Google Struggles with the Web and Trump Copies Xi Jinping

Google says goodbye to the web, even for Chromebooks

Google is presenting its 'Google Books' as the future of AI laptops. Android-based notebooks with a 'Magic Pointer' that activates a full-screen Gemini mode when wiggled back and forth. The devices are set to be launched this year by Acer, Asus, Dell, HP, and Lenovo. Google consistently avoids the word 'Android,' speaking instead of 'Gemini Intelligence.' The kicker: Apps from the smartphone are streamed directly to the laptop, and files are transferred seamlessly. A glowing 'Glowbar' on the lid signals... something; Google hasn't revealed what exactly. So, after 15 years of the Chromebook philosophy ('everything runs in the browser'), the native app world is finally coming to Google laptops. → arstechnica.com

Synthszr Take: Google is executing a spectacular U-turn here without explicitly acknowledging it. In 2011, Chromebooks launched with a radical promise: The browser is the operating system, local apps are dead. Now come Google Books with Android apps, phone streaming, and a Gemini cursor that demotes the web to mere content logistics. The Magic Pointer is less of an innovation and more of a capitulation to Microsoft's Recall debacle (remember that?). Google has apparently realized that the web-first philosophy is becoming a hindrance in the AI era. If Gemini is taking over the entire screen and pulling context from all apps anyway, why bother with the detour through Chrome? The hardware partners are playing along because, after the niche existence of Chromebooks, they can finally sell proper laptops again. The web is becoming the dark web of APIs, while users live in AI-curated app bubbles.

Google reinvents the mouse pointer

Google DeepMind has reinvented the mouse pointer. After half a century as a silent tool, it can now understand what it's pointing at and why that's important. The experimental AI-powered technology combines pointing with natural language: you point at a building and say, 'Show me the way there' — done. No more cumbersome prompts, no more copying between windows. The four design principles behind it: The workflow is maintained (AI works across all apps), 'Show and Tell' replaces long text commands, natural language like 'Fix this' or 'Move that' is sufficient, and pixels become interactive entities. A photo of a note becomes a to-do list, a still frame in a travel video becomes a booking link for the restaurant. Google is already integrating the technology into Chrome and the new Googlebook laptops, where the 'Magic Pointer' makes Gemini available directly under the cursor. → deepmind.google

Synthszr Take: The mouse pointer was already a revolution in 1984 — now Google is finally turning it into what Xerox failed to achieve back then: a truly intuitive interface between human and machine. The irony is brutal: While everyone is talking about chat interfaces, Google is solving the real problem right where it arises — directly in the workflow, without 'AI detours.' This is reminiscent of Apple's 1-Click patent from 1999: The most brilliant ideas are often the simplest. Google's 'This-and-That' principle finally turns the mouse into what it always should have been: an extension of our intention, not just our hand. Who needs command lines or chat windows when the context is already there? The real disruption isn't in new interfaces, but in finally getting the old ones right.

Trump copies Xi Jinping's AI control strategy

Donald Trump is flying to China this week, and his guest list reads like a Who's Who of the American tech elite: Tim Cook (Apple), Elon Musk (Tesla/SpaceX), Dina Powell McCormick (Meta), as well as the CEOs of Micron, Cisco, and Qualcomm. Jensen Huang of Nvidia is conspicuously absent — the very man who warned in April that a 'loser mentality' on chip exports would jeopardize America's AI dominance. Without the world's most important chip manufacturer, a semiconductor deal becomes less likely, even though announcements from Micron seem possible. Despite shifting production to India and Vietnam, Apple remains the big winner in China: The iPhone 17 is driving quarterly figures to record levels. But the real story lies elsewhere: While Trump presents his tech magnates as trophies of a 'hands-free' innovation policy, his administration is secretly adopting Beijing's control model for artificial intelligence. → MIT Technology Review

Synthszr Take: For years, China has required AI companies to submit their models for review — both for security and political sensitivity. Trump's planned Executive Order would establish the exact same thing in the US: Frontier labs would have to submit their latest models to the White House for review. The Commerce Department has already struck deals with Google DeepMind, Microsoft, and xAI for 'national security reviews.' Meanwhile, the Pentagon is battling Anthropic in court over military use. What's happening here is the Americanization of Chinese industrial policy: state control over AI development, just with a private-sector veneer. The irony is: Trump portrays himself as a freedom fighter against Big Tech, while simultaneously orchestrating their closest ties to the state since the Manhattan Project.

