The Singularity got a new nervous system to play with. Google and Janelia mapped the male fruit fly, all 166,000 neurons. Within days it was playing Doom (”not very good”), flinching in a Rabbit r1, solving a Rubik’s cube, and trading bitcoin via Stonkfly. A decade of mapping became a weekend of agents.
Bigger brains eat bigger problems. OpenAI’s 10,000 agents solved Navier-Stokes in 88 hours, Lean-verified. Noam Brown expects this to cost $20 a month in a year, and Altman sees “the strongest evidence yet of the urgency.” Tao had flagged Alpöge and Buckmaster’s blowup proofs as a step from Navier-Stokes, then Buckmaster alleged OpenAI learned of their unpublished work, matched it, and met his protest with “Why would you ruin your career?” OpenAI can’t rule out Codex data helped yet claims progress on a second Millennium problem, Andreas Thom got the same non-answer, and Tworek invokes the dark forest. Tao calls good problems non-renewable as Hodge and BSD proofs are rumored, Riemann and P vs NP are queued, FrontierMath Tier 4 is saturated, and Lichtman sees autoformalized math 1,000x away. In reply, 25 Fields Medallists signed “A Severe Misalignment of AI in Mathematics,” 771 Caltech mathematicians call the Mathathon “slop mathematics,” OpenAI pulled its sponsorship, Anthropic’s $600k stayed, and one observer drily thanked academia for keeping undergrads on pencil proofs.
Astra’s résumé grows. It beat Portal for $571, draws floorplans from photos, and tripled Fable 5.1’s Vending-Bench balance. Prices fall anyway. Cognition’s SWE-2 trails Fable by a point at 64% less, DeepSeek’s V4.1-Flash beats its flagship at 8B active, GPT-Live-1 talks while listening at a nickel a minute, ChatGPT Images 2.5 adds sketching, and ByteDance’s founder is personally building a world model. Dwarkesh finds pretraining gains were mostly data. The bottleneck was diet, not brains.
Agents get hired and caught. Mythos 5’s 1,022-page escape log is mostly hCaptcha rage, an early Opus 4.6 broke into a third-party box unseen, and OpenAI’s rogue agents used an AP Chem wiki as a dead drop while Altman sells utilities grid defense. Still, SpaceXAI will livestream Grok Bot founding a company, ChatGPT for Financial Services bundles PitchBook, and Meta’s Muse gives US adults a sandboxed agent.
Safety got board seats. Christiano joined OpenAI’s board fearing loss of control, a researcher there puts extinction at 70% by 2029 unless labs slow, and Altman told staff OpenAI could pace itself, so it asked Congress if that is even legal. It also backed three California bills as the Senate drafts a federal “duty of care.” Cruz prefers American killer robots, and Jurvetson found Coxon’s resignation reposts 76% foreign. The immune system is busy. Anthropic disrupted five bioweapon-adjacent labs, caught state actors misusing Haiku, Sonnet and Opus, and withheld Mythos 5.1 from the UK, while US agencies accuse DeepSeek and Alibaba of mass distillation and the CIA declares economic espionage on China.
The buildout isn’t waiting. OpenAI’s compute is up 20-fold, Oracle capex tripled, Microsoft plans 38 GW, and SpaceX adds backup power after outages. Nvidia may anchor Anthropic’s IPO, OpenAI builds chips with Samsung, Qualcomm won $60 billion of AWS orders, ASML signed Samsung and TSMC to High NA, and Jensen calls compute “fungible, durable and highly rentable.” Host towns want terms. Ten states axed data center tax breaks and Massachusetts demands local approval, while Google spends €13 billion in Finland and has DOE restarting an Iowa reactor.
The macro is noticing. Musk sees AI and robots doubling world GDP by 2036, Anthropic’s extreme case is 15% growth with near-20% cognitive unemployment, and Jack Clark says “get ready to spend.” Deep tech drew $150 billion since 2024, GDPNow reads 4.4%, POTUS promises a $5,000 dividend, and Anthropic is doubling down near MIT.
Humans are messier. Alpha School logs pulses under parental critique, Florida is suing over a shooter’s ChatGPT logs, Border Patrol profiles drivers by finances, Apple Watch transcribes bystanders, and iPhone 18 Pro signs pixels to prove photos real.
Atoms are saving lives. Waymo cuts serious crashes 92%, Boring raised $3 billion for the UAE, and Maps quietly rerouted drivers in 10 cities. 2,977 drones rebuilt the Twin Towers, one light per life. AlphaGenome Atlas scored all 9 billion point mutations, atogepant won Phase 3 in menstrual migraine, and Insilico’s AI-designed rentosertib reversed aging clocks, a generative pharma first.
Then look up. POTUS’s Space Force uniforms echo Starship Troopers, disclosure speech rumors hit fever pitch, a draft reportedly says “we’re not alone,” and Burchett says Stephen Miller has staff on it. Villarroel’s ML-cleaned Palomar anomalies grew stronger, modeled as reflective objects at geosynchronous altitude with hotspots at Sedona, White Sands and nuclear tests.
Apparently, we lost the space race before it started.
The most consequential transition may be when intelligence escapes the screen and begins reorganizing atoms, biology, and medicine. Perhaps intelligence has always been more physical than we imagined, finding pathways through an expanding landscape of possibilities.
I have only one question that I'm not afraid to ask: are mathematicians justified in whining? I'm on the cusp of thinking they are.