China pulls a Silicon Valley on… Silicon Valley
When Chinese artificial intelligence startup DeepSeek rocketed to the top of the AI race over the weekend, it upended not just a ruthless corporate competition for tech supremacy but a geopolitical one. America was supposed to be the world leader on AI; suddenly, China had a credible claim.
Wall Street was shocked. So was Washington. And yet for anyone familiar with previous digital upheavals, it shouldn't have come as much of a surprise.
Silicon Valley's own mythos is built on the tech-world challengers who beat the odds and slew corporate Goliaths through sheer technical ingenuity. If anything is remarkable about the seeming DeepSeek coup, it's that companies like OpenAI — which were scrappy, disruptive startups just a couple of years ago — have already matured into the kind of big, connected firms that get caught on their back foot by faster-moving rivals.
'There's a certain extent to which maybe [large AI firms] thought this was harder to do than it was,' Dean Ball, a research fellow at Mercatus and author of the Hyperdimensional newsletter, told DFD.
Given America's strict export controls on high-end chips, analysts are still trying to figure out the exact details of how DeepSeek managed to match the top American models on a shoestring. DeepSeek itself claims it was done with ingenuity rather than fancy equipment — based on good old-fashioned optimization and low-level programming grunt work.
'You can keep compute away from China, but you can't export-control the ideas that everyone in the world is hunting for,' Ball said. 'This is just 1,500 lines of Python code, and it's going to spread.'
Silicon Valley, and much of our modern digital landscape, was built on exactly such ingenious workarounds. In the 1980s IBM had a seemingly permanent stranglehold on the nascent personal computer market, until entrepreneurs and tinkerers across the country reverse-engineered the company's vaunted firmware to make their own cheaper, faster competitors. In the Web era, Google Chrome dethroned its clunky, monopolistic predecessor Internet Explorer using open-source components. (The David-and-Goliath effect can sometimes even spill over and revolutionize other, non-"tech' industries, as in the case of the Napster founders realizing that people might prefer to listen to music on their computers rather than by buying it on pricey plastic discs.)
When this kind of out-innovation happens in the private sector, it's celebrated as 'disruption,' the lifeblood of American capitalism. When it happens at the hands of a Chinese firm and embarrasses a multi-billion-dollar national champion industry, as President Donald Trump's new administration wants AI to be, it's cause for alarm. Senate Majority Leader John Thune (R-S.D.) told POLITICO today that the DeepSeek model's performance is 'a wake-up call for us that we've got to step up our game,' and venture capitalist and Trump ally Marc Andreessen declared it 'AI's Sputnik moment.'
There's a huge irony at the heart of this problem, and Andreessen's post (unintentionally) gets right at it. The tech world, particularly its right-leaning elements, have for years used the threat of Chinese supremacy to advocate for more government investment and support. The result is a tech industry whose relationship with government has slowly grown to resemble that of stalwart aerospace giants like Boeing and Lockheed Martin, their products given special treatment by a government that sees them as crucial to national security.
Eric Schmidt, the longtime Google CEO and founder of the Special Competitive Studies Project focused on U.S.-China geopolitical competition, wrote in an op-ed published in The Washington Post today that America should invest more in research and development to maintain the U.S.' 'competitive edge,' and that 'there is clearly mounting pressure on America's Big Tech players if DeepSeek can compete with them using far fewer resources.'
Despite his newfound closeness with American AI companies, Trump actually embraced the DeepSeek news, calling it a 'positive development.' He said that he hoped 'instead of spending billions and billions, you will spend less and you'll come up with hopefully the same solution.'
So which part of the industry is going to prevail? Ball told DFD that likely the compute heavyweights and their flyweight counterparts do ultimately need each other — the breakthroughs achieved at great cost by companies like OpenAI, then inevitably replicated by a cheap competitor, and vice versa, in a never-ending cat and mouse game that speeds up innovation overall.
