Tech Brew Ride Home - Let's Regulate This AI Stuff?

Episode Date: July 14, 2026

Demis Hassabis proposed a US-based frontier AI standards body modeled on FINRA. IBM's stock cratered 20% on a Q2 miss from chip-spending shifts, Spotify launched a voice-control feature, Kalshi debuted an AI compute forward curve, and Anthropic studied Claude's values. Demis Hassabis proposes a US-based Standards Body for "Frontier-class" AI, modeled after FINRA; labs would share models for review up to 30 days before release (X) Demis Hassabis proposes a US-based Standards Body for "Frontier-class" AI, modeled after FINRA; labs would share models for review up to 30 days before release (The Verge) IBM reports preliminary Q2 revenue up 1% YoY to $17.2B, below $17.9B est., as CEO Arvind Krishna says customers are shifting spending to chips; IBM falls 20%+ (Bloomberg) Spotify launches a Talk to Spotify feature that lets users create playlists and more, rolling out in beta to Premium users 18+ in the US, Ireland, and Sweden (Engadget) Kalshi launches a forward curve tool for AI compute, using event contracts to track the future rental costs of GPUs, storage, and memory (Bloomberg) Simulating everything, sort of: The promise and limits of world models (Ars Technica) Subscribe to the ad-free feed. Learn more about your ad choices. Visit megaphone.fm/adchoices

Welcome to the Tech Rewrite home for Tuesday, July 14th, 2026. I'm Brian McCullough today. Demis Hasabas has proposed a U.S.-based frontier AI standards body. IBM's stock cratered 20% on a Q2 miss from chip spending shifts. Spotify launched a voice control feature. Kalshi debuted in AI compute forward curve and what are world models? Here's what you miss today in the world of tech. To all the IT pros out there, today's sponsor has a special gift for you. That gift is Ace of Uptime, an online card-based game where you go up against the problems that threaten uptime on a daily basis. Each level pits you against a recognizable adversary. We're talking about alert overloads, heat that's trapped in a cramped server room, and systems that look fine but are far from it.

Choose your move and see whether your decisions resolve the issue or escalate it. The game is fast, fun, and designed to let you test how you'd respond when uptime is on the line. Think you can beat the villains of downtime? Head over to Eaton, com slash ace to find out that's eaton.com slash ace. Demis Hasabas is proposing a U.S.-based standards body for frontier class AI models modeled after FINRA. AI labs would share models for review up to 30 days before release under this scheme, quoting the verge.

Demis Hasabas thinks the world needs an AI watchdog with the power to hit the brakes if frontier models become too dangerous. Writing at a blog post, the Google DeepMind CEO and co-Fo founder said the U.S. should lead the initiative arguing that the country is the best place to set global standards, quote, given its economic and technical standing. The organization, which could resemble existing regulators like the Financial Industry Regulatory Authority, would be made up of leading independent experts and representatives from open source communities and would have the authority to evaluate frontier models before they are released and coordinate an industry-wide

slowdown if they were judged too risky to deploy. The post titled A Framework for Frontier AI and the dawning of a new age argued that the need for global regulation is becoming more urgent as AI systems grow in sophistication. Artificial general intelligence quote is probably only a few short years away. Hasabas said, when we look back on this time in the decades to come, I think we will realize we were standing in the foothills of the singularity, nothing less than the dawning of a new age for humanity. According to Axios, Hasabas has spent months quietly building support for his proposal, including briefing the Trump administration, other AI and European officials and hopes to have the new organization up and running before the end of the year.

He told Axios that the noises I've been hearing from the Trump administration are very positive. The proposal is the latest effort by Hasabas and other industry leaders to establish a coherent framework for governing increasingly powerful AI systems, as well as mitigate the risks they pose. As of yet, there is no global set of rules governing AI specifically, nor a comprehensive set of rules nationally in the U.S. Hasabas, the joint winner of the 2024 Nobel Prize for Chemistry for his work on AI-based protein prediction, also signed his name to a statement calling for tougher protections against AI-aided bioweapons production last month. Asabas's most recent comments, follow a statement from top economists and tech titans, including Anthropic co-founder Jack

Clark and former Google CEO Eric Schmidt, urging world leaders to take the looming economic impacts of AI seriously, end quote. from the AI bubble watch folder, quoting Bloomberg. IBM shares slid by the most in at least 58 years after that company reported preliminary second quarter sales that fell short of expectations, attributing the miss to customers shifting their spending to chips and servers amid AI-fueled shortages. Shares in IBM fell as much as 26% in New York, their biggest intraday loss since at least January 3, 1968. The results weighed on other software companies with Workday and Service Now following

about 6%. Chip stocks, including those of SKHenics and Arm Holdings, were up. An unprecedented global buildout of data center's key to powering artificial intelligence systems has brought about severe shortages of semiconductors, particularly memory chips. The supply squeeze has raised costs for manufacturers of everything from iPads to Xbox consoles, and IBM's results show it's also forcing companies to shift their spending toward servers and chips, leaving less money for other technologies such as IBM's mainframes and software. Discretionary IT spending is worsening and will likely be the main theme across most software companies when they report results. Bloomberg Intelligence Annalists Anrag Rana said in a note, IBM chief executive officer Arvind Krishna said the company

