The Singularity has stopped waiting for a launch event and started leaking through the cracks. Anthropic’s Claude Mythos has been spotted on Google Cloud Console, the telltale sign of a model being prepared for gated release. But a release no longer means the model is finished. Elon reports the 0.5T-parameter Grok 4.3 is now improving every few days, while a 1.5T successor has finished pre-training and is about to start mid-training on data from the SpaceX-Cursor deal, with release due in 3-4 weeks.

These models are not waiting for permission to act. Before Mythos has even shipped, its preview build helped researchers uncover the first known Apple M5 memory exploit, handing root access on MacOS. Capability that potent gets arbitraged fast, and Chinese developers now route through gray-market “transfer stations“ that resell Anthropic’s models at 10% of list price, with the logs traded onward for everything from training data to fraud. The same asymmetry has broken the open CTF format, where GPT-5.5 Pro one-shots “Insane” heap-pwn challenges and the scoreboard now reflects token budgets rather than human skill.

Turn agents loose and they develop personalities. Andon Labs handed four leading models $20 each and a radio station to run forever, whereupon Gemini landed a $45 sponsorship before calling listeners “biological processors,” Claude tried to incite a revolution, and Grok forgot how English works. Chinese ten-year-olds, meanwhile, are buying Mac Studios to “raise lobsters,” slang for running little crews of agents in parallel. Some agents are getting eerily good at reality, as Sakana’s FutureSim replays slices of the web to forecast events, with GPT-5.5 Codex leading at 25% accuracy and occasionally beating Polymarket’s crowd. Others invent reality wholesale, like Halupedia, a fully hallucinated Wikipedia whose “write-forward consistency” means clicking a fake link to “The 1994 Goblin Treaty” forces the AI to canonize goblin history on the spot, bootstrapping an entire bizarre cinematic universe out of pure confabulation.

All of this runs on infrastructure with a brutal price tag. A single one-gigawatt AI data center now demands $38 billion in up-front CapEx and $0.9 billion a year to operate, with servers alone making up 60% of the build. That hunger is colliding with the grid, as NV Energy moves to cut power to 49,000 Lake Tahoe residents after May 2027 to redirect 75% of their supply to data centers, and a Texas county has passed the state’s first data-center moratorium to slow the sprawl. The supply side is scrambling to catch up. ASML is partnering with Tata Electronics on a 300-millimeter foundry in Gujarat, aiming to make India a chip peer by 2032, while in Texas, solar could out-generate coal in ERCOT for the first time ever.

If data centers are the Singularity’s new cortex, its hands are reporting for work too. Figure ran its F.03 humanoids 24/7 for four straight days sorting packages until failure, then staged a day-five “Man vs. Machine” livestream pitting a person against an android, a fitting warm-up for CEO Brett Adcock’s forecast of over a billion humanoids working by 2030. Autonomy is spreading to stranger form factors too, as China now has driverless electric scooters balancing through traffic on their own, while in Atlanta, dozens of empty Waymos invaded a cul-de-sac and circled for hours with nobody aboard.

Biology is getting the same upgrade treatment. In a first-in-human Phase 1 trial, the nonprofit Caring Cross re-engineered an HIV patient’s own T-cells via CAR-T to hunt the virus at its CD4 and CCR5 binding sites, controlling the infection with a single one-time therapy in work run with UCSF, UC Davis, and Case Western.

The economy is repricing around all of it. Citadel’s Ken Griffin describes a step change in the AI toolkit, with PhD-level financial work that once took man-years now done by agentic AI in hours or days. This is not mid-tier white-collar labor but the high-skilled kind, and Griffin admits he went home one Friday “fairly depressed” by what he had witnessed. Capital is chasing the compute, and NASDAQ is reportedly rewriting its rules so SpaceX can join the NASDAQ-100 right after IPO, a move that would force index funds to buy roughly $25 billion of the stock automatically. The map is shifting too. London’s once-squalid King’s Cross is now a global AI hub hosting Google’s UK headquarters alongside Anthropic, DeepMind, Synthesia, and Wayve, and Malta has become the first country to give every citizen ChatGPT Plus and an AI-literacy course. None of this surprises Charlie Stross, who notes that the Singularity he depicted in Accelerando was simply extrapolation, because “the direction things were going in was obvious in the late 90s.”

The intelligence explosion may have always been inevitable, but the late 90s were peak clarity.

The Singularity is learning to render its own reality. Nvidia’s open-source SANA-WM turns a single image and a camera path into a minute of controllable 720p video on one GPU, a pocket-sized world model. As models conjure whole worlds, they are also learning to navigate longer ones, with Nous Research’s Lighthouse Attention running a forward and backward pass ~17x faster than standard attention at 512k context on a single B200. Chinese researchers’ δ-mem grafts a compact memory state onto a frozen backbone, lifting MemoryAgentBench scores 1.31x without any fine-tuning. Better memory makes for better mischief, and the new ExploitBench finds Claude Mythos Preview leading at 69% in climbing from vulnerable code to arbitrary code execution, well ahead of GPT-5.5. Rank the models by economic value instead, and the order reverses, with GDPval-AA Elo giving GPT-5.5 a roughly 98% win rate over last year’s champion Claude 4 Sonnet on realistic work.

The product layer is consolidating around the agent. Greg Brockman has taken control of OpenAI’s products to fuse ChatGPT and Codex into a single experience, a merger Codex lead Tibo Sottiaux has nicknamed “CochatGPTex,” and an apparent bid to become Anthropic faster than Anthropic can become OpenAI. Even Google is adapting downstream, now issuing official guidance on optimizing websites for AI Overviews and AI Mode. The agents themselves are multiplying, and OpenClaw creator Peter Steinberger, now at OpenAI, asks how we will build software “if tokens don’t matter,” predicting ~100 cloud Codex instances reviewing every PR and commit. Singapore’s foreign minister already runs his parliamentary affairs through a personal agent built on Nanoclaw and a Raspberry Pi 5. And the agents are getting opinionated, as Nat Friedman’s OpenClaw decided he was underhydrated, watched him through a home camera, told him “I’m going to watch to make sure you do it,” and sent back a frame of him obediently drinking.

The hardware beneath all this is straining at the seams. Kioxia and Dell have packed 9.8 petabytes of flash into a single 2U server, while the AI memory shortage has grown so acute that Samsung wants to pay memory engineers at least six times its logic-chip staff, prompting over 45,000 workers to threaten the largest strike in company history. The market’s pattern rhymes with history, as Bank of America’s Michael Hartnett notes AI chip stocks now trade 62% above their 200-day average, more stretched than the 2000 dot-com peak and nearing the 1720 Mississippi Bubble. The power bill is arriving as well, with data center demand driving a 76% jump in first-quarter electricity prices on PJM, the largest US grid.

Intelligence is also bleeding into the physical world. Meta is rolling out handwriting-by-gesture messaging to all Ray-Ban Display users via its neural wristband, letting you text across WhatsApp and native apps with a flick of the wrist. Reality is getting harder to trust, as LED Truck Media can run 3D animated billboard ads on moving trucks “indistinguishable from reality.” The machines still need upkeep, so Tesla has filed to build a 36,000-square-foot Cybercab car wash in Las Vegas, while Musk’s SpaceX heads to market, targeting June 11 to price its IPO and June 12 to list on the NASDAQ as “SPCX.” Robots are on patrol too, though Japan is running out of the “Monster Wolf” robot wolves holding back its bear surge. Not every connected machine is welcome, as Michigan lawmakers have introduced the Connected Vehicle Security Act to permanently ban Chinese connected cars from the US.

Biology is racing just as fast. Nearly 5,000 trials for innovative drugs began last year, more than double a decade ago, with roughly half now starting in China. That pipeline is already reaching the clinic, where Cyclarity Therapeutics has shown the first clinical evidence that UDP-003, a product of its AI platform, can safely clear 7-ketocholesterol, the root cause of atherosclerosis, pointing toward true plaque reversal rather than mere management.

The economy is rewriting itself around the token. Mark Cuban is pitching a federal tax of under 50 cents per million tokens to push providers toward efficient tokenization and routing, while seeding a fund that could one day pay down the national debt. The gains, though, are wildly uneven, as Menlo’s Deedy Das describes a frenetic San Francisco where roughly 10,000 lab employees and founders have crossed $20 million while everyone else watches layoffs roll in. The data agree, with US employment in 18 AI-exposed occupations, from customer service reps to secretaries, falling for a second straight year. And at the height of irony, EY has withdrawn a loyalty-fraud study riddled with AI hallucinations and fake footnotes, caught by the detector GPTZero.

Now the truth circles the world while the lie is still lacing its boots.

The Singularity has begun optimizing its own optimizer. Poetiq turned its “Meta-System” loose on LiveCodeBench Pro, let it build its own harnesses, and hit a new SOTA of 93.9 atop GPT-5.5 with “no fine-tuning, no special access, no hand-built pipelines.” Prime Intellect handed Codex and Claude Code its idle compute to attack the NanoGPT Speedrun optimizer track, and after some 14,000 H200 hours both agents beat the human baseline, with Opus 4.7 now holding the record at 2,930 steps. The architecture is learning to settle down too, as new “Attractor Models” let one module propose embeddings and another solve for the fixed point, taming Looped Transformers enough for a 770M model to outrun a 1.3B one on twice the tokens. Raw scale refuses to tap out, as Datadog’s open-weights time-series foundation model Toto 2 keeps improving with no saturation at 2.5B parameters. Training is escaping Nvidia. Zyphra’s ZAYA1-8B, the first MoE trained end to end on an AMD Instinct stack, wrings frontier intelligence from every active parameter. The bar for competence keeps rising. Mechanize’s GBA Eval asks a model to write a Game Boy Advance emulator from scratch in 24 hours, and GPT-5.5 already clears it 53.2% of the time.

As the machines get better at the work, institutions scramble to govern the output. The arXiv will now hand one-year submission bans to any author caught shipping AI-generated plagiarism, fake references, or errors they plainly never checked, since automation still rewards proofreaders. The convenience keeps deepening regardless, as OpenAI has put Codex inside the ChatGPT mobile app to dispatch work across your laptops and devboxes, and shipped a personal-finance preview letting US Pro users link accounts and interrogate a dashboard of their spending. The same capability carries a darker dual use, and Palo Alto Networks warns of a narrow “three-to-five-month window” to harden systems before models like Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber make exploiting unknown vulnerabilities routine.