'There's not some finish line we're racing to,' Ball said. 'We're going to get models that are more capable, and the country that quote-unquote 'wins' will be the country that has the most creative and wide-ranging uses for AI, and that's going to be compute-expensive and require a huge infrastructure buildout.'
Still, if America's established AI players hope to keep up and avoid surprises like this most recent one — and, maybe more importantly, maintain the political credibility that allows them to make big asks of their federal government — they might need to pay more attention to their competitors, and reacquaint themselves with the improvisational, cheaper-is-better ethos on which Silicon Valley was built.
mr. musk goes to washington
Lawmakers are skeptical of Elon Musk's plans for an overhaul of government, given his apparent lack of familiarity with it.
POLITICO's Andres Picon reported for E&E News this morning on a chorus of Washington lawmakers who support his goals but think the tech mogul is in for an education in how the capital (and indeed the Capitol) works.
'I think what [Musk is] learning is that the rules around here, how you cut the government — there's political realities,' Senate Budget Chair Lindsey Graham (R-S.C.), said. House Agriculture Chair Glenn 'G.T.' Thompson (R-Pa.) said he 'would welcome Mr. Musk to come and sit down so we can have a really good conversation about agriculture, because I want him to understand American agriculture.'
And furthermore, even some Republican lawmakers are concerned about potential conflicts of interest: 'Say you're the chairman of a committee. … Would you be concerned that Donald Trump at some point would say, 'Hey, I've got this great idea: Elon's gonna give me $10 and I'm gonna lease NASA to him for the next 50 years'?' one anonymous longtime Republican lawmaker said. 'You know that kind of shit is possible … You have to wonder about the separation between the best interest of the country and the best interest of Elon.'
fact-checkers under fire
Fact-checkers have been under siege since Meta's Mark Zuckerberg made his high-profile call to stop working with them.
POLITICO's Mathieu Pollet spoke to five people at fact-checking organizations Meta once partnered with. They all said they've faced increased online harassment, attacks from politicians and even death threats in recent weeks.
'The first thing we've noticed after Zuckerberg's statement … was a huge spread of harassment toward fact-checkers,' Aistė Meidutė, editor of the Lie Detector fact-checking project at the Lithuanian news outlet Delfi, told Mathieu. 'It was a huge beat to our credibility. We see the after-effects right now … We are very worried.'
Although Meta's announcement was rolled out in the U.S., observers think it won't be long before the policy shift reaches Europe as well — even though that will set the U.S. company up for a clash with the European Union's Digital Services Act. 'It's hard for me to understand why they would keep the program here in its current form if they won't have it in the U.S. where they are under the most direct political pressure anyway,' said Morten Langfeldt Dahlback, the head of innovation at Norway's fact-checking firm Faktisk.
post OF THE DAY
The Future in 5 links
Stay in touch with the whole team: Derek Robertson (drobertson@politico.com); Mohar Chatterjee (mchatterjee@politico.com); Steve Heuser (sheuser@politico.com); Nate Robson (nrobson@politico.com); Daniella Cheslow (dcheslow@politico.com); and Christine Mui (cmui@politico.com).
