel, one of our cofounders, led the open-endedness and self-improvement teams at Google DeepMind and particularly worked on the world model Genie 3, which is a great example of open-endedness. You can tell it any concept, any world, any agent, and it just creates it, and it's interactive. In biological evolution, animals adapt to the environment, and then others counter-adapt to those adaptations. It's just a process that can evolve for billions of years, and interesting stuff keeps happening, right? That's how we developed eyes in our [heads]. Another example is rainbow teaming, from another paper from Tim. Have you heard of red teaming? In cybersecurity, it means-- So, red teaming also has to be done in an LLM context. Basically you try to get the LLM to tell you how to build a bomb, and you want to make sure that it doesn’t do it. Now, humans can sit there for a long time and come up with interesting examples of what the AI shouldn't say. But what if you tested this first AI with a second AI, and that second AI now has the task of making the first AI [try to] say all the possible bad things. And then they can go back and forth for millions of iterations. You can actually allow two AIs to co-evolve. One keeps attacking the other, and then comes up with not just one angle but many different angles, and hence the rainbow analogy. And then you can inoculate the first AI, and you become safer and safer. This was an idea from Tim Rocktaeschel, and it’s now used in all the major labs. How do you know when it’s done? I suppose it’s never done. Some of these things will never be done. You can always get more intelligent. You can always get better at programming and math and so on. There are some bounds on in

Neil Batlivala has spent seven years building a healthcare company that most of the tech industry has never heard of and that serves a patient population most of Silicon Valley ignores. But last month, that work put him at the center of something much bigger. His company, Pair Team, announced on April 30 it had been accepted into ACCESS, a Medicare program — as one of 150 participants chosen by the Centers for Medicare & Medicaid Services to test what AI-driven medical care could look like at federal scale. The program goes live July 5. “The government is creating swim lanes for AI innovation in traditionally regulated industries,” he told me over a Zoom call a few days later. “The best solution wins, which, in regulated industries like healthcare — that’s not been the case.” ACCESS — Advancing Chronic Care with Effective, Scalable Solutions — is a 10-year CMS program testing a payment model that rewards health outcomes rather than required activities (like a certain number of check-ins). Participating organizations like Pair Team receive predictable payments for managing qualifying conditions and earn the full amount only when patients meet measurable health goals, like lower blood pressure or reduced pain. It covers diabetes, hypertension, chronic kidney disease, obesity, depression, and anxiety. That payment structure is the real news. Traditional Medicare reimburses based on time spent with a clinician. There’s no mechanism to pay for an AI agent that monitors a patient between visits, calls to check in, coordinates a housing referral, or makes sure someone picks up their medication. ACCESS creates that mechanism for the first time. “It’s a payment model transformation,” Batlivala said. “You just couldn’t do this before.” The first cohort spans a wide range of participants — AI doctor startups, virtual nutrition therapy providers, connected device companies, and wearable makers like Whoop. Batlivala is skeptical of some of them. "I'm a big fan of wearables, but for a senior who's struggling with food insecurity, I don't know how much Whoop is going to be able to do," he said, adding of his own company, "We've been building toward this for five-plus years now." Pair Team launched in 2019 with a specific kind of patient in mind: people managing chronic conditions who were also dealing with unstable housing, too little food, or lack of transportation. About a third of Americans fall somewhere in that category. The company's premise was that you can't improve health outcomes without addressing the full context of someone's life. It now employs roughly 850 clinical professionals, runs what it describes as the largest community health workforce in California, and, per Batlivala, generates revenue above nine figures. It has raised about $30 million, backed by Kleiner Perkins, Kraft Ventures, and Next Ventures. The model has peer-reviewed evidence behind it. A study, co-authored by Pair Team researchers and peer-reviewed by the Journal of General Internal Medicine, evaluated Pair Team's community-integrated model, which blends medical, behavioral, and social care for Medicaid members with high rates of homelessness, serious mental illness, and chronic disease and it showed strong patient engagement and significant reductions in avoidable emergency and inpatient utilization. Batlivala says one in four hospital visits and one in two ER visits don't happen when a patient is in his company's care. But for years, delivering that level of care required human teams, which limited how fast and cheaply it could scale. Then, about nine months ago, Pair Team deployed a voice AI agent called Flora as its primary patient-facing interface. Flora is available 24 hours a day, handles intake, coordinates referrals, and does the check-ins that keep patients engaged between clinical visits. The first call that shifted his thinking was with a 67-year-old woman living out of her car, managing PTSD and congestive heart failure. She spoke w