had expected supply chain issues to weigh on results, but he said the company failed to predict that its customers would also end up shifting their spending away from IBM's products to servers, storage, and memory purchases to hedge against further price increases. What played out was worse than our expectations, Krishna said in a letter to investors, adding that its Z mainframes and associated software accounted for much of the shortfall. These conditions require our teams to execute perfectly, and this quarter we faltered. We did not adapt and move quickly enough, and numerous large deals failed to close on the timelines we expected, end quote. The blow to IBM's hardware sales threatens to stymie its efforts to refashion itself into a

high-growth software company through major acquisitions of Red Hat, HashiCorp, and Confluent. Even the company's new focus has made it a target for investors' concern that artificial intelligence tools will replace many current software products. In February, IBM saw a steep sell-off after AI startup Anthropic unveiled a tool that may help modernize a dated programming language that runs on IBM mainframes. IBM, like most software providers, has integrated AI into its products and touted its ability to provide customers with the latest technology. The company has tried to convince investors that AI will strengthen its business, not replace it. IBM executives have said AI-related work increases demand for IBM's infrastructure software,

which lets clients work with leading AI models. Krishna said customers were also distracted by rapidly evolving cybersecurity concerns. Anthropics mythos model alarmed governments and corporations around the world earlier this year with its ability to unearth vulnerabilities that could be exploited by bad actors. banks, technology companies, and other institutions were given early access to the model in an effort to shore up defenses before Mythos was more widely released, end quote. Spotify has officially launched a Talk to Spotify feature that lets users create playlists rolling out in beta to premium users, aged 18 plus in the U.S., Ireland, and Sweden, quoting in Gadget. Spotify already uses a lot of AI too much, some might say, for things like remixing and even generating your own personal podcast via prompts. Now the company is finally letting paid users control the app with their voice or text to do things like create playlists,

learn about songs, or explore their listening history. By typing or speaking directly in the app, you can have a back-and-forth conversation to choose what's playing, learn about the music you love, revisit your listening history, and go deeper on podcasts and audiobooks, all without leaving Spotify, the company wrote. The new feature works from within the home or now playing views on mobile. From the Talk to Spotify feature, you can issue commands like play some artists I haven't heard, before, then fine tune it by saying, add some bad bunny or make it more upbeat. When you hear a song you like, you can ask it to do things like save this song, add it to my cue, or follow this artist.

From the now playing view, you can also learn more about a song or artist. For instance, you can pose questions like, what is the inspiration behind Dua Leipa's radical optimism? When was this album released or what genre is this? Talk to Spotify can then answer those questions and steer you to related artists or stories. It also works with podcasts and audiobooks, letting you learn more about a podcast guest or author. The feature can even tell you about your own taste and history via questions like, when did I first listen to the song or what genres have I been into recently. To use Talk to Spotify, simply press the mic button in the search field to talk or type commands instead. It's now rolling out gradually in beta to premium users 18 or older in

the U.S., Ireland, and Sweden across iOS and Android devices in English. It looks like an appropriate use of AI to help users control and learn about their music, though more features to help us avoid Slop would be nice too, end quote. Only on Hulu on Disney Plus. When critical company knowledge isn't documented, there's a major ripple effect. Work becomes inconsistent. Tools don't get adopted, and knowledge walks out the door when someone leaves. Thankfully, our sponsor, Scribe, was built to fix that.

Their workflow AI platform is trusted by nearly half of the Fortune 500 to capture workflows in real time. Here's how it works. You turn on the Scribe browser extension or desktop app, do a process as you normally would, and Scribe will build a guide as you go. It automatically redacts sensitive information and even suggests improvements to your existing workflows. To see what Scribe could look like for your org, head to Scribe.how slash ride home and mention ride home for your first month of Scribe capture free on select plans. That's S-C-R-I-B-E. dot how slash ride home.

CalShe has launched a forward curve tool for measuring AI compute using event contracts to track the future rental costs of GPUs, storage and memory. Sounds useful, quoting Bloomberg. The prediction markets exchange will offer a forward curve tracking compute, a shorthand for the power, storage, memory, and other resources used in AI processes, according to executives. The shape of the curve should give a sense of where the price to rent GPUs or graphics processing units is headed. We are using prediction markets to build the forward curve, which will provide the market with a view of what compute costs will be in the future for different grades and timeframes of GPUs. Udash Jha, Kalshies chief risk officer said in an interview, forward curves are a snapshot of where traders expect the price of an asset or commodity to be at a specific moment in the future. They plot the price for future delivery and are important metrics

