20260510
#212
“In capitalism, the future is an asset and it’s already been sold.” — Total Refusal, Hardly Working
“The idea of Progress is seductive, and so intuitive that we never question that there is a direction to it.” — Karl Schroeder on Stephen J. Gould, My Library: Full House
People: A national survey on the state of shame. China’s sad toy horse meme. Record 45% of US adults now identify as political independents. The One-Time Pad — on codes, human error, and the Cold War. Public benches. Jobless futures.
Geopolitics: Battle deaths from cross-border conflicts averaged 15,000/year from 1989–2014 — now over 100,000/year. China’s economy 30% larger than the US by purchasing power, industrial base twice as large. Record $1.19 trillion trade surplus.
Futures: The Ark Builders — on deep time, global hiding places, and the case for forgetting.
AI: The labs are rationing compute. Codex went from 100,000 to 2 million developers in three months. Alibaba closed-sourced Qwen. H100 rental prices at 18-month high. Economists forecast AI adds 1–1.5 percentage points to US annual growth by 2050 — plus 10 million fewer jobs and inequality at its highest since 1939. GitHub Copilot injected promo content into code reviews.
Random: Modern life as garbage time — the sports metaphor taking over Chinese internet.
The Rationing
Avery first noticed the rationing on a Tuesday, when her session hit a wall at 2:47 PM.
She had been asking Claude to help draft a procurement report — nothing unusual, nothing that would have troubled the system three weeks earlier. But the interface returned a polite note explaining that she had reached her usage limit for the period. Would she like to try again later?^1
“Later,” she said aloud, to nobody. The office was half-empty. It had been half-empty since March, when the team shrank from twelve to seven. The official term was “right-sizing.” Derek, who had survived two previous right-sizings, called it “wrong-sizing in a nicer font.”
Derek was in procurement, which meant he spent his days tracking things most people never thought about — GPUs, mostly, and the electricity to run them. H100 rental prices had just hit an eighteen-month high. OpenAI’s CFO had publicly admitted they were turning away business. Anthropic had tightened its limits. Alibaba had closed-sourced Qwen, its open-weight model, like a restaurant pulling its recipe off the internet because too many people were cooking at home.
“The labs are rationing,” Derek told Avery over lunch. “Not because the technology got worse. Because it got too good and too many people showed up.”
“So it’s a queue,” Avery said.
“It’s a queue where the building is also on fire and somebody is selling tickets to watch.”^2
Their colleague Matt, who worked in what the company still optimistically called “strategic foresight,” had been tracking a different kind of rationing. Battle-related deaths from cross-border conflicts had risen from an average of fifteen thousand a year to over a hundred thousand. The future, Matt often observed, was being spent faster than it could be replenished. He had pinned a quote from an artist collective to his monitor: In capitalism, the future is an asset and it’s already been sold.
That evening, Avery’s daughter showed her a plushie she’d seen online — a toy horse from a shop in eastern China with its mouth accidentally sewn upside down, turning its smile into a frown. It had gone viral. Young Chinese workers were buying them in pairs: the crying one for the office, the smiling one for home.
“Why is it sad?” Avery’s daughter asked.
“Because it works too hard,” Avery said, which was both a simplification and the entire truth.^3
On Thursday, GitHub Copilot started injecting promotional content into code reviews. The developer community erupted. The feature was pulled within hours, but Derek found the incident revealing. “They didn’t run out of ideas,” he said. “They ran out of revenue models. So they tried putting ads in the thinking.”
Matt, who had been reading about the Karpathy loop — a method for turning AI into an automated research assistant — pointed out that the real shift wasn’t the rationing itself. It was what happened when people learned to work within the constraints. The earliest adopters, the ones who’d been using these tools for years, had developed instincts that newer users hadn’t. They knew when to push, when to rephrase, when to step away. Falling behind in AI, he said, was not a technology problem. It was a learning problem.
Avery reopened her session at 5:30 PM. The limit had refreshed. She typed her question again, more precisely this time — fewer tokens, tighter framing, the kind of economy you learn when resources are finite.
The answer came back in seconds. It was better than her morning attempt.
“There’s probably a lesson in that,” she muttered, and went home to the smiling horse.
^1 The politeness of AI refusals is perhaps their most unnerving quality. A human gatekeeper might at least have the decency to look uncomfortable.
^2 Derek had a gift for metaphors that were simultaneously inaccurate and entirely correct.
^3 The sad horse plushie became, in the span of a single lunar cycle, the most honest corporate mascot in modern history — a creature that understood its job, resented it, and showed up anyway.
^4 Alibaba closing Qwen’s weights is the AI equivalent of a neighbour who used to leave their Wi-Fi open suddenly installing a password. You knew it would happen. You’re still annoyed.