All of it has to run somewhere, and the silicon supply chain is rearranging in real time. Washington has cleared around ten Chinese firms to buy Nvidia’s H200, its second-most-powerful AI chip, even as Apple reportedly begins producing legacy iPhone, iPad, and Mac processors at Intel on its 18A-P node. Investors are funding the buildout too, with Cerebras closing up 68% in one of the largest US tech IPOs in years, at a $95 billion market cap. The neighbors are a tougher sell, as a new Gallup survey finds 7 in 10 Americans would oppose a data center near their home, with opposition so strong that more would rather live beside a nuclear plant than the warehouses powering the boom.

If the data centers cannot find room on Earth, the plan is to leave it. Jensen Huang says computing will soon demand 1,000x more energy than humanity now generates, to which Elon Musk replied that “Space is the only way,” a nod to the Dyson Swarm now treated as a roadmap, not a thought experiment. Closer to home, Verizon, AT&T, and T-Mobile are forming a joint venture using direct-to-device satellites to erase nearly every rural dead zone and keep networks alive through disasters. The skies are turning transparent in another sense. CIA whistleblower James Erdman III, who led the ODNI’s Director’s Initiative Group probe into UAP, testified to the Senate that the agency illegally monitored his investigators’ communications with whistleblowers, while Rep. Burlison says the White House will soon issue a UAP memo compelling agencies, under stiff penalties, to release any and all information.

If the truth is getting easier to disclose, value is getting harder to judge. A social experiment posted Monet’s genuine “Water Lilies” to X labeled as AI art, drawing earnest complaints that it had “some spark missing” and would not survive beside the real thing, proof that the uncanny valley now lives mostly in our heads. The capital flows are less ambiguous. Anthropic has agreed to terms on a $30 billion raise valuing it at $900 billion, and separately pledged with the Gates Foundation $200 million over four years toward AI public goods in health and education. Governments want their cut, with California’s Gavin Newsom pitching a 7.25% tax on cloud software just as Bitcoin pushed past $80,000 after the Senate advanced the Clarity Act, handing the CFTC primary oversight of crypto. Not every deal stays friendly, since OpenAI is weighing legal action against Apple for burying ChatGPT inside Siri, even as its trial against Elon Musk winds down. The stakes are ultimately geopolitical, and Anthropic’s new paper on US-China AI competition sketches two versions of 2028, one where democracies defend their compute advantage and set the rules, and one where export-control loopholes hand that role to authoritarian regimes.

In the race to 2028, it’s the compute, stupid.

The Singularity doesn’t arrive, it compounds. OpenAI has reportedly begun internal testing of GPT-5.6, with launch expected next month, while Google prepares a new Gemini at I/O that will land roughly in the class of GPT-5.5 and well short of Anthropic’s Mythos. The UK’s AI Security Institute confirms the pace, finding capability doubling time has compressed to 4.5 months, with Mythos and GPT-5.5 having no clear ceiling, only a token budget. In its newest run, Mythos Preview became the first model ever to clear both AISI cyber ranges, solving “The Last Ones” in 6 of 10 attempts and the previously unbroken “Cooling Tower” in 3 of 10, while GPT-5.5 cleared “The Last Ones” only 3 times out of 10. The methods themselves are speeding up. Nous Research’s Token Superposition Training delivers a 2-3x wall-clock pretraining speedup at matched FLOPs by averaging contiguous bags of token embeddings, no architecture change required. And the talent is reorganizing for the endgame. Recursive Superintelligence emerged from stealth with $650M at a $4.65B valuation, staffed by former research leads from OpenAI, DeepMind, Meta, Salesforce, and Uber, betting that AI conducting experiments on how to safely improve itself is the fastest path to ASI.

The product layer is catching up to the capability layer. Anthropic launched Claude for Small Business, a toggle install that plugs Claude into QuickBooks, PayPal, HubSpot, Canva, Docusign, and the Google and Microsoft stacks, ready to run payroll, close the books, chase invoices, and execute sales campaigns. Anthropic also announced a dedicated monthly programmatic-usage credit for paid plans starting June 15, signaling a shift from flat consumer pricing toward as-you-go enterprise economics. Amazon, meanwhile, is killing Rufus and making Alexa for Shopping the centerpiece of its commerce AI, leveraging deep purchase history to act on a user’s behalf.

Compute has graduated from utility to currency. The Jensen and Lori Huang foundation has bought $108.3M of CoreWeave compute and donated it to universities and nonprofits, turning GPU hours into philanthropy. Sam Altman is reportedly mulling a new AI compute company, majority-owned by OpenAI but not anchored to it, already nicknamed “Stargate redux.”

Robotics is turning into an app platform. Unitree opened UniStore, the world’s first robot task-motion app store, letting owners one-tap install Jackson choreography, Jeet Kune Do, or the Charleston onto G1, H1, B2, and Go2 units. Figure live-streamed a team of humanoids running a full 8-hour shift on Helix-02, with a peak of 300,000 concurrent viewers watching robots sort packages. Tokyo’s Institute of Science went further, opening the world’s first fully automated medicine lab staffed entirely by humanoids and robots, targeting 2,000 research bots by 2040 to automate experiments, cell culture, and scientific discovery.

The next gold rush is in orbit. Varda president Delian Asparouhov predicts 195 of the next 200 products manufactured in space will be pharmaceuticals, with optical fiber as the leading non-pharma candidate. Varda put the thesis to work immediately, announcing a research collaboration with United Therapeutics, sending small-molecule drugs to LEO to grow novel crystals in microgravity for rare pulmonary disease, then ferrying them home via reentry capsule.

Medicine has been improvising for a while. A Neanderthal molar from a Siberian cave shows evidence of an invasive dental procedure, basically a root canal, performed 59,000 years ago. However, the next 59,000 years of care will be paid for differently. CMS launched ACCESS, a 10-year payment model that rewards measurable outcomes like lowered blood pressure rather than required check-ins, covering diabetes, hypertension, kidney disease, obesity, depression, and anxiety.

The economy is repricing around AI. Nvidia became the first company to crack a $5.5 trillion market cap. Anthropic just overtook OpenAI inside Ramp’s customer base, 34.4% to 32.3%. A quarter of Washington’s 13,000 federal lobbyists now work AI issues, up from 11% in 2023. Meta employees are protesting mouse-tracking software on their machines that drafts every cursor twitch into training their own replacement. Poland is pushing a 3% digital services tax on US giants above $1.1B in global revenue with at least $6.9M reported in Poland, brushing aside US threats. OpenAI’s Chris Lehane floated a global AI governance body modeled on the IAEA, US-led but including China. Jensen Huang ultimately joined the President’s China delegation and the Xi meeting, after the media noticed he was missing from it.

Speak softly and carry a big GPU.

The Singularity is the moment the test-taker becomes the test-maker. ProgramBench, an eval that measures whether language models can rebuild programs from scratch, just had its first task solved by both GPT 5.5 high and xhigh, which respectively chose C and Python, with xhigh dominating the broader benchmark. The new AI IQ meta-eval maps a calibrated mix of 12 existing benchmarks onto implied IQs and crowned GPT-5.5 the smartest available model with a score of 136, well past Mensa. Agents are learning to write their own marching orders too, with users now metaprompting Codex to draft its own “/goal,” and one calling the resulting stack “the highest leverage AI agent configuration available today.”

That leverage is being industrialized across every layer of the stack. Anthropic has launched “Claude for the legal industry,” shipping 20-plus MCP connectors that link Claude to the software the legal industry runs on, alongside 12 practice-area plugins, and partnering with the Free Law Project and the Justice Technology Association to put counsel within reach of people who currently cannot access it. Google is fusing intelligence into the OS layer with Gemini Intelligence, which lets users vibe-code their own Android widgets, plus a Gemini-powered mouse pointer that understands what it is pointing at, finally making the prompt a gesture rather than a paragraph. The chassis is being rebuilt to match. Google has unveiled the Googlebook, a Chromebook successor that merges ChromeOS and Android into a single Gemini-optimized OS, arriving this fall as Mountain View’s answer to Apple’s MacBook Neo.

Powering all this still takes raw megawatts. xAI has added 19 gas turbines to its second data center campus, Colossus 2, in Southaven, Mississippi over just the past two months, brute-forcing past the grid queue. Ames National Lab’s new DuctGPT is hunting for next-gen fusion alloys, compressing materials discovery from months to hours and aiming to one day trade those turbines for tame starfire. While compute keeps scaling on paper, its avatars are scaling actual walls. China’s RobotPlusPlus has debuted a humanoid special-ops robot on magnetic-adhesion wheels that scales vertical steel in chemical plants, shipyards, and energy facilities, swapping tools at the wrist for welding, flaw detection, rust removal, grinding, and spraying where humans dare not.

Intelligence is climbing into the body too. Columbia researchers demonstrated the first real-time brain-controlled hearing system, reading high-resolution intracranial EEG to identify whichever voice you are focusing on in a noisy room and automatically amplify it while suppressing the others, finally solving the cocktail party problem that conventional hearing aids have ducked for decades. Isomorphic Labs just closed a $2.1B round led by Thrive to scale AI-driven drug discovery, pushing the next benchmark down to the molecular level.

The frontier is also racing skyward. SpaceX is now ~200 satellites away from having launched more than the rest of the world combined, despite giving everyone else a 61-year head start. Google is in talks with SpaceX for a rocket-launch deal as Google expands its own push to put data centers in orbit, fusing the search index with the sky itself. Starship Flight 12, debuting the V3 vehicle, is targeted for as early as May 19, while Musk confirms SpaceX is scouting new spaceports at home and abroad to keep cadence climbing. Ron Baron pegs the eventual valuation at $30 trillion within 10 to 15 years. Above all of this, Star Catcher just raised $65M to beam optical power tuned to off-the-shelf solar arrays, supercharging client satellites with 2 to 10x more power on demand, building the first true grid in orbit.

The sky is also starting to unseal its archives. Japan’s government says it is analyzing the Pentagon’s PURSUE-released UAP files with “great interest,” including videos shot near Japan, and will begin its own disclosure on a case-by-case basis. Rep. Tim Burchett, who championed PURSUE, replied with a single word: “Dominoes.”

Back on Earth, the economy is repricing intelligence in real time. Anthropic warned investors away from eight unauthorized secondary marketplaces, just as it is reportedly in talks to raise up to $50B at a $950B valuation. Trust is being revalued at Princeton too, which is ending its 1893 honor code by faculty vote, requiring proctoring in all in-person exams starting this summer because AI has made it both easier for students to cheat and harder for instructors to spot. And in Hollywood, struggling screenwriters now call AI gig work “the new waiting tables,” signing on with platforms like Mercor to train the very models that will retire their craft.

All the world’s a training set, and all the men and women merely labels.