Hashtags

Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles


CNET
a few seconds ago
- CNET
I Gave GPT-5 Some Work Tasks in Copilot. Here's What's New
OpenAI's latest GPT-5 model promises to be the most advanced AI assistant yet, but for the millions of workers stuck using Microsoft Copilot for their daily routines, the reality is a bit more nuanced. Microsoft announced last week that GPT-5, OpenAI's latest ChatGPT AI model, is now integrated across its Copilot services, including Microsoft 365 Copilot, Copilot Studio, GitHub Copilot and the standalone Copilot app. While GPT-5 does deliver some noticeable improvements in polish, detail and helpfulness, the differences feel more like a software update than an impactful revamp. For professionals weighing whether to supplement their work Copilot toolkit with ChatGPT's premium offering, the question comes down to whether or not GPT-5 is enough of a leap forward to make any difference in your work. The update introduces "Smart Mode," a real-time routing system that directs tasks to either a fast model or a deeper reasoning variant based on each user's intent and the complexity of the request. I spent hours testing GPT-5 in Microsoft Copilot to explore what you couldn't do before but can now, and how the latest model stacks up to the regular Copilot Quick Response mode. Here's what I found. Read also: ChatGPT Image Generator Is in Microsoft Copilot Now. Here's What You Can Do With It What you couldn't do before, but can now with GPT-5 GPT-5 is now accessible across all platforms (web, Windows, macOS, mobile), getting a truly universal rollout in Microsoft. The goal was for Microsoft platforms to benefit from "deeper reasoning" as the underlying architecture includes one model for quick responses and another for deeper "thinking" that would be used for more complex tasks, with a router selecting between them automatically. GPT-5 has already experienced backlash since its release on Aug. 7. Users quickly complained that the new model has a less conversational, more stuffy and corporate tone than previous models, which were initially replaced by GPT-5. In response, OpenAI made older models like GPT-4o available again. While casual users were dissatisfied with the more corporate tone, it might be useful in the workplace. GPT-5 is also integrated into GitHub Copilot and Visual Studio Code, enabling developers to handle larger-scale coding, refactoring and agentic workflows. Access is available through paid Copilot plans and can be enabled via administrative settings. Copilot Studio users can also now select GPT-5 for their agents, using Smart Mode or opting for deeper reasoning when building automations or workflows. "With the new Smart Mode, Copilot automatically adapts to your task and routes you to the most capable model, removing the need to switch or guess," The Copilot Team wrote. "Whether you're solving complex problems, writing creatively or turning ideas into visuals, Smart Mode delivers the smartest, most intuitive experience yet." Read also: ChatGPT's Boss Says You Still Shouldn't Trust It as Your Main Source of Information How to get started with Copilot You can follow these steps to get started with Copilot: Open any Copilot interface (web, desktop or mobile). Choose Smart Mode in the composer. Start interacting with GPT-5. There's no extra setup required and it's free to try. Read also: GPT-5's Voice Mode Can Hold a Decent Conversation, but Please Don't Talk to ChatGPT in Public Side-by-side comparison of three use-case experiments To get a clear sense of how Microsoft's GPT-5 mode inside Copilot stacks up against the standard Copilot model, I spent hours running the two modes through three real-world prompts. I wanted to measure distinct skills like reading and summarizing a large block of text, creating an image and analyzing a dataset. The goal was simply to see if GPT-5 feels like a big leap forward or an incremental step -- if there's any substantial change at all. For the first test, I played the role of a law student staring down an intimidating reading assignment. My prompt asked Copilot to read a 137-page textbook chapter and generate a study guide from it. Both models delivered summaries in the same go-to GPT format of bullet points with emojis sprinkled in. GPT-5's version was a touch longer and just slightly more thorough, but the difference wasn't dramatic. If you are a student or employee looking to summarize large chapters of text quickly, then any mode of Copilot would suffice. The second test was pure creativity. I prompted that I am a painter looking for a reference photo that doesn't exist, and I asked the two Copilot modes to create a fantasy city at dusk, filled with dragons and a dramatic evening sky. Here again, the results were surprisingly close. The non-GPT version of Copilot produced a decent rendering that would have helped as a reference photo, while GPT-5's output was noticeably richer and more vivid, with stronger perspective. It wasn't a radical improvement, but you could still see the upgrade in image quality. Quick Response versus GPT-5. Screenshot by Macy Meyer/CNET Finally, I tested the versions on data analysis, handing over figures on air travel passenger growth, airfare changes and yield shifts from Q4 2020 to Q4 2021. Both modes broke down the numbers in a useful way that would have been helpful had I been presenting my findings in a business or finance meeting. The major difference was in presentation. GPT-5 created a "leaderboard" of the winners and laggards, making the insights pop with more clarity and readability. The standard model stuck to a straightforward design with three separate tables, more so organizing the data rather than interpreting it with key findings or providing context for the data figures. Across all three scenarios, it became obvious that GPT-5 is a little sharper, a little more polished and a little more helpful. It