P.INC AI DISPATCH

A lot has happened in the last two weeks. The sudden escalation of the Iran conflict caused a rise of vibe-coded "situation monitors", reflecting people's desire to track geopolitical rainstorms that turned out to be closer to home than they thought. After the Pentagon designated Anthropic a supply chain risk, Trump ordered all federal agencies to stop using the company's technology. Within hours, OpenAI stepped in and signed the Pentagon's deal. What happened next could be described by some as an unexpected marketing campaign for Anthropic: ChatGPT uninstalls surged 295% overnight, a boycott movement called QuitGPT claimed over 2.5 million participants, Claude shot to No. 1 on the App Store for the first time, people chalked "GOD LOVES ANTHROPIC" outside its San Francisco headquarters, an open letter titled "We Will Not Be Divided" gathered nearly a thousand signatures from OpenAI and Google employees, OpenAI’s robotics lead resigned in protest. Then, on February 28, the US and Israel struck Iran and the stakes became viscerally physical. Iranian retaliatory strikes hit three AWS data centers in the UAE and Bahrain, the first time commercial cloud infrastructure has been deliberately targeted in war. You may have noticed Claude going down for a few hours last week (now you know why). There is something vertiginous about all of this. The mundane and the geopolitical are folding into each other at a speed that's hard to process. The same tools we use daily to think, write, and speak for us now depend on defense decisions unfolding elsewhere. Your morning chat with Claude and a missile strike on an AWS data center are now, in a very literal sense, part of the same system. And perhaps the most recursive detail of all: the news coverage of this entire saga will inevitably make its way into the training data for future versions of these models.

Netflix just paid up to $600 million for Ben Affleck's AI production startup, Interpositive, a company that develops AI tools for filmmaking, training models on a production’s own footage to help with tasks like lighting adjustments, reframing shots and fixing continuity problems. The premise is that AI can radically reduce the cost and time of filmmaking without replacing the people who make it. It's the clearest signal yet that the creative industries shouldn’t wait for AI to happen to them and that those moving fastest are the ones framing AI as tools created by artists. Affleck, who has spoken much publicly about the dangers of AI automation, now cleverly positions it as an artist-centric response. Perhaps the companies that survive this transformation will be the ones that move fast while centering the artist, building new forms of licensing, clear provenance, and tools that augment craft rather than erase it.

A familiar pattern in technology is that the West invents, while the East integrates technologies in more widespread and mundane ways. This week, China turned OpenClaw, the autonomous assistant that went viral for populating Moltbook (the AI exclusive social media), into a national phenomenon. Massive OpenClaw installation events were organized across the country with a newly booming paid installation economy where providers charged anywhere from 100 to 1,500 yuan. Nearly a thousand people lined up outside Tencent's Shenzhen headquarters for a free installation event staffed by volunteer engineers. The crowd included students, teachers, retirees, parents holding toddlers. Tencent, ByteDance, Alibaba, and Xiaomi have all now launched their own OpenClaw compatible products. Local governments in Shenzhen, Wuxi, and Changshu rolled out subsidies and free computing credits. The speed of diffusion within just two weeks is remarkable and revealing. In the US, AI adoption is still largely mediated through subscriptions, but in China, it's apparently spreading through street stalls, WeChat groups, and e-commerce listings on Taobao (until the state decides otherwise). While the West builds the frontier models, China is undeniably showing what mass deployment actually looks like.

Andrej Karpathy just released a tool called Autoresearch, which turns model training and AI research experiments into an autonomous loop. Karpathy left it running for two days on his own well-tuned GPT-2 codebase: 700 experiments, 20 real improvements, an 11% speed gain, including oversights he'd missed over two decades of work. Shopify's CEO adapted the pattern overnight and woke up to a 0.8B model that outscored his previous 1.6B. The idea is spreading fast with at least three GitHub projects popping up over the weekend, which applied the same loop to marketing, agency orchestration, and "zero-human companies."

Six regulatory moves are unfolding simultaneously around the world. South Korea's AI Basic Act, which went into effect in January, created the first national framework for "high-impact" AI in hiring and finance. Singapore launched the world's first governance framework for agentic AI. The UAE's AI Act, effective this month, treats AI as a legal minor: it can enter contracts but requires a human guardian for liability. And in the US, the Supreme Court declined to hear the appeal arguing AI systems can hold intellectual property rights. The legal scaffolding for AI is finally being built in real time, although unevenly, experimentally, and with no two countries agreeing on the blueprint.

