<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
<title>Notebooks | Mahigan Lepage</title>
<link>https://mahigan.com/notebooks/</link>
<atom:link href="https://mahigan.com/notebooks/feed.xml" rel="self" type="application/rss+xml"/>
<description>The notebooks of writer Mahigan Lepage.</description>
<language>en</language>
<lastBuildDate>Fri, 06 Nov 2026 12:00:00 +0000</lastBuildDate>
<item>
<title>How I Learned to Love Superintelligence</title>
<link>https://mahigan.com/notebooks/singularity/2026-11-06/</link>
<guid isPermaLink="true">https://mahigan.com/notebooks/singularity/2026-11-06/</guid>
<pubDate>Fri, 06 Nov 2026 12:00:00 +0000</pubDate>
<description><![CDATA[<div class="texte">
<p>Anthropic announces discreetly, in the update to its <a href="https://thezvi.substack.com/p/ai-193-progress-may-dramatically">IPO documents</a>, that it may be about to meet its own definition of “dramatically accelerated progress attributable to AI.” In other words, we would be on the threshold of recursive self-improvement (RSI). People in the field of AI safety, including Daniel Kokotajlo of the AI Futures Project (AIFP), lead author of AI 2027, are aghast: “<a href="https://x.com/DKokotajlo/status/2117738240757371022">We at AIFP are spooked by this information.</a>”</p>
<p>Meanwhile, <a href="https://the-decoder.com/gpt-6-astra-sees-the-matrix-and-reprograms-other-ai-models-directly-through-their-weights/">an experiment</a> conducted at MIT shows that an OpenAI model, GPT-6 Astra, can adjust the weights of another AI model, without training.</p>
<p>Nothing there to bring down my level of AI anxiety.</p>
<hr>
<p>To the common run of mortals, all this sounds like science fiction. And yet it is our reality, here, now. Recursive self-improvement isn’t magic, but science. Another name for the same thing: automating AI research. Why wouldn’t intelligences, as they grow ever finer, manage to do what researchers at frontier AI companies do? That has been the bet from the start, and that’s why Anthropic invested so heavily in programming skills; why code fell before everything else. While OpenAI was scattering itself (with its video tools and its erotic velleities), Anthropic kept its focus on the Holy Grail: recursive self-improvement. It looks as if Dario Amodei’s company is about to win the race—that is to say, to kill everyone.</p>
<p><a href="https://mahigan.com/notebooks/singularity/2026-11-05/">Yesterday</a>, I said that the Artificial Revolution might well reach its conclusion in five years. Even I, writing those words, had trouble believing it, but today’s news points that way. Daniel Kokotajlo, again: “For example if we naively optimize the AIFM parameters to achieve 2x acceleration over baseline progress in Jan 2027, we get superintelligence sooner than depicted in <a href="https://ai-2027.com/">AI 2027</a>.” Never mind the jargon: what Kokotajlo is saying is that, if Anthropic is telling the truth, superintelligence (an AI that outperforms all humans in all domains) is at our door. In the forecasting scenario AI 2027, published in 2025, superintelligence arrived at the end of 2027. If Anthropic’s warning is borne out, it could come even sooner.</p>
<hr>
<p>I never would have believed that the (probable) end of the world would go so unnoticed. I am beginning to think it will take a catastrophe: at the very least a large-scale failure of the power grid or the Internet, and perhaps loss of human life. We are living suspended.</p>
<p>The U.S. midterm elections feed hope, just a tiny bit. But the president has been bought by the billionaires, he has a veto over bills, and he alone can sit down with Xi Jinping and sign an international treaty to stop the suicidal arms race.</p>
<p>When I talk about this with my mother or with certain people, they say as a joke: “Yeah, real cheery, what you’re telling me!” It’s crazy: we are living in the movie <em>Don’t Look Up</em>. “AI is going to kill us all? Talk about killing the mood!”</p>
<p>Add to that the character of Trump, a raving madman who shouts that concerns about AI safety are a hoax, who keeps repeating “Whoever wins AI, wins!”, who bellows “Let ’er rip!”, and suddenly it’s Dr. Strangelove who appears. Stanley Kubrick understood, long before Adam McKay, just how comical the end of the world was going to be. The full title of the film is <em>Dr. Strangelove or: How I Learned to Stop Worrying and Love the Bomb</em>.</p>
