Situational Awareness, Copilot Summer, and the Decade That Decides

Earlier this year I wrote that corporations and governments need to start planning for a post-labor economy now — that waiting until people feel real pain is a recipe for upheaval. This month made that argument harder to dismiss.

A former OpenAI researcher named Leopold Aschenbrenner published Situational Awareness, a 165-page case that the people closest to the technology genuinely expect machines smarter than most humans at most cognitive work by the end of the decade. You can quibble with his timelines. What you cannot quibble with is the trendline he charts: each leap from GPT-2 to GPT-4 took capabilities from "toddler" to "smart high schooler," and the compute curves driving those leaps have not bent.

Meanwhile, the copilot quietly became infrastructure. Apple put an intelligence layer into two billion devices at WWDC. Anthropic shipped Claude 3.5 Sonnet, and for the first time a mainstream model writes production code many professionals would sign off on. Nvidia — the company selling shovels for this gold rush — briefly became the most valuable company on Earth.

The Challenges

Notice what none of these announcements mentioned: employment. We are industrializing cognition the way the nineteenth century industrialized muscle, and the public conversation is still about chatbot hallucinations. The Luddites were not wrong that the looms would change everything; they were wrong about what could be done with that fact. Smashing the machines failed. Preparing people for the machines was never seriously tried.

The Advantages

I remain, stubbornly, an optimist. Intelligence too cheap to meter is the raw material of a Star Trek economy — medicine, education, and expertise that no longer gate on wealth. But the bridge between here and there is not self-building. It is policy, and policy runs on lead time.

Actionable Steps

If you run a company: start the workforce-transition planning this quarter, not after the first AI-attributed layoff round. If you write policy: the studies on income floors and retraining need to be commissioned now, because their results take years. If you are an individual: get your hands on these tools this summer. Fluency is the only moat an individual can dig alone.

The decade that decides has already started. Plan before the pain.

LLAP. 🖖