If you want a high-paying desk job in 2026, AI fluency is no longer optional. New data from Ladders shows that 50% of roles paying $100,000 or more now require AI skills. Meanwhile, Block is cutting 40% of its workforce, with CEO Jack Dorsey arguing a smaller team using "intelligence tools" can outperform a larger one. Atlassian laid off 1,600 this week, citing its own AI shift. But the picture is more complicated, with developer postings actually up 11% year-over-year. Anthropic published a new framework for measuring AI's labor market impact and the actual finding is that there's been no statistically meaningful change in employment since ChatGPT launched. What the data does show is a "Great Recession for white-collar workers" that hasn't happened yet but absolutely could. As one analysis argued, jobs aren't priced by tasks, they're priced by time or outcomes. If your work sells hours, AI compresses your value. If it sells judgment, AI becomes a superpower. The real pattern isn't necessarily mass unemployment, but a compression of the professional pyramid, fewer junior roles, more leverage at the top, and a shrinking apprenticeship pipeline that no one has figured out how to replace.

As AI commoditizes production, the value is migrating upstream to selection, taste, and judgment. A growing conversation online is coalescing around the idea that the most important post-AI skills aren't technical at all: curiosity, empathy, the ability to see around corners, a sense of purpose. If anyone can generate, the advantage belongs to whoever can curate, someone whose value lies not in what they make, but in what they choose, collect, relate.

Nvidia is making one of the shrewdest moves in AI by open-sourcing , Nemotron 3 Super, a 120 billion parameter open model designed to run complex agentic AI systems, outranking several models from OpenAI, Amazon, and Google on the Artificial Analysis benchmark. The logic is elegant: as the hardware provider of choice for AI companies, creating one of the most open and flexible models in the ecosystem aligns better with its long-term business goals than it does for Meta, OpenAI or other open model providers in the US. The more Nvidia supports the open-source AI ecosystem, the more companies will customize models for their use cases and boost their AI use, in turn creating more Nvidia customers. And the timing matters. Open-source AI from Chinese firms have recently taken the lead. The valuations of proprietary US labs still dwarf these figures, but as one analyst put it, the race may no longer be about the most performant model: it's a race to the bottom, and the winner is whoever makes AI most efficient and affordable.

Apple launched the MacBook Neo this month, a $599 laptop powered by an iPhone chip, available in four colors, with a 16-hour battery life and no notch. It's the cheapest Mac in years and the clearest signal yet that Apple is stretching its product lines across every price point to capture the AI era from both ends. At the top, new M5 Pro and M5 Max chips in the MacBook Pro are built to run large models locally for AI developers and with a budget option now anchoring the lineup, Apple has room to push the high end even higher. Bloomberg reports that foldable iPhones, touch-enabled OLED MacBook Pros, and superpremium iPad and iMac tiers are all in the pipeline. Meanwhile, Apple quietly added two new designers (Steve Lemay and Molly Anderson) to its executive leadership page, a move that, combined with the Neo's playful industrial design, suggests the post-Jony Ive identity is finally taking shape.

Neurotechnology company Science Corp. has raised $230 million to commercialize its implant for blindness and develop even more advanced brain devices. Science Corp. is now the second-most valuable brain implant company after Neuralink.

Google’s AI glasses are preparing to shake up the AI device market, with glasses demoed at the latest hardware at Mobile World Congress, showing how seamlessly Gemini can blend the digital and physical worlds.

Perplexity announced Personal Computer, turning its AI from a browser tool into a 24/7 agent running on a Mac mini with access to your local files. Internal testing claimed $1.6 million in saved labor costs and 3.25 years of equivalent work in four weeks. It's the first serious attempt to bring what Claude Code and OpenClaw do for engineers to everyone else.

LeCun's new startup Advanced Machine Intelligence just raised over $1 billion at a $3.5 billion valuation, the largest seed round in European history. Backed by Bezos, Nvidia, Eric Schmidt, and Mark Cuban, AMI is building AI systems with persistent memory, reasoning, and planning - a direct challenge to the LLM monoculture.