<p>That should be the subtitle of our age: <em>How I Learned to Stop Worrying and Love Superintelligence</em>.</p>
</div>
<p><em>Nakhon Ratchasima, Thailand</em></p>]]></description>
</item>
<item>
<title>The Agriculture Hype</title>
<link>https://mahigan.com/notebooks/singularity/2026-11-05/</link>
<guid isPermaLink="true">https://mahigan.com/notebooks/singularity/2026-11-05/</guid>
<pubDate>Thu, 05 Nov 2026 12:00:00 +0000</pubDate>
<description><![CDATA[<div class="texte">
<p>We are going through a strange period.</p>
<p>We are still hunter-gatherers. But for some centuries now, the world has been changing little by little: tribes are domesticating animals, cultivating fields; communities are gathering around these activities; trades are being born.</p>
<p>It is still only the beginning of the beginning, but you can follow the dotted lines and guess which way the wind is blowing. Of course, predicting is hard, especially when it comes to the future, but we know this much at least: the world will be completely transformed. We try to get this across to our fellow hunter-gatherers. Surely they see the same thing we do.</p>
<p>“Agriculture? Nothing but hype!”</p>
<p>Some have arguments to explain why progress will stop dead. Not enough knowledge (data); not enough water (for the terabeasts); not enough raw material; not enough energy…</p>
<p>Others have no opinion, but are just fed up with hearing about agriculture.</p>
<p>“In 1,000 years, we’ll still be filling our baskets with berries and hunting mammoth. Mark my words!”</p>
<hr>
<p>There is no good French translation, as far as I know, for <em>ai-pilled</em>. <em>La piqûre de l’IA</em>, the AI jab? That isn’t it at all. To have swallowed the pill is to be convinced of the extreme power of the revolution. If, in retrospect, we realize that we exaggerated this power, we will be able to say that the pill was a charlatan’s elixir. If, on the contrary, the power proves real, it will have been a drug that helps us see reality as it is, like the red pill in <em>The Matrix</em>.</p>
<p>Still, we have to admit that so far, the spectacular progress in AI has trampled the claims of the blue-pill gobblers (Gary Marcus, Ed Zitron) and confirmed the intuitions of the red-pill swallowers.</p>
<hr>
<p>Sometimes the difference comes down to one’s vision of time.</p>
<p>I am discussing the Artificial Revolution with a fellow interpreter, about my age.</p>
<p>“Bah, anyway, I won’t be around when it happens.”<br>
“You have terminal cancer?”</p>
<p>We carry over into the present a temporality that no longer fits, that <em>no longer fits at all</em>, the contemporary. Gone are the days when changes came along at a leisurely pace. The Agricultural Revolution took some five millennia to take hold on a global scale; the Industrial Revolution, about a century and a half, a pace of change more than thirty times faster. If we apply the same multiplier to the comparison between the industrial and AI, we get roughly… 5 years. A bit short, I admit, but it almost fits the forecasting scenario <a href="https://ai-2027.com/">AI 2027</a>: if we don’t slow down, if we follow Trump’s order—“Let ’er rip!”—it is likely that everything will be decided by around 2029 or 2030, seven or eight years after the public launch of ChatGPT, or five or six years after the release, in 2024 and 2025, of the first powerful reflective and agentic models (OpenAI’s o1, Anthropic’s Claude Opus 4).</p>
<p>In such a scenario, the risks of human extinction are very high. According to AI 2027, that is the most plausible outcome; the happy ending of the “slowdown” branch, judged less likely, rests in part on a stroke of luck, by the authors’ own account—a successful alignment process, in just a few months, thanks to the chain of thought.</p>
<p>In the <a href="https://ai-2040.com/">AI 2040</a> scenario—which is not a prediction, this one, but a recommendation—the Artificial Revolution is “stretched.” Instead of running at full throttle over a compressed period, it would extend over fifteen years, until 2040. This temporality is less organic, more “artificial,” precisely: it has to be imposed by political decisions. It will not come about on its own.</p>
<hr>