OpenAI Daybreak is security as a business model

OpenAI is launching Daybreak, turning cybersecurity into a service product. CEO Sam Altman announced plans to work with 'as many companies as possible' to continuously secure software. The timing is no coincidence: One month after Anthropic's mysterious Project Glasswing, which grants only select partners access to its 'dangerously capable' Claude-Mythos-Preview model, OpenAI is taking the opposite approach. Instead of exclusivity, there are two buttons: 'Request a vulnerability scan' and 'Contact sales.' The message is clear: Security is becoming a scalable service, not a secret project for insiders. Daybreak uses Codex Security to first create a threat model (who has access, where are the vulnerabilities) and then search for exploits in the actual code — and theoretically, patch them too. → MIT Technology Review

Synthszr Take: OpenAI is making a virtue of necessity and turning the security narrative 180 degrees. While Anthropic stages its model as 'too dangerous for the general public' (software developer Daniel Stenberg calls this an 'astonishingly successful marketing stunt'), OpenAI is selling security as a self-service. It's clever: The very same capabilities that make models dangerous are being repurposed as the solution. But the real move is deeper. OpenAI is building a new revenue stream alongside API calls — continuous security scans as a subscription model, likely with automatic patching as a premium feature. In a world where 86 percent of code will soon be written by AI (Gartner forecast for 2028), Security-as-a-Service is becoming mandatory insurance. OpenAI is positioning itself as the provider that both writes AND secures the code: Vertical integration at its finest.

Claude Opus 4.7: most people are only using 20 percent of its capabilities

Anthropic already released Claude Opus 4.7 in April 2026. The model can reason more precisely, follow instructions literally, and execute autonomous tasks more reliably than all previous versions. Still, most users are wasting 80 percent of its power because they prompt the model like ChatGPT or carry over habits from Opus 4.6. The new Effort-parameter — the most important new feature — directly controls the model's intelligence allocation. Anyone who ignores it or sets it to a low level will receive correspondingly superficial answers. The model now interprets instructions strictly literally: an instruction for a single element is no longer tacitly generalized to all elements. The tone is more direct, with fewer of the warm qualifications that characterized Opus 4.6. → Linas from Linas's Newsletter

Synthszr Take: Claude Opus 4.7 illustrates the classic innovation dilemma of artificial intelligence: The infrastructure is evolving faster than users can keep up. While Anthropic is (probably) already working on the next model, most people are still experimenting with copy-paste prompts from the ChatGPT era. The Effort-parameter is more than just a technical detail. It makes it clear that AI models are not magic oracles, but precision tools that demand precise operation. The problem lies less in the technology and more in our mental inertia: we treat new systems like old ones and are surprised by mediocre results. The real challenge for companies will be to train their teams faster than the next model generation appears.

Kyutai makes language models obsolete: and runs on a laptop

The French AI lab Kyutai has quietly ignited the next stage of speech processing. Their Pocket-TTS model has only 100 million parameters — one-tenth the size of comparable systems — and runs in real-time on a standard CPU. No GPU cluster, no cloud, just local computing power. The model supports voice cloning and achieves the quality of systems ten times its size. In parallel, they have developed Moshi: the first speech-native dialogue system that processes speech directly, without the detour through text. The latency is in the single-digit millisecond range. Moshi understands emotions and non-verbal communication — things that are lost in text conversion. The spin-off Gradium now wants to make this technology production-ready → AINews

Synthszr Take: Kyutai shows what happens when you ask the right question: Why do we even need billions of parameters? The answer is brutally simple: We don't, if we approach the problem differently. Speech-native instead of a text detour, 100 million instead of 10 billion parameters, CPU instead of a GPU farm. This isn't an incremental improvement; it's a completely different approach. While everyone else is betting on bigger models, Kyutai is making them smaller and more specialized — and suddenly, professional speech processing runs on any office computer. The consequence: Voice interfaces will become a commodity. In two years, every business app will be controllable with natural language, locally and in real time. The major cloud providers are losing their most important differentiator: computing power.