produces longer and more detailed summaries, slightly richer visuals and more thoughtful data presentations. But the keyword here is "slightly." This isn't a night-and-day transformation. If you were already impressed with Copilot before, GPT-5 may feel like a welcome refinement rather than a revolution. Can you link your ChatGPT account to Copilot? The short answer is no. ChatGPT and Copilot remain separate platforms, and there's currently no integration allowing you to link your ChatGPT account to your Copilot account. GPT-5 powers both ChatGPT and Copilot, but usage, user interfaces and account contexts remain siloed. For instance, the ability to change the bot's personalities introduced with the GPT-5 release is not carried over to Copilot. In ChatGPT, GPT-5 is the default model for free and paid users, with varying usage limits based on which plan you have. In Copilot, GPT-5 appears via Smart Mode in the app ecosystem, which is distinct from ChatGPT workflows. "We continue to use the latest models from OpenAI and we are also beginning to apply our own in-house technology to the experience," a Microsoft spokesperson said. "This combination gives us more flexibility to apply the best model to each customer's interaction with Copilot." So, if you're hoping to mix and match features -- like shopping plugins, the Code Interpreter that's available in ChatGPT Plus or a custom prompt you'd developed on one of the platforms -- that cross-platform compatibility isn't available right now.


New York Post
a few seconds ago
- New York Post
Car company creates hilarious tool to channel drivers' anger — and avoid dangerous road rage incidents
A car company in China has a cartoonish approach to curing road rage. XPeng has unveiled a quirky new feature designed to subdue angry drivers — though it may also prove to be a distraction. Suppose you've ever thrown a digital banana peel at an opponent in Mario Kart. In that case, you might know the satisfaction that throwing something like a digital shoe or an angry face would illicit. Advertisement 3 The company developed an augmented reality game that allows the user to fire emojis at the wrongdoer. Xpeng The Guangzhou-based car company has developed an augmented reality game that allows the user to fire off emojis at the wrongdoer, which will project over the windshield in 3D space and give the illusion of hitting the cars. The steering wheel will have a customizable button that serves as a trigger, and the system will identify the target through its camera array before launching the emojis that appear to explode against the real car. Advertisement While they might not be as satisfying as throwing up the middle finger or watching the cup of coffee you hurled splash against your enemy's car window, it could provide drivers with some temporary relief from pent-up anger. The program, dubbed the 'Road Rage Reliever,' was unveiled earlier this month during a presentation in China. XPeng CEO He Xiaopeng said the bizarre feature represents 'technology-driven emotion,' Fast Company reported. 'We used to prioritize technology first, but starting this year, we will prioritize experience first,' Xiaopeng said, adding that the AR game is a way to 'be civilized and experience 'civilized frustration' rather than engaging in dangerous behaviors.' Advertisement 3 Road Rage Reliever converts the windshield into a virtual battleground featuring an 87-inch-wide augmented reality heads-up display. REUTERS Road Rage Reliever converts the windshield into a virtual battleground featuring an 87-inch-wide augmented reality heads-up display (AR-HUD) that takes up the driver's entire field of vision and then some. According to the developers, XPeng and Chinese electronics manufacturer Huawei, the program has advanced optics and calculates the exact distortion necessary for the eyes to believe that things are not displayed on the windshield, but rather floating in the space 33 feet in front of the car. Advertisement The optical illusion convinces the brain to interpret the 3D objects as real, the companies claim. The development comes as road rage has become an increasing problem around the globe. In the U.S. alone, 92% of Americans have said they witnessed road rage in 2024 and 89% said they have been a victim, according to Consumer Affairs. Though road rage isn't new, there has been a significant spike in violence a result. In 2024 alone, 116 people were killed in road rage shootings and 362 were injured, according to Siegfried and Jensen. But the big question is whether or not its safe, with already growing concerns about driver distractions with touchscreen infotainment systems in cars. 3 XPeng CEO He Xiaopeng said the bizarre feature represents 'technology-driven emotion.' AP Photo/Ng Han Guan XPeng said that 'drivers are responsible for judging when it's safe to use.' Studies have shown that heads-up display systems can significantly improve the subjects' attention to risky areas, particularly when compared to traditional displays in the middle of the dashboard that force the driver to take their eyes off the road. The HUD placement keeps eyes in the driver's forward field of view, but some research shows that drivers were more likely to glance at HUDs compared to traditional displays, which can make them 'potentially distracting … Because the HUD is in the driver's field of view, drivers may fixate on it and fail to perceive events in the environment,' the National Highway Traffic Safety Administration said. Advertisement However, that study refers to traditional car HUDs that are essentially small dashboards in the field of view, like those in high-end luxury cars like BMW and Mercedes. The XPeng feature overlays real AR elements onto the road and makes them part of the environment — which could potentially increase attention, improve response time and reduce the difficulty of processing information in dangerous scenarios.


