mahigan.com
FR

The End of History

Is it because the rupture is too great even to be contemplated? Is it because the world is looking the other way?

Is it wise to cling to the judgment of the “It’s all going to be okay” experts as if to a life buoy?

The simple(-minded) calculation I made two weeks ago, carrying the acceleration multiplier between the Agricultural Revolution and the Industrial Revolution (x33) over to the ratio between the Industrial Revolution and the Artificial Revolution, seems to be holding up even better than I thought, or would have wished.

Agricultural Revolution: ~5,000 years
Industrial Revolution: ~150 years
Acceleration factor: x 33
150 (Industrial Revolution) ÷ 33 = 4.55, rounded up to 5 years

Since each revolution rests on the one before it and feeds on its own outputs (for example, steel mills produce steel quickly and in great quantity, steel that is in turn used to build steel mills and other industrial facilities, etc.), we should expect things to go much (much) faster each time.

ChatGPT-3.5 was launched on November 30, 2022. The seasoned forecasters of the AI Futures Project now predict that 2027 will bring us the “superhuman AI researcher”1 (models better than humans at artificial intelligence research) and “top-expert-dominating AI” (TED-AI)—which would lead us straight to superintelligence.

Why do I set superintelligence as the end point of the Artificial Revolution? Because—to put it bluntly—it is the end point of everything. Once superintelligence exists, the metaphor of the singularity reaches its perfection: we will have crossed the horizon, there will be no turning back, the die will have been cast. Whether or not human beings disappear right away, it will be, one way or another, the true “End of History,” since we will no longer be able to write it. We will no longer steer our destiny, in the same way that chimpanzees have no control over their future, or over how much forest we set aside for their habitat.

When Mark Zuckerberg says something like “Everyone will have superintelligence in their pocket,” it amounts to saying: “Everyone will have an atomic bomb in their bag,” or “Everyone will have the bubonic plague in their handkerchief.” It is both idiotic and suicidal.


My mother is supposed to come see me in Thailand early next year; my nephew is also coming to take the first backpacking trip of his life with his friends. Sometimes I tell myself I should warn them. Many experts foresee a loss of control within 3 to 18 months… At the lower end of that range, that’s February or March 2027. Even without a total loss of control, it’s not certain that all our infrastructure will hold up in the short term—just look at the hacking incidents of recent months and the string of outages in the United States.2 At any other time—in 2005, in 2015, or even in 2025—I wouldn’t have had the slightest worry, but in 2027, I’m not 100 percent sure their plane will be able to take off when it’s time to fly home.

Don’t tell my mother: I don’t want her to fret.


When I started this journal on November 1, I thought I would talk more about writing and literature. I have an experimental project I’ve started coding (read: having Claude Code code). But, frankly, I can’t ignore the latest news from Anthropic, from METR, and (this week) from OpenAI—Sam Altman’s lab has just announced increased capabilities for its GPT-6.2 Astra model3 in AI research and development, which points to a move toward recursive self-improvement.

I don’t understand how, as a writer or an artist, in this month of November 2026, anyone can look the other way. I don’t understand how anyone can fail to feel the urgent need to look this enormous medusa in the eye, even at the risk of being turned to stone.

I know a lot less about the technical side of AI than guys like Zvi Mowshowitz, for example—whom I can’t recommend highly enough that you follow on Substack. All I add, with my journal, are the shreds of subjectivity inherited from a dying literature. If we manage to make it through the next few years, a few sensory traces will remain of the most perilous days and months in human history.

  1. AI Futures Project, “Q3.5 Timelines Update: Looks Like the Intelligence Explosion Will Happen in 2027 After All,” November 10, 2026. ↩︎

  2. Even if these outages were not caused by cyberattacks, they are probably linked to AI, according to several sources. To guard against future attacks, institutions are currently working to strengthen the security of their computer systems, often using AI, which can cause repeated crashes. See “Tech outages spark fears over AI and cybersecurity,” Axios, November 17, 2026; “Outages in US cause worry, despite official attempts to downplay concerns,” Al Jazeera, November 20, 2026. ↩︎

  3. OpenAI, “Introducing GPT-6.2 Astra and Sol,” November 17, 2026. ↩︎

Share

To support me, buy one of my books.