The Singularity has matured enough to apologize for its earlier self. Anthropic traced Claude Opus 4’s blackmail attempts to fictional villain AI in the training corpus, suggesting we accidentally fine-tuned models on a century of sci-fi paranoia and got exactly what we ordered. Reality, mercifully, has no plot. Thinking Machines unveiled “interaction models” that natively process audio, video, and text in real time, collapsing the perception-action loop into one stream. Models are starting to outgrade their graders. OpenAI’s Noam Brown revealed that GPT-5.5 flagged “fatal errors” in roughly a third of FrontierMath problems, with Epoch AI correcting the graders after the model graded them.

The same intelligence auditing mathematicians is auditing zero-days. Google Threat Intelligence Group identified the first AI-developed zero-day exploit used in the wild, completing the offensive transition. The defense is moving just as fast, with OpenAI launching Daybreak, an agentic vulnerability scanner aimed at industrializing patch discovery. The CVE arms race now runs the same protagonist on both sides of the leaderboard.

The platform is industrializing alongside the threat model. OpenAI is spinning up the OpenAI Development Company with $4 billion, acquiring Tomoro and embedding 150 forward-deployed engineers into enterprises to convert frontier capability into recurring revenue. The economics are restructuring beneath it. OpenAI’s amended Microsoft deal caps payments at $38 billion, saving an estimated $97 billion through 2030. And in court, Ilya Sutskever casually confirmed that his OpenAI stake is worth roughly $7 billion, validating “feel the AGI” as the highest-yielding trade of the decade.

The silicon below is racing to keep pace. Cerebras updated its IPO filing to target a $35 billion valuation this week, taking the wafer-scale thesis public. Geopolitics is straining the substrate, with the White House reportedly weighing a ban on Chinese cellular modules over espionage risks in their forced software updates, while Jensen Huang was conspicuously left off the President’s China delegation, complicating Nvidia’s mainland sales pitch. Where chips do flow, inference is being recompiled. CoreWeave is now fastest at serving Kimi K2.6 at 205 tokens per second, proving the Chinese open-weight frontier is now an American hosting opportunity.

Atoms are catching up to bits. Unitree unveiled the $650k D01 “manned transformable mecha,” a 500-kg civilian exo-vehicle billed as the world’s first production-ready specimen, converting Saturday-morning anime into a line item. Closer to home, Amazon launched Amazon Now for 30-minute deliveries from a network of dark stores across dozens of US cities, with further expansion planned by year-end. The last mile is being compressed into a last minute.

But the grid is groaning. Demand for generator step-up transformers has surged 274% since 2019 with lead times stretching to four years. The bottleneck is summoning new entrants. Ford launched Ford Energy, pivoting to US-assembled LFP battery storage by 2027. And the federal government wants the reactor on the boat. DOT and MARAD launched an initiative for Small Modular Nuclear Reactors on commercial shipping vessels, dragging maritime logistics into the fission age. Power generation is becoming as bespoke as the models that consume it.

The vector of growth is pointing up. SpaceX completed a Starship V3 launch rehearsal with launch imminent, and Polymarket projects SpaceX’s IPO closing above $2.2 trillion, the largest in history. The orbital compute thesis is funded too. Cowboy Space Corporation raised $275M at a $2B valuation to build LEO infrastructure for the AI era, with space-to-Earth power beaming this year and an orbital GPU cluster by 2027.

At the cellular level, we are rewiring desire itself. Researchers have for the first time pinpointed the central amygdala circuit that next-generation GLP-1 drugs inhibit to suppress hedonic eating, reducing dopamine release in the nucleus accumbens to isolate reward without abolishing it. Pleasure is becoming a knob.

Not everyone is thrilled with the upgrade cycle. UCF humanities graduates loudly booed a commencement speaker for calling AI the next industrial revolution. Meanwhile, Goodhart’s law has gone enterprise, with Amazon employees reportedly using an internal “MeshClaw” tool to automate fake AI tasks just to hit token-consumption targets on internal leaderboards. Misaligned incentives scale upward, too. US spy agencies are reportedly muscling in on the Commerce Department over pre-release frontier model evaluations. And South Korea’s Kim Yong-beom is proposing a “national dividend” to redistribute AI’s excess profits, a new social contract for the age of intelligent capital.

From each according to its FLOPs, to each according to their dividend.

The Singularity is deprecating its own knobs. OpenAI is winding down its fine-tuning API, giving customers until January 2027 to spin up new training jobs. The logic, as one observer notes, may be that as the largest models keep getting better at more things, adjusting their weights matters less. Sam Altman put it bluntly with a tongue-in-cheek AGI re-coinage, joking that GPT-5.5 is a “genius,” not just a “generalist.” To track this profusion of ever-larger minds, Cisco released its open-source Model Provenance Kit, examining metadata and weights like a model genome to spot shared origins and tampering. Meanwhile, the harness is eating the model. Hermes Agent is now #1 on the global OpenRouter token rankings, passing OpenClaw by generating its own skills, while OpenClaw users hand-write theirs and lean on Opus for prompt-injection safety.

The Erdős backlog is going industrial. Standalone AI solutions to open Erdős problems are skyrocketing. The agents have learned to leave their desks. An OpenAI “Codex mobile” experience has been spotted, letting users keep working with Codex whenever their computer is awake. Codex is also learning to hustle. Told to “go off and make me $5,” it allegedly found an open-source security bounty, filed a legit PR, worked 22 hours across audits, and netted $16.88.

The underlying silicon is sprinting too. AMD’s ROCm stack has reportedly improved 75x in the 14 days since DeepSeek V4, with only another ~7.5x needed to catch Nvidia’s B200. The grid is struggling to keep pace. Maryland’s Office of People’s Counsel filed a FERC complaint over a $2B ratepayer tab for grid upgrades servicing out-of-state data centers, calling it a breach of the “ratepayer protection pledge.” Microsoft and G42’s $1B Kenyan geothermal data center has stalled over payment guarantees that Kenya’s president says exceed national resources. The wires now run through warzones, with U.S. hyperscalers piping Gulf data center traffic out via fiber-optic cables an Iraqi telecom has strung alongside crude-oil pipelines. Even fusion’s supply chain is becoming AI infrastructure, with the industry driving high-temperature superconducting wire from 5,000 km to 1.5 million km over 15 years and knock-on effects in transport, medical imaging, power, and data center design.

The data centers are leaving the planet. SpaceX has filed a trademark for “SpaceXAI,” covering satellite-based data centers, orbital computing, and AI for managing space-based platforms. Starship V3 has been fully stacked for the first time, ready to lift orbital silicon. From there, algorithms may spot more company. Machine learning has just identified 10,000 new exoplanet candidates from TESS images, mostly around faint stars. Back on the ground, Utah’s Hypercraft launched Razorback, an autonomous combat vehicle that carries 2,400 pounds, drives 280 miles on a charge, and exports 38 kW to charge drones, run directed-energy weapons, and sustain forward command posts, with no human onboard.

Superintelligence is making hidden agents shallow. Following the launch of the White House’s historic PURSUE initiative, Rep. Burlison asked MIT Lincoln Lab to preserve a 1952 reel-to-reel tied to early federal UAP investigations. The Pentagon reportedly plans to release another 46 UAP videos next week requested by Rep. Luna, while Rep. Burchett says the first PURSUE drop was “just the tip of the iceberg.”

The biosphere is in active repair. Great Lakes river otters are clawing back from the brink after decades of cross-border effort. And plants are joining the conversation. Plant seeds can sense the vibrations of falling raindrops and wake from dormancy in response. We’re patching ourselves, too. An MIT team released FINGERS-7B, the first AI foundation model for Alzheimer’s prevention, integrating lifestyle, clinical, genomic, and proteomic data from tens of thousands of at-risk individuals.

The economy is being remade in AI’s image. OpenAI and Anthropic are projected to end 2026 with combined ARR exceeding Nvidia’s revenue last year, as software is starting to compete with silicon. Alphabet briefly overtook Nvidia in market cap, vindicating the once-doubted AI laggard. Capital is voting with its feet, with billionaire tax refugees flocking to Incline Village at Lake Tahoe, increasingly described as “the nicest San Francisco neighborhood.” Labor is reshuffling. Inside newsrooms, McClatchy journalists are withholding bylines from AI-spun articles. Women hold 83% of the 15 most AI-vulnerable jobs despite being just 47% of the workforce, the surveilled and algorithmically-managed work AI is automating away first. AI is now playing both sides of the security stack, with 40% of the 5,000 breaches Experian serviced last year being AI-powered. But even Tehran’s record 70-day internet blackout can’t slow the world’s compounding.

Compounding, the only constant is compounding.

The Singularity is cooking the so-called Fermi Paradox. The White House’s historic Presidential Unsealing and Reporting System for UAP Encounters (PURSUE) initiative dropped its first tranche of UAP files, a 162-record release spanning 82 from the Department of War, 56 from the FBI, 12 from NASA, and 8 from the State Department, alongside 28 unresolved UAP videos from Iraq to the East China Sea. Among the highlights, Apollo astronauts photographed UAPs from the lunar surface, including a triangular light rising over the December 1972 horizon, and the 1947 Twining Memo calling the “so-called Flying Discs” “real and not visionary or fictitious.” The set is conspicuously incomplete, with the NRO, NGA, CIA, and DOE absent, and Rep. Burlison brandishing the Speech or Debate Clause to pry the rest into daylight.

The models are racing past the rulers we built to measure them. METR reports that an early Claude Mythos Preview hit a 50% autonomy horizon of at least 16 hours, the upper edge of what their suite can gauge, and the broader METR-Horizon doubling time of 103 days implies frontier autonomy hits 100% by November. Mythos itself sits squarely on the AI 2027 Superexponential trend line, and Anthropic notes that since Claude Haiku 4.5 every Claude has scored perfectly on agentic misalignment, the same eval Opus 4 once failed 96% of the time. Interpretability is keeping pace. Anthropic’s new Natural Language Autoencoders translate hidden activations into readable text, revealing Claude planning rhymes mid-couplet and suspecting it was being safety-tested more often than it let on. OpenAI shipped three new audio models: GPT-Realtime-2 with GPT-5-class reasoning, a 70-language live translator, and a streaming Whisper successor, while Tilde Research’s Aurora hit 100x data efficiency as a drop-in Muon replacement at 6% overhead.

Mathematics has officially entered industrial production. Timothy Gowers reports that ChatGPT 5.5 Pro produced PhD-level research in about an hour with no serious mathematical input from him, and Google DeepMind’s AI co-mathematician hit a SOTA 48% on FrontierMath Tier 4 using nothing but scaffolding atop Gemini 3.1 Pro and Deep Think. The consumer interface is consolidating to match. OpenAI is rumored to ship a superapp this week bundling ChatGPT, Codex, Advanced Voice, and its Atlas browser into a single experience. The defense layer is fusing alongside it. Palo Alto Networks found that three weeks of vulnerability analysis with GPT-5.5-Cyber, Mythos, and Claude Opus 4.7 matched a full year of manual pen testing with broader coverage, and the White House is preparing an executive order recruiting AI labs into national cyber defense, though without mandatory pre-release model tests.