for buyers and sellers as they seek to manage exposure and potential risks. The instruments are used to plot expected future interest rates and moves by central banks. They are also essential tools for companies that need to buy large quantities of natural gas or jet fuel and want to lock in a price in advance. Compute is becoming a commodity in its own right as demand grows for the resources is needed to build and run AI models. The tech rush has fueled investment in new data centers, with some researchers projecting that spending on new infrastructure will one day reach trillions of dollars. Kalshi's forward curve is based on weekly and monthly event contracts related to compute

costs going as far as a year into the future. Based on these contracts, an algorithm builds a forward curve that gives a single price, which can be used to launch other products, including futures and options, Joss said. It's a key enabler for a lot of subsequent hedging, risk management, and even speculative activities, just said. Derivatives exchanges are also looking to list compute futures contracts. CME Group said in May that it would launch computing power futures linked to an index compiled by Silicon Data. Intercontinental Exchange, the owner of the New York Stock Exchange is teaming up with financial infrastructure firm Orne to add futures contracts for computing power as

well, end quote. Finally today, a bit of a long read, but I've mentioned a few times how so-called world models are maybe the next big thing in AI beyond LLMs? What are world models? Well, ours Technica is glad you asked. Quote, there are many parallels between LLMs and world models in terms of architecture and how people expect them to improve over time. For some, though, they're seen as a potential answer to the limitations of LLMs, even though work on them predates that contemporary narrative. The idea that you're going to extend the capabilities of LLMs to the point that they're going to have human level intelligence is complete nonsense. former meta chief AI scientist Jan Lacoon told Wired earlier this year,

Lacoon has made waves with an opinion that some working in AI and LLMC as contrarian, but he's actually speaking for a sizable segment of the field. Over just the past few months, world models have advanced from a research topic, which they still are, of course, to the basis for new commercial projects and huge funding rounds. World Labs and AMI reportedly raised around $1 billion each in February and March, respectively, and Runway also raised $315 million in. February. It's important to note that world models is an umbrella term that is often thrown around without a clear definition, though. It's definitely an overloaded term, Vincent Sitzman

told me in a lengthy conversation about the research and concepts underlying world models. Sitzman is an assistant professor at MIT who has published research on neural rendering, visual computing, and robotics. He leads the scene representation group within MIT's computer science and artificial intelligence lab. When asked to give a definition, he simply described a world model as any model that takes in an interaction, and given that interaction, it enables you to simulate what would happen next in some environment. When announcing its GWM1 family of models in December runway, defined a world model as an AI system that builds an internal representation of an environment and uses it to simulate future events within that environment. Further,

the aim of general world models is to represent and simulate a wide range of situations and interactions like those encountered in the real world runway added. I also spoke with Ben Mildenhall, co-founder of World Labs, a former Google Computer Vision and Physics researcher and co-creator of neural radiance fields, Nerth, a method for constructing navigable 3D scenes from 2D images in a format that is differentiable and therefore useful in machine learning contexts. The key things that are distinguishing it from an LLM are demonstrating degrees of spatial and maybe for lack of a better word, continuous understanding, he said. A very distinguishing aspect of interacting with an LLM is they are turn-based, meaning users type some text, there's a pause, and then they get

a block of text back. By contrast, he sees a world model as synchronous real-time as a system. Something that would define a world model is the degree of freedom that you have in interacting with the spatial world where you do not have this mediated linear journey of A-than-B, then A-then-B, he said. a user or agent is utilizing a world model, they are actually able to interact with it like it's some sort of world and you are taking continuous actions where there are parallel things happening at the same time. Milden Hall's co-founder Fifi Lee has written that she believes there are three criteria that define a world model. World models can generate worlds with perceptual, geometrical, and physical consistency, are multimodal by design and can output the next states

based on input actions. Today, what most people mean when they say world model is generating pixels, so like generating a realistic video, conditional of the actions, Sitzman said. We've seen video generation models gain traction over the past couple of years. Runway, for example, built its reputation as a company that makes video models and related tools that are used by filmmakers, advertisers, and others in creative fields. Now that focus has shifted as the company has made clear its intention to expand beyond those areas to focus on world models like GWM1 as well with an eye toward applications in robotics and beyond in the coming years.

That might seem like a jargon. lateral pivot, but if you consider how these world models are being developed, it's a natural next step. In many cases, they're a direct extension of the work previously done on video models. Physics are key for robotics and other areas of physical AI. Any autonomous physical unit must demonstrate something, at least analogous to an understanding of the laws of the physical world. Intuitively, it might seem odd to say that something built primarily from video data could develop anything resembling physical understanding, but many researchers and the company's building products on top of their research, believe that video models appear to demonstrate a

genuinely useful ability to reflect or predict real-world physics, at least when looking at their inputs and outputs. In order to solve this problem of predicting the next frame of a video very well, you need to basically predict so many aspects of the physical world, so many aspects of how objects objects move, how people move in an environment. Physics, Jermondis told me, end quote. This is a bit of a long read, so I'm only quoting a bit from it. But if you find this interesting and want to see what people think the next big thing is,

click through to read the whole thing. Nothing more for you today. Talk to you tomorrow.