The Hill
a few seconds ago
- The Hill
The ‘big beautiful bill's' smart tax reform will grow wealth in every zip code
Democrats joined together to vote 'no' on the One Big Beautiful Bill Act, with one senator calling it 'bad for our country' and 'bad for our families.' But is that an accurate critique? While the legislative package signed into law by President Trump on July 4 traces its lineage to the Tax Cuts and Jobs Act, this updated version goes well beyond tax reduction — it's about expanding opportunity. From child savings accounts to stronger investment in Opportunity Zones and education choice, the new law gives American families the tools to rise — and to build. At its core, the 'big, beautiful bill' expands economic opportunity in every zip code. One of the law's most immediate benefits is the massive tax relief it delivers to working- and middle-class households. For example, a married family of four earning $100,000 is expected to see its tax bill shrink by close to $3,000 annually. That's not just down to lower tax rates, but to the doubling of the standard deduction. Moreover, a single restaurant server will realize up to $1,300 in tax relief with the act's No Tax on Tips provision. Additionally, a veteran working an overtime job, like 4.8 million other veterans, will receive $1,400 in tax savings. These savings aren't just pennies to the dollar. This is money for child care, car repairs, rent or groceries — real expenses that a family relies on to survive. Small businesses stand to benefit from these tax breaks too, giving entrepreneurs the breathing room needed to hire at a decent wage, expand and reinvest in their businesses. What sets this law apart is how it pairs near-term relief with long-term wealth building. The new child savings account provision is a prime example. Every American newborn will receive an account with $1,000 from Uncle Sam that grows tax-free over time, generating the kind of seed capital that could one day fund higher education, a first home, or a small business. It's a simple but transformative tool. Whether a child is born in Seattle or South Bend, this provision offers a path to generational wealth that doesn't depend on zip code, income level or political affiliation. It provides every American access to a head start in life. Just as access to financial opportunity is essential for every child, so too is the freedom to pursue a quality education. Through Opportunity Scholarships, the legislation empowers parents to choose the learning environment that best fits their child's needs. Families will receive financial support to send their children to private, charter, or faith-based schools — a powerful lifeline for families in communities where public education has failed their children. That kind of flexibility shouldn't be a luxury. With this bill, it will become a national standard. The new law also extends and expands the Opportunity Zones program. Originally established under the Tax Cuts and Jobs Act, Opportunity Zones channel private capital into historically overlooked communities through smart tax incentives. Under the new provisions, Opportunity Zones will prioritize rural communities and include enhanced transparency measures to ensure investments translate into real impact. According to the White House Council of Economic Advisers, the program's expansion is projected to drive more than $100 billion in private investment into underserved regions across the country. Taken alone, these provisions support American workers, parents, and entrepreneurs. Put together, the new law forms a unified strategy for upward economic mobility — one that aligns tax relief, education reform, savings, and local investment into a coherent framework for growth. The One Big Beautiful Bill Act lowers costs for families today while investing in the next generation's ability to grow, save, and thrive. That's a law that creates real opportunity for every American.