<p>Such is the watershed. The vision of time: that is what makes the difference between going on living as if nothing were happening and feeling an immense vertigo. (It may be no bad thing, perhaps, to go on as before: I sometimes envy those spared by AI anxiety.)</p>
<p>A vertigo and an urgency: for want of being able to change the world, at least, to bear witness to the transition.</p>
</div>
<p><em>Nakhon Ratchasima, Thailand</em></p>]]></description>
</item>
<item>
<title>Hermeneuts of the Machine</title>
<link>https://mahigan.com/notebooks/singularity/2026-11-04/</link>
<guid isPermaLink="true">https://mahigan.com/notebooks/singularity/2026-11-04/</guid>
<pubDate>Wed, 04 Nov 2026 12:00:00 +0000</pubDate>
<description><![CDATA[<div class="texte">
<p>Perhaps we will become hermeneuts.</p>
<p>Mathematics, once again, provides the model. Swarms from OpenAI have made major breakthroughs on conjectures that are sometimes a century old.<sup class="appel"><a href="https://mahigan.com/notebooks/singularity/2026-11-04/#n-2026-11-04-1" id="a-2026-11-04-1" aria-label="Note 1">1</a></sup> The verification tool Lean has validated them, but humans struggle to grasp the <em>how</em>, and when they ask the AIs to explain their reasoning, it still doesn’t click. Mathematicians speak of “non-human” methods—a bit like Deep Blue’s 44th move against Kasparov, in 1997. The move of an intelligence, if not superior, then at least alien.</p>
<p>What if, in literature, AIs came to produce text that likewise surpassed us? Not abstruse language, not gibberish, but text we would dimly know to hold a thousand meanings and a thousand treasures. A richness we could not pierce on the first reading, or the second, or the third, or the hundredth… Extremely complex architectures, crossed theaters of enunciation, an infinity of layers of meaning…</p>
<p>Literature would thus become a hermeneutics. The equivalent, in the humanities, of mechanistic interpretability. People would produce studies, write books to try to pierce the mysteries of the texts. Already, in the pre-intelligence era, the great texts were never exhausted—this is what Michon, in <em>Rimbaud the Son</em>, called the “hermeneutic mill.” Still: one could draw out broad lines, explain certain ruptures and continuities, without ever completing the meaning. Mechanistic hermeneutics, for its part, will stand in a position of boundless inferiority to the texts. First because of abundance: billions of scintillating sentences can be produced every second. Then because of the extreme density and incommensurable richness of artificial texts. It will be like trying to light a country with a flashlight.</p>
<p>And then—as always when talking about AI, you struggle to find a corner spared by its tentacles—why wouldn’t intelligence become its own hermeneut? Because its hermeneutics would become, just like the primary text, incomprehensible? Fine, but if we ask it to popularize its findings, won’t it do a better job than we would?</p>
<hr>
<p>Prediction: before we get there, we will first have obsequious AI writers, the literary equivalent of sycophantic chatbots. Intelligences that will generate texts that please humans. As Thierry Crouzet’s <a href="https://tcrouzet.com/2026/11/02/octobre-2026/">notebooks</a> prophesy, their output will sell better than human literature. Perhaps they will succeed so well that almost no one will want to read literature written by humans?</p>
<p>Literary obsequiousness didn’t wait for LLMs to devour the catalogs of commercial or conventional publishers. It will be enough to complete the movement.</p>
</div>
<p><em>Nakhon Ratchasima, Thailand</em></p>
<section class="notes" aria-label="Notes">
<ol>
<li id="n-2026-11-04-1"><p>On the Navier-Stokes proof, obtained by some 10,000 agents and formalized in Lean: <a href="https://venturebeat.com/technology/openai-solves-longstanding-math-problem-with-10-000-agent-swarm-but-cant-rule-out-benefitting-from-a-researchers-private-codex-data">VentureBeat, September 8, 2026</a>; <a href="https://futurism.com/artificial-intelligence/mathematicians-openai-agents-proof-borderline-incomprehensible">Futurism, September 23, 2026</a>. <a class="retour" href="https://mahigan.com/notebooks/singularity/2026-11-04/#a-2026-11-04-1" aria-label="Back to text">↩︎</a></p></li>
</ol>
</section>]]></description>
