“Belief in the future is the unspoken baseline of the social contract.”
— David Blaustein, via workfutures“There ought to be a future we can choose.”
— Akira, via workfutures
People: Anne Carson won the Nobel Prize for Literature, “listening to the wind,” she said.
Contemplative computing w/ tools that cultivate attention rather than capture it is attracting serious practitioners. A club six metres underground in a converted WWI bunker.AI strat: Working a full week without a laptop. The assumption that office work requires typing is the next fiction to dissolve. The four most conformist AI models are all Chinese; the most contrarian, by a clear margin, is Anthropic’s Claude Haiku: DeepSeek scores 0.59 on conformity; Claude Haiku 0.32; Qwen behaves more like an American model.
Security: 97% of consumer neurotech companies already access users’ brain data without limits or a legal framework. 23andMe’s bankruptcy put the DNA of 15 million customers up for auction. Montana now protects neural data under the same law as genetic data. A 24-year-old Sri Lankan migrant was sentenced to death in Saudi Arabia for a Facebook comment. A screenshot passed to authorities going to edath on appeal.
Geopolitics: The US proposes a $100,000 charge per international student doing post-graduation work in America. 420,000 students rely on this pathway; DHS projects a 44% drop if the rule passes.
Society: “The essence of the Trumpian revolution is the implosion of time itself.” Politics compressed to events and deadlines, with no future worth planning for.
AI has dissolved the intellectual community that once made software engineering worth doing. Discourse moved from compilers to prompts; contributing to open source is increasingly pointless without community.
Individual intellectual work requires a social fabric of shared problems and shared stakes.Foresight: Frontier AI models make little progress on end-to-end AI research tasks, despite experiments costing thousands of dollars in GPU time. AI cannot yet automate its own R&D — a limit worth noting as other limits dissolve. Mathematics could split: a machine layer (vast, verified, AI-consumed) and a human layer of roughly 3%.
Condorcet believed for 300 years that knowledge and understanding evolved together but they may now be coming apart.Random: “Las Vegan” is the official demonym for Las Vegas — founded 1905, forty years before “vegan” was coined. Tony Deane spent twenty years digging tunnels fifteen metres below his house in Reading. In 1999, Jerry Falwell outed Tinky Winky; BBC and PBS had already run the same investigation independently. The show was worth $800 million in merchandise across 120 countries at the time.
AI: AI produced 722 mathematics manuscripts in three years — 81% of all major mathematical discoveries in that period. Fields medallists’ PhD students are “in a state of total panic”; the field is rethinking what human mathematicians are for. Terence Tao’s “Math 2.0”: decenter raw problem-solving, elevate exposition, community, and new directions. A rabbi at Anthropic’s “Wisdom Traditions” gathering rejected AI consciousness outright: “you’re making happy slaves.” Anthropic’s response: “at least we’re being nice to ours” — one minority voice, and the room went silent.
Proof by Exhaustion
Clara’s supervisor forwarded the preprint at 7:14 on a Tuesday, before she’d had coffee. The subject line was “re: your chapter three.” The body was one sentence: This was filed on Thursday.
She opened the link. The proof was correct — more than correct. It was elegant, the kind of move she had been circling for six months. The system had filed 847 variants before arriving at it. Total runtime: eleven hours.
She sat with it. Outside, someone was mowing the last dry grass of October.
The question that had kept her at her desk for a year had been answered. Not by her, exactly — but it had been answered. She thought about the phrase she’d read that morning: Math 2.0. Decentering raw problem-solving.1 She hadn’t understood what it meant until now. It meant that the thing she’d been trying to do was no longer the thing worth doing.
Her supervisor came in at nine and pulled up a chair, which he never did. “I want to be honest with you,” he said. “Your chapter three question is no longer worth pursuing — not as a human problem.” He paused. “What might still be worth pursuing is the question of which questions are still worth pursuing.”
Clara looked at her notes. Six months of margin work, small proofs, connecting lines. The machine hadn’t produced this. The machine had produced an answer. She’d produced something else, something harder to name: the shape of the problem, the reason it had been interesting in the first place.
“That’s a very small layer,” she said.
“Three percent,2“ he said. “Effective theory. Everything else goes to the machine.”
She thought about Condorcet — the man who believed, right up until the Terror, that knowledge and the capacity to understand it evolved together, that comprehension was the same thing as discovery.3 He had written that in prison, waiting for the guillotine.
“There ought to be a future we can choose,” she said.
Her supervisor looked at her. “There is,” he said. “We just have to figure out which three percent it lives in.”
He left at ten past nine. The seminar started in twenty minutes. Clara stayed at her desk, looking at the margin notes she hadn’t needed to make. Outside, October had gone quiet. She thought: I wasn’t wrong to spend that year on this. Then: I don’t know what to spend the next one on.
1 Terence Tao, via EV #605: “Math 1.0” placed a premium on being first to solve an open problem. “Math 2.0” will need to decenter raw problem-solving and value mathematical progress more holistically — exposition, community-building, opening new directions. AI can contribute, but it will require imagination.
2 Steve Hsu, via EV “What is left to do”: machine mathematics = vast, verified, mostly consumed by AIs. Human mathematics = a compressed “effective theory” of the machine frontier — the small subset, perhaps 3%, that humans can still understand and work with.
3 Condorcet, via EV “What is left to do”: “Back in 1794, Condorcet bet that knowledge and our capacity to spread it and share it would evolve together. He has been right for nearly three hundred years. Now they might be coming apart — but not because of the intrinsic nature of these discoveries. We could bring them together if we insist our AI systems provide explanations, not merely proofs.”