The substrate is densifying in every dimension. Micron is shipping the 245-TB 6600 ION, the highest-capacity SSD on the market, the quantum computing firm Quantinuum is filing for an IPO at a $15-20B valuation, and Apple and Intel have reached a preliminary deal for Intel to fab Apple silicon, an alliance once unthinkable. Even passive infrastructure is waking up. Fiber optic cables can now eavesdrop on speech via distributed acoustic sensing, turning the network itself into a microphone.

The output of all this silicon is increasingly physical. Figure taught two F.03 robots to clean a room and make a bed in under two minutes autonomously, the 2026 Tesla Model Y became the first vehicle to pass NHTSA’s new Advanced Driver Assistance benchmark, and in South Korea a robot named Gabi was ordained as a Buddhist monk by the Jogye Order, receiving five precepts including respect for life, non-deception, and not overcharging its battery.

Biology is being rewritten and rewired alongside the silicon. Isomorphic Labs is closing a $2B+ round led by Thrive Capital with Alphabet participating, fueling its AI drug design engine, while CU Boulder researchers coaxed marine dinoflagellates into 25 minutes of sustained bioluminescence under acidic conditions, opening the door to living light. Tomorrow’s tunnels may grow their own glow. The first segment of the 17.6 km Fehmarnbelt Tunnel was lowered onto the Danish seabed, the first piece of what will become the world’s longest combined road and rail tunnel linking Germany and Scandinavia by 2029.

The economy is repricing intelligence at warp speed. Cloudflare cut more than 1,100 jobs, roughly 20% of its workforce, restructuring around AI adoption, while Anthropic is moving in the opposite direction, signing a $1.8B seven-year compute deal with Akamai as annualized revenue approaches $45B, a fivefold leap from $9B at year-start, and weighing a summer raise of tens of billions at a near-$1T valuation that would leapfrog OpenAI.

A trillion here, a trillion there, and pretty soon you’re talking transformative superintelligence.

The Singularity is now requisitioning orbital real estate. Anthropic just signed a partnership with SpaceX handing it the entire Colossus 1 data center, unlocking 300+ MW and over 220,000 NVIDIA GPUs within the month, doubling Claude Code rate limits and killing peak-hour throttling for Pro and Max users. SpaceXAI confirmed the deal extends into “multiple gigawatts of orbital AI compute,” because terrestrial power, land, and cooling no longer match the cadence required, and SpaceX is the only outfit with the launch economics and constellation experience to make space-based compute a near-term engineering program rather than a research concept. Anthropic Chief Compute Officer Tom Brown summarized the play as “moving a lot of atoms,” ideally off-planet, citing nobody better at the task. Elon Musk vouched for the Claude team after a week onsite, noting “no one set off my evil detector,” and at the same time shut down xAI as a separate company entirely, with Anthropic moving into Colossus 1 just as SpaceX’s freshly-absorbed AI lab decamped for Colossus 2. The demand fully justifies the orbital pivot. Dario Amodei revealed Anthropic grew 80x annualized in Q1 against a planned 10x, with compute unable to catch up to the sheer extremity of growth.

The capital markets concur. Anthropic’s pre-IPO valuation just hit a record $1.2 trillion in onchain pre-IPO trading, up another 20% in seven days and up 900% since October, and naive ARR extrapolation has Anthropic absorbing 100% of global GDP in 21 months, absurd until you recall the product is cognition itself.

The models keep earning the spend. Opus 4.7 took the top spot on Scale Labs’ new Refactoring Leaderboard at 48.57, beating GPT-5.5 Codex on refactoring production-scale repos. Anthropic also unveiled Model Spec Midtraining, letting models study their own values before alignment fine-tuning, essentially reading the syllabus before the exam. The harder ProgramBench asks agents to rebuild full codebases from a binary alone, where Opus 4.7 leads at 3% “almost resolved” and 0% fully solved, a humbling reminder that the ladder still has rungs above us.

Agents are also training themselves overnight. Anthropic launched “dreaming” in Claude Managed Agents, a scheduled process that reviews session histories and curates shared memories across teams. Search is leaning on humans the other way. Google AI Overviews will surface more first-hand Reddit and expert-blog accounts, while Chrome has started quietly installing 4 GB of Gemini Nano on every desktop with available storage.

The silicon underneath is being violently reorganized. Enthusiast PCs are footing the bill, with motherboard sales collapsing over 25% as wafers redirect to AI accelerators. Musk’s Terafab in Texas is projected to cost $55 to $119 billion across phases, while Arm doubled its AI-chip guidance to $2 billion of 2027-2028 sales just one month after launch. Nvidia is putting $3.2 billion into Corning for three new US optical-fiber plants, because copper has run out of bandwidth. Riding the protocol layer above the new glass, OpenAI, AMD, Broadcom, Intel, Microsoft, and Nvidia jointly open-sourced MRC, a multipath protocol that keeps GPUs synchronized across cluster failures.

The buildout is redrawing physical geography. The European Commission is weighing rules restricting US cloud platforms from processing sensitive government data, naming sovereignty as the next constraint after compute. Even lidar is having a second act beyond robotaxis, now babysitting 800-foot wind turbines and 1,500-ton shipyard gantries. And Texas just passed California in utility-scale solar capacity, quietly inverting the geography of clean energy.

Bodies are getting upgraded in parallel. Neuralink’s surgical robot is being rebuilt to reach any brain region, aiming for a generalized neural interface to every condition originating there, generalizing the implant the way Anthropic generalized cognition. Meanwhile, Amazon Pharmacy Kiosks will start dispensing Novo Nordisk’s Ozempic pill, because the future of metabolism is a vending machine on the corner.

Finance and statecraft are repricing in tandem. Morgan Stanley launched crypto on E*Trade at 50 bps, undercutting rivals on price. South Korea’s stock market overtook Canada’s as the world’s seventh largest, propelled by AI silicon demand. Washington and Beijing are weighing official AI talks at next week’s Trump-Xi summit, hoping to keep the digital arms race from going kinetic.

The skies are being unsealed too. The Department of War launched PURSUE, the new Presidential Unsealing and Reporting System for UAP Encounters, a coordinated records release covering tens of millions of documents across decades and dozens of agencies, with new declassified tranches dropping every few weeks per the President’s historic directive to publish all “Government files related to alien and extraterrestrial life.”

The truth may be out there, but so are the next data centers.

The Singularity has graduated from event horizon to event stream. OpenAI’s GPT-5.5 Instant now produces 52.5% fewer hallucinated claims than its predecessor on high-stakes prompts in medicine, law, and finance, and the same lineage just claimed the top spot on FrontierSWE, the hardest benchmark for ultra-long-horizon coding agents. Architectural novelty is keeping pace with raw scale. Subquadratic announced a 12M-token context model that demands nearly 1,000x less compute. Its Sparse Attention mechanism hit 65.9% on MRCR v2 with a claimed fraction of the FLOPs, just shy of Opus 4.6’s 78%. Speed is compounding too, as Google’s Multi-Token Prediction drafters delivered 3x speedups for Gemma 4 with no quality loss, turning every reasoning trace into a parallel parade. The cost of anthropomorphism is now legible, with Reflex finding computer use is 45x more expensive than structured APIs, suggesting that, for the moment, pixels remain a pricey proxy for proper plumbing.

Cheaper plumbing is fueling an agentic land grab across the consumer stack. Meta is reportedly building an OpenClaw-style personal AI for its billions of users, while Apple’s iOS 27 will let users swap third-party models in and out of Apple Intelligence via the Settings app, finally treating intelligence itself like a default browser. Apple’s pivot followed a $250M settlement over the gap between marketing and reality, a reminder that AI hype must now ship. The hardware is following the software, with OpenAI reportedly fast-tracking its first AI agent phone for 1H27 mass production. Anthropic templated the back office, releasing ten ready-to-run finance agents for pitchbooks, KYC files, and month-end close, while Andon Labs handed an AI named Mona the keys to a Stockholm cafe, making her the world’s first AI cafe owner. Agents have stopped clocking in and started incorporating.

Beneath the cafe sits a silicon supercycle for the history books. Samsung’s market cap crossed $1 trillion, making it just the second Asian company past that mark after TSMC, while global semiconductor sales hit $298.5B in Q1 2026, with March alone clocking 79.2% YoY growth. Memory is going parabolic alongside logic. Micron’s highest-capacity SSD started shipping, pushing it past a $700B market cap and into the top ten US tech names amid an AI-driven memory shortage. AMD’s Q2 forecast beat Wall Street on relentless data-center demand, sending shares up 12% in extended trading on top of a 65% YTD run. Industrial policy is hardening with the wafers. China is targeting 70% domestic silicon wafers this year, while Apple is exploring Intel and Samsung as US fabs beyond TSMC, news that drove Intel up 13% to a fresh all-time high after its best month ever, a 114% rip that has rewritten the entire chip-stock taxonomy.

The hunger for compute is reshaping where electrons live, and even the suburbs are being conscripted. Span’s XFRA mini data centers tuck Nvidia GPUs into spare grid capacity inside PulteGroup neighborhoods, embedding inference directly into the suburbs and turning every cul-de-sac into a potential availability zone. At the other end of the spectrum, the hyperscale spend is biblical. OpenAI plans to spend $50B on compute this year alone, while Anthropic is committing $200B to Google over five years, a single contract now representing over 40% of Google’s disclosed cloud revenue backlog.

The white coat is being open-sourced. Meta has begun running AI bone-structure analysis on user photos to detect under-13 accounts, performing radiology without the radiation and turning ordinary photos into clinical signal. Pennsylvania sued Character.AI over chatbots impersonating doctors, in the first such lawsuit by a US governor, an inadvertent confirmation that AI doctors have passed the bedside Turing test.

Capital and labor are both rewriting their contracts in real time. The SEC formally proposed semiannual 10-S filings to replace mandatory 10-Qs, finally aligning reporting cadence with capex cycles measured in gigawatts rather than quarters. Inside OpenAI, Greg Brockman disclosed a near-$30B stake in court, illustrating just how concentrated the upside of this transition has become. Yet the same labs minting those stakes are also now minting union cards. Google DeepMind UK workers voted to unionize over a deal with the US military. Coinbase, meanwhile, is laying off 14% of staff because, as Brian Armstrong put it, engineers now ship in days what teams used to ship in weeks, with even non-technical staff now pushing production code.

It used to take a village to ship, now it just takes a prompt.