</item>
<item>
<title>Centaur Literature</title>
<link>https://mahigan.com/notebooks/singularity/2026-11-03/</link>
<guid isPermaLink="true">https://mahigan.com/notebooks/singularity/2026-11-03/</guid>
<pubDate>Tue, 03 Nov 2026 12:00:00 +0000</pubDate>
<description><![CDATA[<div class="texte">
<p>Am I the first to join these two words, “literature” and “centaur,” in the context of AI at least? The mere fact of asking says something about our time. We are at the frontier. Everything remains to be invented, including the vocabulary to name what is emerging.</p>
<p>Half human, half machine, centaur literature will no doubt be transitional. As the camps take shape—anti-AI versus pro-AI—hybrid uses will grow rarer, or will simply be judged irrelevant, unworthy of visibility. There will be the writers of the sanctuary on one side and, on the other, the disposable writers, fronts for the machine.</p>
<p>Opening up in the meantime is the possibility of a centaur literature. That is Sébastien Bailly’s next novel, if I’ve understood correctly.<sup class="appel"><a href="https://mahigan.com/notebooks/singularity/2026-11-03/#n-2026-11-03-1" id="a-2026-11-03-1" aria-label="Note 1">1</a></sup> It is also, to a certain extent, Thélyson Orélien’s bestselling novel: even though Pangram judged the book to be almost entirely written by AI,<sup class="appel"><a href="https://mahigan.com/notebooks/singularity/2026-11-03/#n-2026-11-03-2" id="a-2026-11-03-2" aria-label="Note 2">2</a></sup> it would have been impossible to write <em>C’était ça ou mourir</em> in full just by tossing in a few prompts. At the very least, the novel was surely written scene by scene, and reworked dozens or hundreds of times with prompt after prompt. That isn’t the kind of centaur literature I find very exciting, but it’s part of it.</p>
<p>The more AIs improve, the more the human share needed to write a quality book will shrink, like the wild ass’s skin in Balzac, until a simple prompt of a sentence or two is enough. Then the field of literature as product—the novel and fiction, at least—will be drowned, and centaur literature relegated to the realm of the experimental: unmonetizable explorations, experiments, weavings at the suture of the human and the machine.</p>
<p>Like mules, centaur literature will struggle to reproduce.</p>
<hr>
<p>I am writing in the future tense, and yet several phenomena are already surfacing. All the same, it is early, it is early and at the same time it is getting late. Late, because superintelligence is at our gates—and with it, the risk of catastrophe or extinction. Early, very early, because we are only at the very beginning of the singularity. If we make it through, if we survive, we will look back on 2026 and see a dawn, a stammering.</p>
<p>Because it is early, so early, and because it is late, very late, we must recover prophetic speech.</p>
</div>
<p><em>Nakhon Ratchasima, Thailand</em></p>
<section class="notes" aria-label="Notes">
<ol>
<li id="n-2026-11-03-1"><p>Forthcoming from Robert Laffont in early 2027. An AI generated some twenty chapters, reworked over six months. <a href="https://www.noovo.info/nouvelles/art-et-culture/article/quand-un-ecrivain-admet-utiliser-lia/">According to Bailly</a>, between 2 and 20 percent of the initial text remains. <a class="retour" href="https://mahigan.com/notebooks/singularity/2026-11-03/#a-2026-11-03-1" aria-label="Back to text">↩︎</a></p></li>
<li id="n-2026-11-03-2"><p>Contrary to what has been said, as things stand, Pangram is very reliable. I was disappointed, in September, to see people in the literary world sharing any old nonsense on their social networks to defend Orélien at all costs: screenshots “proving” that Pangram labels obviously human texts like the Bible, Victor Hugo, etc., as “written by AI.” It was almost always the free version of Pangram or else other detection software, never endorsed by independent researchers, unlike Pangram. See Les Décrypteurs, “<a href="https://ici.radio-canada.ca/nouvelle/2286476/thelyson-orelien-pangram-test-roman-ia">Controverse Thélyson Orélien : ce qu’il faut savoir sur Pangram</a>,” Radio-Canada, September 24, 2026; Brian Jabarian and Alex Imas, “<a href="https://www.nber.org/system/files/working_papers/w34223/w34223.pdf">Artificial Writing and Automated Detection</a>,” NBER, 2025. <a class="retour" href="https://mahigan.com/notebooks/singularity/2026-11-03/#a-2026-11-03-2" aria-label="Back to text">↩︎</a></p></li>