Amazon launches 30-minute delivery with AI

Amazon is rolling out 'Amazon Now' in major American cities: 30-minute delivery for everything from groceries to household goods. The fee is $3.99 for Prime members, $13.99 for everyone else. Additionally, orders under $15 incur a fee of $1.99 (Prime) or $3.99 (non-Prime). The service is launching in Atlanta, Dallas-Fort Worth, Philadelphia, and Seattle, with other cities like Austin, Denver, and Houston to follow. Amazon speaks of 'smaller fulfillment locations close to customers' — which sounds like more warehouses, but it's something different. The 24/7 availability suggests fully automated processes, not shift work in mini-depots. → Techpresso

Synthszr Take: Amazon is selling this as a logistics innovation, but the real story is AI-driven inventory optimization. 30 minutes only works if you know in advance what customers will order — and that's exactly what machine learning does based on billions of data points. The 'smaller locations' are not traditional warehouses, but algorithmically stocked buffers with the top 500 products for each area. Amazon is using the same pattern here as with AWS: first build it for yourself, then sell it as a service. The pricing structure shows where this is headed: Prime members pay 70% less than non-members. This is platform economics in its purest form. DoorDash and Uber Eats continue to compete with human drivers and variable fees — Amazon is automating the entire supply chain.

OpenAI makes hundreds of employees millionaires

In October, 75 OpenAI employees each cashed in $30 million. That's the maximum amount OpenAI allowed in the stock sale — a total of $6.6 billion flowed to over 600 employees. Average: $11 million per person. Most are parking their remaining shares in foundations to save on taxes. OpenAI requires a two-year holding period for shares, so for many, this was their first chance. Some employees already earn a base salary of over $500,000; Meta offers top talent $30 million packages over four years. The Wall Street Journal calls it the largest pre-IPO payout in tech history → Tech Brew

Synthszr Take: These numbers mark a turning point: AI wealth is no longer just being created by founders and VCs, but also by regular employees with lucky timing. The rush to sell suggests a panic about the bubble bursting – those who waited for an IPO during the dot-com boom often lost everything. OpenAI is valued at $852 billion (twice as much as SAP and Siemens combined) and is aiming for a trillion at its IPO. The Bay Area is already feeling it: real estate prices have risen by 14% in one year. What's emerging here is a new two-class society of the AI era — the 75 lucky ones now belong to the richest 1% in the US. The rest are hoping the next funding round will reach them too, before AI automates their jobs.

Young tech founders go into 'Monk Mode' and dump their girlfriends

A new phenomenon is sweeping the Valley: Young founders are breaking up with their partners with the excuse, 'It's not you, it's my startup.' Lee Beckman, 30, founder of an EdTech startup, ended his five-month long-distance relationship because he was 'so drained from building the company.' The trend has a name: 'Founder Mode' — a toxic mix of self-exploitation and performance theater. According to a survey of 200 founders, 73% have sacrificed their relationships for their startup. The reasoning: One must go 'all in,' tolerate no distractions, and be fully dedicated to the company. Dating apps are already reacting: Bumble is testing a 'Founder Mode' badge so that like-minded people can find each other. → Business Insider

Synthszr Take: The real irony is that AI tools are supposed to make work more efficient — instead, founders are using the time saved to work even more. This is the Jevons paradox in its purest form: More efficiency leads to more consumption of the resource, not less. These guys aren't sacrificing their relationships for groundbreaking innovation, but for a LinkedIn-worthy self-image. 73% have left their partners — for what, exactly? For the next calendar app that nobody needs? The problem isn't the workload, but the confusion of busyness with business. Someone who has no time for a relationship probably also has no time to understand user problems.