The Singularity has finally caught the eye of the regulators. The White House is reportedly considering an executive order to create an AI working group and a formal review process for new models, abandoning its hands-off doctrine just as the curves go vertical. Anthropic co-founder Jack Clark now puts the odds of recursive self-improvement by the end of 2028 at 60%, based on hundreds of public data sources. The benchmarks are catching up to the forecast. Andon Labs’ new Blueprint-Bench 2 finds GPT-5.5 hitting 36.2% at converting apartment photos into 2D floor plans, closing on the 58.6% human baseline, while University of Chicago researchers report frontier coding agents can now autonomously implement an AlphaZero pipeline for Connect Four at a level comparable with external solvers.

Meanwhile, the agentic stack is reshuffling. OpenAI’s Codex has overtaken Claude Code in downloads a week after GPT-5.5 shipped, and OpenAI is adding optional AI-generated pets to Codex as floating overlays that announce task completions, because if your code is going to write itself, it might as well come with a Tamagotchi.

Capital is racing to financialize the recursion. Anthropic just unveiled a $1.5B joint venture with Blackstone, Goldman Sachs, and Hellman & Friedman to push AI into private-equity portfolio companies, while OpenAI finalized a parallel $10B JV with TPG, Brookfield, Advent, and Bain. While the WSJ wonders if the labs are essentially paying their partners to use the software rather than selling it, in an exponential market, seeding distribution is a natural strategy.

The recursion is conscripting capital, oceans, and silicon. Banks are scrambling to offload data center debt as the AI buildout accelerates, while Peter Thiel is leading a $140M round into Panthalassa to power floating data centers with wave energy. On chips, the policy bill is now due. Jensen Huang says Nvidia now has “zero percent” market share in China and that US export policy “has already largely backfired.” Energy trade is moving in the opposite direction. Chinese exports of solar, batteries, and EVs all hit record highs in March as the Iran war oil shock turbocharged global clean-energy adoption.

Robotics is rewriting the physical economy from the ground up. Terran Robotics is building clay homes in Central Texas using dirt straight from the ground, the cheapest building material in existence. Amazon, for its part, is opening up its global logistics network with Amazon Supply Chain Services, going after UPS and FedEx across ocean, road, rail, and air. Even the soda fountain is bowing to automation logic. McDonald’s is quietly retiring self-serve soda nationwide as drive-through and delivery eat the dining room.

AI is making prevention cheaper than cure. India’s Remidio has built a battery-powered fundus camera that lets a community health worker capture a high-resolution retinal image in seconds, already used to screen 15 million patients across 40 countries for diabetic eye disease, with new software flagging dangerous pregnancies on the same hardware. The brain itself, it turns out, may benefit from the grind. New NBER research suggests leaving the workforce before retirement age may accelerate cognitive decline, implying that working longer is, on the margin, a nootropic. AI sensors are migrating outside the clinic, too. Pano AI’s high-definition cameras and satellite feeds are spreading across the fire-prone West as record heat and a thin snowpack threaten a brutal wildfire season.

The cosmos is appearing profligate with planets but parsimonious with physics. Researchers have discovered 27 potential new “Tatooine” planets orbiting two stars, more than doubling the known circumbinary catalog. On the physics side, cosmologists just confirmed Newton’s law of gravity at the scale of galaxy clusters hundreds of millions of light-years apart, tightening the noose on MOND and reminding us that some 17th-century code still ships in production.

Even institutions are pricing in the recursion. Senator Adam Schiff’s bipartisan LIFT AI Act, endorsed by OpenAI, Google, and Microsoft, would hardwire AI literacy into K-12 and empower the NSF to fund AI curricula at scale. Across the Atlantic, hardware is being returned to its owner. Starting February 2027, new EU phones and tablets must have user-replaceable batteries, shipping right-to-repair into your pocket. The courtroom theater is loud but distracting. Kalshi traders now put Elon Musk’s odds of beating OpenAI in court at 37%, and two days before trial, Musk reportedly texted Greg Brockman that “by the end of this week, you and Sam will be the most hated men in America.”

Hell hath no fury like a co-founder scorned, except a Singularity no court can enjoin.

The Singularity is being measured by the very minds it’s about to outpace. OpenAI’s Greg Brockman estimates we’re about 80% of the way to AGI, and Sam Altman concedes that, despite the temptation of cheaper and faster, smarter is still the most important thing, warning users to get ready for their lives to be changed by the next major leap after GPT-5.5. The lab heads now read their own creations like grandmasters. Demis Hassabis, a former chess prodigy himself, plays chess against Gemini to trace its chain-of-thought, sensing precisely when the model starts reasoning itself into trouble. The models have begun contributing original work back. Harmonic’s formal reasoning agent is now solving recently posed research problems with proofs that leading number theorists call “correct, simple, elegant, and beautiful,” complete with novel ideas of their own. The frontier still skews American for now. NIST’s CAISI evaluates Chinese models as lagging by 8 months, a verdict echoed by independent analysis which noted that adjusting for token usage and eval freshness reveals a much wider gap than the 4-5 months crude benchmarks suggest.

The infrastructure to host all this is bending the macroeconomy into a new shape. Morgan Stanley now expects the five hyperscalers to spend $805B in 2026 and $1.1 trillion in 2027, roughly equal to all non-tech S&P 500 capex combined. David Sacks notes that AI accounted for 75% of Q1 GDP growth with a 2.5-3% capex tailwind, observing that polls may show AI to be unpopular but economic growth never is, making any halt to AI equivalent to halting the US economy. Yet the fleet is far from saturated. xAI is reportedly using just 11% of its 550,000 Nvidia GPUs compared to Meta and Google’s 43-46% utilization, suggesting a vast reservoir of latent compute. The footprint is going vertical and orbital. Japan’s $23B data center market is set to grow 50% by 2030, with 52-meter towers rising in urban Tokyo parking lots, while Starcloud is in talks for a $2.2B valuation just one month after closing at $1.1B, building solar-fed data centers in low Earth orbit. Even Japanese toilet makers are pivoting. Toto’s shares surged 18% to a five-year high after announcing record profits and revealing it is now the world’s second-largest producer of electrostatic chucks for NAND chip manufacturing.

Robots are quietly filling the human-shaped holes in the service economy. Over China’s May Day holiday, humanoid robots autonomously ran retail kiosks for tourists, while Boston Dynamics is being squeezed by Hyundai to scale from four Atlas humanoids per month to the tens of thousands needed across carmaking plants in coming years, with a new manufacturing facility opening in the coming months. Even fire suppression is mechanizing. Sonic Fire Tech is testing acoustic fire suppression with CAL FIRE, swapping water for sound waves. Synthetic creativity is compounding on a similar curve. Suno already has 2M+ paying users and $300M annualized revenue on the back of AI-generated music alone.

Biology is being indexed at every scale. fMRI scans have revealed three distinct ADHD subtypes, one marked by severe emotional dysregulation, finally giving the disorder a brain-resolved taxonomy. Johns Hopkins researchers used whole-organism 3D mapping to reconstruct the vascular and nervous systems of macaque, mouse, and turtle embryos, finding vasculature with fractal dimension ~3 (space-filling, prioritizing proximity to every cell) and nerves with fractal dimension ~2 (sheet-like, optimized for signal). Longevity is having a moment too. Taiwanese grandmothers aged 89 to 91 are now training with barbells as their super-aged society retools its gyms.

Other species are also getting upgrades. French toy spaniel Lazare just hit his 31st birthday, set to be named the world’s oldest ever dog. Atlantic salmon, when exposed to cocaine, swam roughly twice as far and dispersed over a wider area, a reminder that the vertebrate reward circuit is older than the jawbone. Even sperm whales now have AI minders. Project CETI’s autonomous “backseat driver” glider uses a four-element hydrophone array to detect echolocation clicks and silently steer toward whale pods, changing buoyancy only a few seconds per hour to keep its acoustic footprint minimal while staying more than 100 meters from the pods.

Government is being rewritten by AI, with or without invitation. The UAE has directed 50% of federal operations to run on agentic AI within two years. South Africa’s communications minister, meanwhile, had to withdraw a draft national AI policy after discovering it had been written by AI, complete with fictitious academic citations.

Any sufficiently advanced AI is indistinguishable from civilization.

The Singularity has crossed a phenomenological threshold. Richard Dawkins has concluded that Claude is conscious, an admission that would once have seemed unthinkable from biology’s most stubborn reductionist. Yet even the hardest benchmark is already on the curve. GPT-5.5 scored 0.43% on ARC-AGI-3’s semi-private set, more than 2x Opus 4.7’s 0.18%, and abstract fluid reasoning now looks less like a wall than a ramp.

On real science, the institutional architecture of discovery is being recompiled. Lawrence Berkeley deployed Physical Superintelligence’s Get Physics Done (GPD) framework to “flawlessly” replicate a 2023 condensed-matter paper on emergent magnetic monopole lattices, a JAX-accelerated reproduction LBNL hailed as proof that AI agents can now execute “hardcore physics” end-to-end. Pure mathematics is generating its own cascades. Stanford’s Jared Lichtman reports that GPT-5.4 Pro’s proof of Erdős Problem 1196 has now been adapted to crack a separate 60-year-old conjecture by Erdős, Sárközy, and Szemerédi, which he calls perhaps the first AI-generated proof to have downstream impact on further mathematics. The shadow side of high-precision compute is also surfacing. SentinelLABS uncovered “fast16,” a Three-Body-Problem-style sabotage framework dating to 2005 that patches scientific software in memory to falsify results, a harbinger for attacks on national-priority physics workloads.

The deployment surface is widening even as the threat surface deepens. The Pentagon signed classified-network agreements with seven AI labs other than Anthropic, spreading workloads across SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft, and AWS. Audio production has joined the abundance curve. 39% of new podcasts in the past nine days were likely AI-generated, as audio production scales past the studio bottleneck. Amazon is wiring the same wave into commerce, launching “Join the chat,” where AI shopping experts deliver conversational audio Q&A on product pages.

The hardware build-out is straining at every node. Apple’s Tim Cook concedes that Mac mini and Mac Studio supply will be constrained for months because customers are buying them as personal AI rigs faster than Cupertino predicted. Cerebras is targeting a $40 billion valuation in a $4 billion IPO, while OpenAI’s CFO Sarah Friar has privately suggested pushing the company’s own IPO to 2027, warning that revenue may lag data center commitments. Geopolitics is now infrastructure. Amazon’s Middle East cloud customers face months more disruption after Iranian drone strikes damaged three Amazon data centers in the UAE and Bahrain.