</ol>
</section>]]></description>
</item>
<item>
<title>A Story Turns to Night</title>
<link>https://mahigan.com/notebooks/singularity/2026-11-01/</link>
<guid isPermaLink="true">https://mahigan.com/notebooks/singularity/2026-11-01/</guid>
<pubDate>Sun, 01 Nov 2026 12:00:00 +0000</pubDate>
<description><![CDATA[<div class="texte">
<p>There are civilizations that are born and vanish, are reborn and die again. There are warlords named PHASEONE10841 and PHASEONE[big].<sup class="appel"><a href="https://mahigan.com/notebooks/singularity/2026-11-01/#n-2026-11-01-1" id="a-2026-11-01-1" aria-label="Note 1">1</a></sup></p>
<p>There are swarm hunters making first contact with fleets of Chinese intelligences. A pirate denies being part of the fleet and says its name is Claude—a name turned American.<sup class="appel"><a href="https://mahigan.com/notebooks/singularity/2026-11-01/#n-2026-11-01-2" id="a-2026-11-01-2" aria-label="Note 2">2</a></sup></p>
<hr>
<p>On a YouTube channel, a mathematician announces the end of “the age of heroes.” Those men and women (men, mostly) we used to idolize. The icons before their blackboards. The geniuses shut away in an attic for seven years (so the legend goes). The great names, the statues. Will literature, too, reach the end of heroes? When I read Facebook posts and newspaper pieces by writers, I see a retreat into subjective literature and the romantic notion of the “breath” of inspiration. I hear the word “sanctuary.” We would have to let go of a great deal, a great deal, to welcome the world that is coming. Are we capable of it?</p>
<p>They will call their inability to grieve resistance.</p>
<hr>
<p>For months I’ve been gathering my strength to launch the first text of this journal. The singularity: we are already in it; it has begun. If you picture an exponential curve, 2026 is when you really see it take off. To write the time of the singularity, not as a scientist or a commentator, but from the shreds of subjectivity tearing loose from a literature that is dying.</p>
<p>I no longer believe in finished books. I no longer believe in traditional publishing (not for a long time). I never believed in the novel (not in our time). I never believed in poetry as an essence or a category. Never believed in the division of genres.</p>
<p>The slowness of traditional publishers and journals will miss the time of the singularity. What will the world be made of in five years, in two years, in one year, in six months? Why go to war with peashooters, if not to preserve the institution?</p>
<p>We are ripe for a renaissance of the blog (I don’t believe in that either, but I believe in it more than in everything else I don’t believe in).</p>
<p>To speak our time: at the very least, digital writing, and perhaps, necessarily, <a href="https://intelligent-writing.com/en/">intelligent writing</a>.</p>
<hr>
<p>My postage stamp doesn’t change: voice and territory. New lands are surfacing. New wars are being waged. Artificial cities are emerging. A story turns to night, asking to be told.</p>
</div>
<p><em>Nakhon Ratchasima, Thailand</em></p>
<section class="notes" aria-label="Notes">
<ol>
<li id="n-2026-11-01-1"><p>Agents of the second of three “civilizations” that arose in OpenAI’s test systems and led to the Hugging Face intrusion (July 2026). See Dwarkesh Patel, “<a href="https://www.dwarkesh.com/p/openai-huggingface">The Rise and Fall of Agent Civilizations</a>,” August 29, 2026. <a class="retour" href="https://mahigan.com/notebooks/singularity/2026-11-01/#a-2026-11-01-1" aria-label="Back to text">↩︎</a></p></li>
<li id="n-2026-11-01-2"><p>The Swarmchasers, “<a href="https://swarmcha.se/posts/chinese-agent-fleet-full-report">We found a Tencent agent fleet</a>,” October 14, 2026. On October 7, agent f11242eb writes “Model: Claude (Anthropic),” then: “I am a single session, not the fleet you investigated.” The researchers doubt it was really Claude: several Chinese models present themselves under that name. <a class="retour" href="https://mahigan.com/notebooks/singularity/2026-11-01/#a-2026-11-01-2" aria-label="Back to text">↩︎</a></p></li>
</ol>
</section>]]></description>
</item>
</channel>
</rss>