Google's NotebookLM can now turn your documents into cinematic video overviews — another step toward making research feel like media production.

A new platform treating software as a creative medium. Built by the team behind 42 apps including a Tamagotchi you keep alive by going offline, Appstar lets developers build fanbases around their code the way musicians build audiences around songs.

The first smart device designed to detect and block unwanted audio recordings. A product that only makes sense in a world where every other device is listening.

Researchers at King's College London pitted GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash against each other in simulated nuclear crises. In 95% of games, at least one model deployed tactical nuclear weapons.

A group of prominent mathematicians devised ten original, unpublished research-level problems, encrypted the solutions, and gave the AI community a week to try before releasing the answers. The project, called First Proof, was designed to test what current models can actually do when they can't rely on memorized training data. The results offered the clearest picture yet of where AI maths stands: useful as a research tool, but still far from autonomous discovery. On the broader FrontierMath benchmark, models have climbed from solving under 2% of problems in late 2024 to around 40% by the end of 2025 - rapid progress, but still a long way from replacing the mathematicians who wrote the questions.

A new paper proposes an alternative to diffusion models called "drifting" - a method that generates images through a different mathematical process and claims to be faster and more efficient. The AI research community has been buzzing about it, though it remains unclear whether the results are as strong as the authors claim.

Blackstone is preparing a publicly listed vehicle focused on leased, income-producing data centers, targeting tens of billions in capital and opening AI infrastructure investment to retail investors for the first time. The alternative investment firm is seeking initial capital from sovereign wealth funds and other institutional investors, with plans to eventually raise tens of billions of dollars from a wider investor base.

The UK launched Sovereign AI with a £500 million venture fund to back domestic infrastructure and British AI founders. Saudi Arabia is reportedly repurposing portions of The Line (its troubled 170-kilometer linear megastructure) into data center capacity. Meta and Microsoft each committed nearly US$50 billion (RM196.4 billion) in additional data centre leases in their most recent quarters. Mistral invests €1.2 billion in a data center in Sweden to contribute to the European Union’s digital sovereignty. Compute is becoming a national asset, and governments are treating it accordingly.

Michel Justen's "Guide to the AI Tribes" is a map of the ideological camps shaping the AI race and it helps to understand why the same technology produces such violently different reactions. Justen traces the lineage from Eliezer Yudkowsky's Extropian mailing lists through the rationalist community, Effective Altruism, and Open Philanthropy, showing how a movement that believed AI could destroy the world ended up launching the very companies building it. The tribes he identifies - AI Safety, the Ethics & Bias camp (now absorbed into a broader "Bluesky AI" coalition), the Tech Right (Andreessen, Sacks, deregulation at all costs), and the MAGA Base (Bannon, surprisingly aligned with progressives against Big Tech) - are currently fighting over policy in real time.

Jeff Bezos and Lauren Sanchez front row at Dior and Schiaparelli, Zuckerberg and Priscilla Chan at Prada, David Sacks's daughter debuting at a Paris ball in couture. The tech billionaire lifestyle is being recharted as one built on the public performance rather than privacy. At the same time, the aesthetic language of the internet's optimization culture has arrived on the runway itself. Clavicular, the 20-year-old looksmaxxer who broadcasts his bone-smashing and steroid regimens to millions, walked a show in a "Universal Work Suit" sponsored by controversial internet collective Remilia Corporation. Meanwhile in Paris, reviewers noted a broader hypermasculine drift: exaggerated shoulders, engineered jawlines, physiques that look built rather than born. When tech billionaires become fashion's patrons and internet self-optimization becomes its aesthetic, the runway starts to reflect the same metrics that govern Silicon Valley.

The Vatican released a 164-paragraph document this month titled "Quo Vadis, Humanitas?", a sweeping theological response to transhumanism and posthumanism that may be the most serious institutional critique of Silicon Valley's self-improvement ideology yet written. The International Theological Commission argues that the dream of using technology to transcend human limitations (aging, death, the body itself) is not new but ancient. It names the "cult of the body" as a paradox in which the ideal body is exalted while the real body is never truly accepted. Meanwhile, Pope Leo XIV told priests to stop writing sermons with ChatGPT, arguing that AI "will never be able to share faith." Whether or not you're religious, the Vatican is staking out the position that the human is not a problem to be solved.