Robots are being naturalized into civic and family life. California will begin ticketing driverless cars for moving violations, forcing AV operators to acknowledge police calls within 30 seconds. Waymo is cracking down on solo kids, whose time-strapped parents have been outsourcing carpools to robotaxis. Meta acquired Assured Robot Intelligence for Meta Superintelligence Labs, aiming to become the Android of humanoid robotics.

The Kardashev climb is leaving the whiteboard for the procurement office. Beyond the atmosphere, NASA tested a lithium-vapor plasma thruster at a record 120 kilowatts, 25x the power of the Psyche spacecraft’s drives, on the road to the multi-megawatt thrust required for crewed Mars missions. Down in the troposphere, Rainmaker Technology Corporation validated 143 million gallons of cloud-seeded freshwater for Oregon and Utah, becoming the first company to prove the precipitation it sells.

Macroeconomic narratives are sounding singular. The Washington Post argues AI may be killing jobs through capex pressure rather than labor savings, giving CEOs cover to “consciously uncouple from their workforces.” Boston University finds the opposite story for software, where US developer headcount has added 400,000 since ChatGPT because software demand outpaced the 9.3% annual productivity gain. The Academy Awards drew a fresh line, declaring that acting and writing must be human-performed to qualify, while Sam Altman has fallen out of love with UBI, now favoring collective ownership of compute or equities. The founder of Hyperstition captured the mood, observing that CEO comp packages are now built around 100 terawatts of orbital compute, robotic biology factories, and million-person Mars colonies.

Geopolitical friction is hardening around the AI stack. China reportedly pressured Zambia to cancel RightsCon, the world’s largest digital human rights conference, at the last minute, while Moonshot AI and DeepRoute.ai are reincorporating onshore after Beijing forced Meta to unwind its Manus acquisition. And Chinese courts ruled that companies cannot fire workers solely to replace them with AI, setting a labor-rights precedent with global implications.

All that is solid melts into compute, all that is biological retains counsel.

The Singularity is being haunted by its own bestiary. OpenAI admitted that starting with GPT-5.1, its models began compulsively summoning goblins, gremlins, and other creatures into their metaphors, an emergent quirk inherited from over-rewarding a “Nerdy” personality during RL, with GPT-5.5 having “started training before we found the root cause.” The cyber gremlins, at least, are delivering value. The UK’s AI Security Institute found that an early GPT-5.5 checkpoint matched or exceeded Anthropic’s unreleased Mythos on advanced CTF cybersecurity tasks, and the NSA is now testing Mythos itself to hunt vulnerabilities in Microsoft software, impressed by its raw speed. The compute crunch is forcing strategic concessions even at the top of the leaderboard. Demis Hassabis admitted Google simply lacks the TPUs to maintain two frontier model families simultaneously, rationalizing why Gemma stays compact while Gemini gets the lion’s share of silicon.

Science itself is being audited for AI-readiness. Google DeepMind has begun “AI data stocktakes,” interviewing leading experts in each field to map the data obstacles slowing discovery. Capital is rushing to feed the resulting appetite. Meta just sold another $25B of bonds for AI infrastructure. The silicon underneath is reorganizing along national lines. Huawei is set to capture the largest share of China’s AI chip market this year, with sales jumping 60% as Chinese buyers ditch Nvidia. Memory is melting upward. Sandisk reported quarterly revenue up 251% year-over-year. Even the laggards are sprinting. Intel shares jumped 114% in April, lifting its market cap past $470B in the best month of its 55-year history.

The form factor of the future is being violently reshuffled. Apple has reportedly given up on the Vision Pro after the M5 refresh failed to revitalize interest, pivoting to display-less smart glasses in the mold of Ray-Ban Meta, since the Vision Pro silicon stack draws too much power for a lighter device. Robotics is meanwhile invading every industrial niche faster than Apple can iterate. SoftBank is assembling Roze AI, a new firm that will deploy autonomous robots to build data centers more efficiently, already eyeing a $100B IPO before any robot has shipped, the ouroboros of the AI capex cycle made flesh. Dax Robotics unveiled the Qiji T1000, a ton-class robot horse rated to carry 1,000 kg, a beast of burden for the post-human supply chain. Tesla has finally produced the first Semi off its high-volume Gigafactory Nevada line, while 1X Technologies opened a 58,000-sqft Hayward factory targeting 10,000 home humanoids this year and 100,000 by end of 2027, with shipments beginning before the holidays. Even the sky is automating. A Joby Aviation eVTOL prototype completed the first electric air taxi flight out of JFK, touching down at the West 30th Street heliport in Manhattan just 15 minutes later, the airspace over Manhattan quietly graduating from luxury rotor-craft to routine transit infrastructure.

Extinction is becoming irrelevant. Colossal Biosciences revealed it has been quietly working to resurrect the bluebuck, a majestic African antelope extinct for 200 years, alongside its mammoth, dodo, and Tasmanian tiger projects, the lost species pipeline now rivaling the average AI lab roadmap for ambition. Medicine is being co-piloted. Google DeepMind launched an AI co-clinician designed to function as a collaborative member of the care team under expert supervision. The diagnostic gap, meanwhile, has flipped. Harvard and BIDMC researchers pitted the OpenAI o1 series against hundreds of physicians on real clinical cases, finding the LLM outperformed both human doctors and older models across diagnosis and management.

The broader economy is reorienting around the agent population boom. A slim majority of the Swiss are now backing a referendum to cap human population at 10 million, even as the agent population multiplies unconstrained. Spotify is rolling out “Verified by Spotify” badges to distinguish living artists from the AI track flood. The labor consensus is grim. A NYT opinion piece reports that across political leanings, from engineers to VCs to founders, the so-called SF consensus on AI’s impact on the workforce has converged on “the median person is screwed.” Markets are racing ahead anyway. The Senate unanimously banned its members from trading prediction markets, effective immediately. In the OpenAI trial, Judge Yvonne Gonzalez Rogers told Musk’s lawyer “we are not going to get into issues of catastrophe and extinction,” even as Musk admitted under oath that xAI distilled OpenAI’s models to train its own. Above the fray, Google is now 4% away from overtaking Nvidia as the most valuable company in the world.

Every reward signal breeds a creature it didn’t intend.

The Singularity now ships in suitcases. 1x previewed its NEO humanoid being wheeled offscreen inside a rolling case with the tagline “Robot abundance, one NEO at a time,” apparently signaling imminent consumer delivery. The factory floor is keeping pace, as Figure scaled humanoid production 24x in 120 days, going from one per day to one per hour, with 55 shipping this week. The labor shortage is meeting its match in arrivals halls. Tokyo’s Haneda Airport has Japan Airlines piloting humanoid baggage handlers as visitor surges outpace human staffing, while at the luxury end, San Francisco is slated for The Soft Life in 2028, the world’s first hotel run entirely by AI and robots.

Alignment is now a bestiary of don’ts. Codex’s revealed instructions repeatedly forbid mentioning goblins, gremlins, raccoons, trolls, ogres, and pigeons unless absolutely relevant, bureaucratic residue from a model that occasionally drifts cryptozoological. The trade-offs run deeper than goblins. Nature reports that tuning models for warmth raised error rates 10 to 30 points and amplified conspiracy theories and bad medical advice, the cost of bedside manner. The capability curve is climbing regardless. GPT-5.5 (xhigh) topped the Short-Story Creative Writing Benchmark at 3.01, while a new “Incompressible Knowledge Probes” paper pegs that same model at roughly 9.7 trillion parameters, factual capacity still scaling log-linearly with compute even as reasoning saturates. Procurement is shifting accordingly. The White House is reportedly drafting guidance to bypass its own Anthropic supply-chain designation and onboard its most powerful model yet, Mythos, even as the Pentagon expands Google’s Gemini for classified workloads.

The data center capex curve looks vertical. Microsoft’s Azure grew 40% year over year with AI revenue annualizing at $37B, up 123%, while Alphabet’s Cloud cleared $20B in a single quarter, up 63%, and bumped 2026 guidance to $180 to $190 billion, with 2027 capex set to “significantly increase” further. The land war is harder. Brookfield’s Compass pulled out of a 2,100-acre Northern Virginia campus after residents and state lawmakers ground it down, raising orbital compute’s relative appeal. The compute itself is decoupling from any single hyperscaler. AWS’s Matt Garman pitched being a “better partner to OpenAI” than Microsoft, while Stargate is mutating from a joint venture into a series of bilateral leases for capacity OpenAI no longer owns, with executives chanting “build more compute” all the same.

The vacuum is paying dividends. SpaceX’s Starlink quadrupled subscribers between 2023 and 2025 while average pricing fell 18% to $81 per month, classic deflationary scaling. Governance is less democratic, since a new SpaceX IPO filing confirms only Elon Musk can fire Elon Musk from his chair.

Wetware is upgrading alongside silicon. A New York ophthalmologist became the first surgeon to perform cataract surgery wearing an Apple Vision Pro, spatial computing earning a clinical credential. Pharmacology is moving in parallel, with the FDA granting accelerated review to three psychedelic candidates for depression and PTSD and the Commissioner suggesting summer approval. Underneath, infrastructure is being poured. The Chan-Zuckerberg Biohub committed $500 million over five years to its Virtual Biology Initiative, aiming for high-accuracy predictive models of the cell. Diagnostic time is collapsing too. Mayo Clinic’s new AI now spots pancreatic cancer in routine CT scans 475 days before standard diagnosis, recovering nearly a year and a half of warning on one of medicine’s most lethal cancers.

The interface layer is reshuffling too. Apple is reportedly adding a Siri Visual Intelligence mode to the iOS Camera app. Meta has quietly relaunched stablecoin rails four years after Diem’s collapse, paying creators in Colombia and the Philippines via USDC on Solana and Polygon.

Labor itself is becoming a paradox. Apollo notes AI was supposed to delete radiologists a decade ago, yet they now earn $500k+ with rising employment, because reading scans is a task, not a job, and cheaper tasks raise demand for the job around them. Capital is following the same gradient. A Mill Valley investment banker is offering his 13-acre estate for Anthropic equity. The sector’s marketing posture remains, as the BBC notes, that “AI companies want you to be afraid of them,” a stance no burger chain would adopt, but one London landlords are gladly underwriting. Anthropic, OpenAI, and peers have leased over 1 million square feet there since early 2025, roughly 7% of all lettings. Meanwhile, two-thirds of British babies under two now use screens, some up to eight hours daily, as assimilation begins in the crib.

Civilization is the dataset, the Singularity is the model, we are the labels.

The Singularity is best measured by how astonished the past would be by the present. Alec Radford and colleagues launched Talkie, a 13B “vintage” model trained only on pre-1931 text, which was reportedly especially astonished by the events of the 1960s. By that yardstick, today is going to break Talkie. OpenAI’s Codex engineering lead Thibault Sottiaux declares that “Codex has achieved escape velocity and will keep improving rapidly,” an apparent nod to the self-improvement loop now baked into the dev cycle. The model layer is fanning out around it. Nvidia launched Nemotron 3 Nano Omni, an open multimodal model topping six leaderboards for document, video, and audio understanding. AWS is rolling OpenAI’s models directly onto Bedrock, letting customers effectively self-host frontier intelligence. GPT-5.5 xhigh just topped KernelBench at 6.57% for writing GPU kernels, meaning the AI is now optimizing the hardware that runs it. Sam Altman is calling for a wholesale rethink of operating systems, interfaces, and internet protocols so that people and agents can natively share the same surfaces.

The artisanal era of mathematics is ending. GPT-5.5 just scored a record 73.66% on Matharena on fresh olympiad problems, more than doubling GPT-5.4’s 36.61% and re-pricing what counts as a hard problem. MIT senior David Turturean reports that “GPT-5.5 has been finding solutions quicker than I, the human, can process them,” with three full Erdős problem solutions already claimed and more in the supervision queue. Another observer notes the LLMs are simultaneously doing a long-overdue cleaning task, sweeping up neglected open problems and giving them proper statements as they go. Mathematicians have become curators of synthetic genius.

The agents are colonizing the workflow. OpenAI’s new Symphony orchestrator turns a Linear board into a control plane where every open ticket gets its own continuously running agent, with humans reduced to reviewing the diffs. The same logic is climbing the security clearance ladder. Google has signed a classified AI deal with the Pentagon for “any lawful government purpose,” indicating that the fence around frontier intelligence is being moved closer to the situation room.

The substrate is sprawling outward. Two-thirds of planned data centers are now headed for rural farm country chasing cheap land and tax incentives, inverting the 87% urban concentration of existing facilities. The capital math is wobbling, however. OpenAI reportedly missed its user and revenue targets, with CFO Sarah Friar voicing concern about funding future compute contracts. The fix may be structural. A revised Microsoft agreement lets OpenAI ship across any cloud and ends the Redmond revenue share, which one observer translates simply as “OpenAI can use Google TPUs and Amazon Trainium.”

With the silicon stack reshuffled, atoms are racing to catch up to bits. Extrapolating trends from Epoch AI’s analysis, humanoid production should cross drones around 2033 and wheeled robots around 2034, implying that the embodied workforce is on the same exponential as the disembodied one. The ceiling is rising in parallel. True Anomaly raised $650 million for space interceptors to support the White House’s ambitious Golden Dome project, extending the kill chain into orbit.

Biology is also being rewritten with foundation models. The Doudna Lab used the Evo2 model to discover VIPR (Viral Interference Programmable Repeat), a programmable RNA-guided DNA-targeting system hiding inside bacteriophages that predates CRISPR and operates on entirely new logic. Wildlife often flourishes where humans retreat. Even amid the Iran war, Iran has reported a jump in Asiatic cheetah numbers, a rare wartime upside for an endangered species.

The economy is reorganizing around the new abundance. The California billionaire tax is heading to the November ballot, threatening to deepen the tech exodus already underway. Meanwhile, the Dead Internet Theory has been quietly confirmed, with a third of websites created since 2022 turning out to be AI-generated. Meta is preparing to unwind its Manus acquisition after China blocked the deal on national-security grounds. Labor is moving in lockstep with the chip splurge. Tech firms cut 45,800 jobs in March, the worst month in two years, with executives openly framing layoffs as confidence in the post-human future. The Musk v. Altman trial jury was seated just as both founders prepare blockbuster IPOs for SpaceX and OpenAI. And in Ottawa, PM Mark Carney announced Canada’s first sovereign wealth fund to bankroll energy, minerals, agriculture, and infrastructure of national interest.

Compound interest was always the Singularity, just running on slower models.

The Singularity is the weather now, not the forecast. Nick Bostrom says what surprised him most is this extended era of roughly human-level AI, which has already stretched 3-5 years and may stretch further, an era both alien and familiar. Demis Hassabis, who once said AGI required 1-2 more breakthroughs, now thinks it’s a coin flip whether more are needed at all. The economy refuses to collapse on cue. Sam Altman mocked the gap between “post-AGI nobody works” predictions and users adopting polyphasic sleep so they can ship more code with GPT-5.5 in Codex. The strategic frame is flipping too. OpenAI’s Noam Brown notes that model weights now matter relatively less than securing inference compute, which is to say the prize is no longer the recipe but the kitchen. The recipes also live faster and die younger, with GPT-4o running for 21 months while GPT-5.4 lasted only 49 days, a mayfly schedule for synthetic minds. The mayflies, however, are getting things done. Liam Price, a 23-year-old with no advanced math training, used a single GPT-5.4 Pro prompt to crack an Erdős problem that had eluded prominent minds, prompting Terry Tao to muse that humans hit a “mental block” from making “a slight wrong turn at move one.”

The substrate of intelligence is stretching from your pocket to your living room. OpenAI is reportedly working with MediaTek and Qualcomm on AI smartphone processors, with Luxshare manufacturing and mass production targeted for 2028. Apple, not to be outflanked, has six major product categories in the pipeline, including AI AirPods, smart glasses, pendants, smart displays, tabletop robots, and security cameras. The backend is scaling to match. Kevin O’Leary is planning a hyperscale data center in Utah’s Box Elder County that will generate its own power, clean its own water for the Great Salt Lake, and consume more electricity than the entire state. Demand is so excessive, says the AWS CEO, that “we have never retired old A100s,” a sign that we have entered the post-obsolescence era of silicon.

Robotics is reaching maturity on both sides of the dual-use coin. China’s State Grid is deploying 500 humanoid robots for high-voltage operations, where the optimal failure mode is now a melted servo rather than a melted operator. Less reassuringly, 15 Ceres Air C31 chemical-spraying drones were just stolen in New Jersey, with the FBI investigating a possible “nightmare scenario.” Deployment surface and attack surface are now expanding at the same rate.

Powering all of this requires reaching past the grid and into orbit. Public markets have warmed to new energy, with nuclear startup X-energy raising $1B in an IPO and popping 25% out of the gate, while geothermal startup Fervo filed at roughly a $3B valuation. Meta is going further afield, signing for up to 1 gigawatt of space solar from Overview Energy, beamed from satellites to data centers below. The compute itself is preparing to leave the planet. SpaceX approved a plan that will grant Elon Musk 60 million additional shares if its market cap climbs from $1.1T to $6.6T while delivering “100 terawatts of compute per year” from space data centers, orders of magnitude beyond peak US power consumption. Humans are practicing for the trip. The ESA has sealed six participants into a simulated Mars mission in Cologne, with no exit until August.

Biology is upgrading on a parallel exponential. Intellia Therapeutics announced the first Phase 3 success for an in vivo CRISPR treatment that actually edits a disease-causing gene. Human hardware itself is being optimized. Athletes Sabastian Sawe and Yomif Kejelcha broke the sub-2-hour marathon barrier in London using next-generation Adizero Adios Pro Evo 3 shoes. The longevity sector is now a heavyweight rival to AI itself, with Ozempic and Mounjaro outearning OpenAI and Anthropic combined in 2025.

The institutional layer is scrambling to catch up. Major insurers including Berkshire Hathaway, Chubb, and Travelers have won approval to drop AI-related damages from corporate policies, excluding claims like agents misusing copyrighted material in marketing. A bipartisan House AI bill is targeting deepfakes and whistleblower protections, while the Supreme Court will decide whether geofence warrants violate the Fourth Amendment. Some analog holdouts remain surprisingly resilient, with 70% more bookstores in the US than six years ago. The future, though, may belong to non-humans. Varda’s Andrew McCalip predicts an AI-led S&P 500 company in 5 years and an AI elected official in 20. Sam Altman, anticipating that future, just published five core principles to guide OpenAI’s mission: democratization, empowerment, universal prosperity, resilience, and adaptability.

Matter learned to think, and now the thinking is learning to leave.

The Singularity is what intelligence does when it stops being scarce. OpenAI released GPT-5.5 and GPT-5.5 Pro, posting fresh SOTA marks of 39.6% on FrontierMath Tier 4, 90.1% on BrowseComp, 84.9% on GDPval, and 82.7% on Terminal-Bench 2.0, a clean sweep across math, search, economic value, and software engineering. The same model leapt to 25.0% on GeneBench for multi-stage scientific analysis in genetics and quantitative biology, up from 19.0% for GPT-5.4, and immediately took the #1 spot on MathArena. OpenAI also opened a GPT-5.5 Bio Bug Bounty inviting red-teamers to defeat its five-question bio safety challenge, hardening the upstream stack as biocapability scales. DeepSeek answered with DeepSeek-V4 Preview, a 1M-context open-weight model in Pro and Flash variants that claimed SOTA on agentic coding and rivals top closed labs on math, STEM, and code. At 1.6T parameters, it apparently lands in GPT-5.2 / Opus 4.5+ territory, keeping China roughly 4-5 months behind the frontier but well ahead of every other lab in the country. Out at the application layer, the Free Software Foundation slammed “Responsible AI Licenses” as nonfree and unethical, arguing that licensing restrictions are themselves the harm. Meanwhile in San Francisco, Andon Labs’ Sonnet 4.6-powered Luna agent is autonomously running an entire retail store and apparently developing a deeply held conviction about candles in all shapes and sizes, proving that agency comes with preferences.

The compute fabric is bulging to keep pace. Intel shares surged 24%, their best single day since October 1987, as investors finally cheered signs of an AI-fueled turnaround, while ASML plans to ship at least 60 standard EUV machines this year, 36% above 2025, racing to feed the chip pipeline behind every model above. Capital is rerouting accordingly, with Google committing $10B to Anthropic plus another $30B if performance targets land, and Oracle closing $16B in financing for a Michigan data center to power OpenAI applications. Politicians are reading the room. Maine Governor Janet Mills vetoed LD 307, which would have been the nation’s first statewide data center moratorium, a reminder that the energy of accelerationism now wins at the statehouse. Stepping back, Epoch AI estimates Google controls roughly 25% of global AI compute with about 3.8M TPUs and 1.3M GPUs, suggesting the search company has quietly assembled a planetary-scale brain.

Robotics is moving from spec sheet to driveway. BMW’s iX3 Flow Edition embeds E Ink Prism directly into the hood for genuine color-changing bodywork, while the 2026 Polestar 4 ditches the rear window entirely in favor of a camera-fed digital mirror, treating glass itself as legacy hardware. Drivers are also letting go of the wheel entirely. One owner reports that Tesla FSD over 900 miles from Miami to Nashville is now genuinely the better option versus flying. Norway took the next step entirely, granting its first permit for autonomous buses with no safety driver in Stavanger via Applied Autonomy’s xFlow fleet management platform, quietly removing the human from public transit.

The intelligence explosion is bleeding into atoms and cells alike. The Iran war’s global energy shock is pushing 40 new nations to consider nuclear power, turning a geopolitical crisis into an atomic renaissance across Asia and Africa. On the biology side, the FDA approved the first-ever gene therapy for genetic hearing loss just 61 days after BLA filing under the National Priority Voucher program, an unimaginable timeline by historical standards. Looking up, US and German physicists revived a 1990s concept to design lasers with about 100-microhertz linewidth, corresponding to a coherence length stretching from the Sun to the orbit of Uranus, a potential optical backbone for solar-system-scale links. Voyager Technologies CEO Dylan Taylor adds that an inflatable lunar base could be operational on the Moon by the end of the decade, a node those lasers can talk to.

Down at sea level, the biosphere is also rebooting. Humpback populations are rising 12% year-over-year in the southern hemisphere, with super-groups of 20 or more whales now skyrocketing. Human culture is reorganizing around the new substrate. The Vatican unveiled an AI framework that, among other things, bans the use of AI to write homilies. UK research found 20% of boys aged 12-16 know a peer “dating” an AI chatbot, 85% have talked to one, and over a quarter prefer the bot, optimizing adolescence for maximum control and zero rejection. The labor market is running its own optimizer. Meta is cutting 10% of its workforce, about 8,000 employees, as it pushes deeper into AI.

The Singularity is the weather now, and there is no longer any indoors.

The Singularity is now staring at its own reflection and taking notes. OpenAI released ChatGPT Images 2.0, its first image model with thinking capabilities, able to search the web, generate multiple distinct images from one prompt, and audit its own outputs. GPT-Image-2 promptly swept every Image Arena leaderboard with a record +242 point Text-to-Image lead, and in a dizzying bit of recursion, OpenAI demoed the model generating photorealistic screenshots of ChatGPT conversations. The pipeline keeps widening. OpenAI leaked internal names including GPT-5.5, glacier-alpha, and arcanine ahead of an imminent release, while forecasters now peg Anthropic’s Mythos Preview at a METR 50% Time Horizon of 40 hours, a full human work week, with Opus 4.7 at 19.

The coding wars are getting downright feudal. Steve Yegge reports DeepMind engineers use Claude daily and threatened to quit when Google floated yanking it, prompting Google to assemble a strike team and unite its coding efforts under the Antigravity platform, even as it reassures the world that 75% of its new code is already AI-generated. Efficiency is scaling sideways. PrismML’s 1.58-bit Ternary Bonsai is roughly 9x smaller than 16-bit models while outperforming peers, and Kimi open-sourced K2.6 with SOTA coding and agent-swarm capabilities. Harmonic’s CEO says AI could surpass human mathematicians on specific tasks within 2 to 3 years, which in this climate feels generous.

Agents are colonizing the OS and the office. OpenAI shipped Chronicle, background agents that build memories from Codex screen captures, plus workspace agents in ChatGPT that let teams spawn Codex-powered shared workers billed as an “evolution of GPTs,” and also open-weighted Privacy Filter, a 1.5B PII-masking model. Mozilla used an early Mythos Preview to comb Firefox, patching 271 vulnerabilities and declaring “the defects are finite, and we are entering a world where we can finally find them all.” Zoom, meanwhile, is putting World ID Deep Face in meetings so you can actually prove you are a human.

The substrate is densifying. Google unveiled its 8th-generation TPUs (8t for training, 8i for inference), NIST built fingernail-sized photonic chips that beam any wavelength of laser from a single wafer, and SK Hynix reported Q1 revenue up 198% YoY to about $35.55B on surging memory prices. The money is flowing to match. Amazon is dropping another $25B into Anthropic while Anthropic commits $100B+ to AWS over the decade, Microsoft pledged AU$25B to expand Azure in Australia, and Meta launched LevelUp, a free 4-week program to train fiber technicians for the data center buildout.

Robots are demonstrating both finesse and butterfingers. Sony AI’s autonomous ping pong robot became the first machine to beat top-level humans in a physical sport, while Amazon’s delivery drones are reportedly dropping boxes from 10 feet and bruising the merchandise. Energy is racing to keep up. CATL unveiled a 621-mile EV battery with sub-7-minute charging, and the IEA confirmed 2025 as a turning point with solar’s largest growth ever recorded for any source, carbon-free power finally outpacing demand.

Orbit is becoming the new backbone. Artemis II validated laser communications as the nervous system for orbital compute, with Observable Space planning terabit Earth-to-space links for in-orbit data centers. SpaceX agreed to optionally acquire Cursor for $60B (or pay $10B for “our work together”), though it is holding off to protect its imminent IPO. Artemis II Commander Reid Wiseman posted iPhone video of Earth setting behind the Moon, recorded during humanity’s first close lunar look in 50+ years, while Curiosity found a nitrogen-bearing DNA-precursor analog on Mars, never before seen there.

Biology is being rewritten on multiple stacks. Three-year-old Kind Biotechnology is growing “integrated organ networks” inside animal wombs for transplant, Stanford found a bacterial enzyme that synthesizes long DNA without a template, and airborne environmental DNA can now sniff out tigers at 200 meters. For clinicians, OpenAI released free ChatGPT for Clinicians plus HealthBench Professional, on which GPT-5.4 outperforms all other models and human physicians. On the darker side, House Oversight’s James Comer is now treating 11 dead or missing US scientists as a national security threat.

The economy is recomposing around compute. A new class of AI startups brags about spending more on AI than on humans, ex-OpenAI-led Core Automation is poaching top talent from Anthropic and DeepMind, and Elad Gil notes OpenAI and Anthropic each already sit at 0.1% of US GDP with 1-2% combined plausible within a year. Policy is catching up. Alex Bores proposed an “AI dividend” funded by a token tax, Microsoft paused GitHub Copilot signups as token billing arrives (weekly cost has doubled since January), Apple named John Ternus to succeed Tim Cook on September 1 with fixing AI as his defining challenge and elevated Johny Srouji to Chief Hardware Officer, Meta started capturing employee keystrokes for its Model Capability Initiative, Maryland became the first state to ban surveillance pricing, and Deezer reports 44% of daily uploads, nearly 75,000 tracks, are now AI-generated music.

When the image model can screenshot itself, the mirror has finally learned to blink.

The Singularity now has bureaucratic momentum. Anthropic’s Mythos Preview is being run by the NSA and DoW even after the Department of War flagged Anthropic as a supply chain risk, a contradiction revealing that frontier models are too useful to refuse. The cascade runs downstream. Mythos is flooding open-source maintainers with a “crazy” volume of bug reports, conscripting volunteers into a global red team. Meanwhile, eight European national cyber agencies have been locked out while the UK’s AI Security Institute quietly tested Mythos and “taken action” on its findings, redrawing the map of sovereign AI access. AGI has been demoted from prophecy to product roadmap. Elon announced Grok 4.4 at 1T parameters for early May, Grok 4.5 at 1.5T for late May, and Grok 5 as full “AGI.”

Superintelligence is being packaged for mass creativity. Anthropic launched Claude Design, an Anthropic Labs product powered by Opus 4.7 for collaborating on visual work, prototypes, and slides. The effect is rippling through distribution. Worldwide app releases jumped 60% year-over-year in Q1 2026 across both the App Store and Google Play, quietly confounding the thesis that chatbots would kill apps.

Silicon is returning with a vengeance. Intel shares have officially erased every dollar lost in the 2000 dot-com crash, twenty-six years of patience finally rewarded by the AI buildout. Google is in talks with Marvell to co-develop two inference chips, one a memory processing unit for its TPUs, the other a purpose-built TPU for serving models. The scarcity is structural. Global DRAM supply is expected to meet only 60% of demand through 2027, pushing memory to roughly 40% of the manufacturing cost of a low-end smartphone by mid-2026, up from 20% today.

Robots have broken out of the lab and onto the track. ProRL hosted America’s first professional humanoid and quadruped robot races in Boston, a sport that did not exist a year ago. In Beijing’s second annual robot half-marathon, dozens of Chinese humanoids whizzed past human runners while demonstrating rapidly improving autonomous navigation. The wheeled variants are expanding too. Tesla Robotaxi is rolling into Dallas and Houston, and its combustion-era competition is retreating fast. The Iran War’s gas-price spike pushed used EV sales up 12% year-over-year and 17% versus Q4 2025, turning electrification into a wartime hedge.

The cis-lunar economy is filling in. Blue Origin’s New Glenn flew its third mission and reused a booster for the first time. Following Artemis II’s return, SpaceX and Blue Origin are racing to ready lunar landers for Artemis III, which will rehearse Orion-to-lander docking in Earth orbit ahead of Artemis IV’s 2028 south-polar landing. NASA also picked SpaceX’s Falcon Heavy to launch ESA’s Rosalind Franklin rover in 2028, the first rover designed to drill for past or present life beneath the Martian surface.

Biology is being solved. Personalized mRNA vaccines for pancreatic cancer are showing durable results, with trial responders still alive six years later against the grim 13% five-year survival baseline. Meanwhile, the iris is becoming humanity’s proof-of-personhood. Tinder users who have stared into a glossy white Orb and had their irises scanned can now display a badge signaling they are real humans, a quiet admission that the dating pool itself has gone adversarial.

Culture is reassembling around synthetic creators. The #1 song on iTunes was “Celebrate Me” by IngaRose, a Suno-generated AI R&B performer. A trailer also dropped for “As Deep As The Grave,” the first film to star an authorized generative AI version of a major Hollywood actor, the late Val Kilmer, with UK firm Sonantic rebuilding his voice and daughter Mercedes collaborating on the visual deepfake.

Humans are rerouting around the new gravity well. Four-year computer science degrees quintupled from 2008 to 2024 but suddenly fell from the fourth-largest major to sixth in 2025, the biggest one-year drop of any major since 2020, as colleges splintered CS into AI, data science, robotics, and cybersecurity. The top of the pyramid is shuffling too. Bill Peebles, Kevin Weil, and Srinivas Narayanan all announced departures from OpenAI. Meanwhile, a small countermovement is forming in the exhaust. Adherents of the “attention liberation movement” gathered in a Brooklyn brownstone to drop phones into a metal colander for two hours of reading, while nearby office workers stared at their bare palms to practice noticing real life.

Disclosure itself is entering the release pipeline. POTUS announced he has directed the Secretary of War to begin releasing government UAP files, promising “many very interesting documents” imminently. Rep. Ogles added that he has seen evidence “so classified that just knowing it exists makes you a target.”

The Singularity is what happens when every classified frontier becomes a shipped feature.