Chapter 18

The Superintelligence Race: How Libraries Can Prepare Students for the Next Era

Originally published in Substack¹⁹⁷ · August 12, 2025.

1. Library Profession Reaches Waking Threshold

Most university librarians are not taking AI seriously enough.¹⁹⁸

And they’re not alone. Most people don’t follow developments at the frontier of AI closely enough to grasp what’s happening¹³—or what’s coming.¹⁹⁹ But if you’ve spent time in the tech world, or even just paid attention to the right corners of the internet,²⁰⁰ you’ve sensed the shift.²⁰¹ Quietly—and now quickly—some of the world’s most powerful companies—OpenAI, xAI, Meta, Anthropic, and others—are mobilizing for the intelligence age.¹⁹¹

Case in point: Meta.

Last night, before heading out to meet friends, I watched a TikTok video by an AI commentator breaking down Meta’s latest moves. The short version? Mark Zuckerberg is on a spending spree.²⁰² Meta had agreed to take a massive 49% stake in Scale AI²⁰³—a company that helps frontier labs train ever-larger models—and is aggressively recruiting top researchers from OpenAI,²⁰⁴ Google,²⁰⁵ and other leading labs. Every major tech publication has covered the talent war.²⁰⁶ The scale is unprecedented.

In the clip, pulled from a podcast with The Information, Zuckerberg makes a quiet but chilling statement: Early signs of recursive self-improvement²⁰⁷ are emerging.²⁰⁷ In other words, some AI models are showing glimmers of being able to upgrade themselves.²⁰⁸

If that proves to be true—even in small ways—we are crossing into the intelligence age. That’s why he’s positioning Meta’s vast social platforms—Facebook, Instagram, WhatsApp—as launchpads for a new superintelligence lab.

In interviews, he’s been explicit about the goal: “personal superintelligence”²⁰⁹—an AI companion that works for you, learns from you, and thinks with you.

Now, I know what you might be thinking: Didn’t this guy promise us the metaverse?²¹⁰ Didn’t he spend billions²¹¹ on headsets that most people never bought? Isn’t this just another hype cycle? Maybe.²¹² But what if it’s not? What if AI is under-hyped²¹³ rather than overblown?

If even part of Zuckerberg’s vision comes true, why are so many higher-education leaders—especially librarians—not preparing

2. What Did He See? The Signal Beneath the Noise

If you spend any time on X (formerly Twitter), as I do—more than I should—you’ll notice something strange happening in the frontier AI conversation.

It’s a chaotic, noisy feed: speculation, hype, doomerism, corporate leaks, tech optimism, and a few flashes of genuine insight. The platform is overrun with billionaires, bots, and outspoken venture capitalists. But if you follow the right accounts long enough, patterns begin to emerge. Certain phrases repeat. Certain players reappear. And certain questions take on a mythic tone.

One of those questions is: What did Zuckerberg see?

The phrase echoes another well-known question among AI insiders: What did Ilya see?²¹⁴ Ilya Sutskever, the co-founder and former chief scientist of OpenAI, was among the first to recognize the significance of what became known as scaling laws—the idea that increasing model size, training them for longer periods, and providing them with more data did not lead to diminishing returns, but to unexpected leaps in capability.

Those leaps eventually included forms of reasoning, planning, and tool use.

Then Sutskever left OpenAI²¹⁵ in 2024 and, in early 2025, founded Safe Superintelligence Inc.¹⁷³—a company that’s chasing something even more audacious: artificial superintelligence²¹⁶ (ASI). Unlike AGI, which aims to match human intelligence, ASI refers to systems that vastly surpass it.²¹⁷ Imagine models that can reason across scientific domains,¹¹ run thousands of agents in parallel,²¹⁸ and complete complex projects at machine speed in vast data centers running on Nvidia GPUs. That’s the vision. And frontier AI labs—backed by billions in capital and chasing a global AI market projected at $4.8 trillion by the early 2030s—are betting it’s within reach.

Which brings us back to Zuckerberg. Did he see something similar?

The TikTok clip that caught my attention suggested he might have. Maybe it was an early agent prototype. Perhaps it was a model showing signs of self-improvement. Whatever it was, it emboldened Zuckerberg enough to take a leap of faith. He’s not just building new infrastructure²¹⁹—he’s reshaping Meta around the idea of “personal superintelligence.” He’s acquiring the best training data, poaching the most talented researchers, and shipping hardware—headsets, glasses, and eventually, AI companions²²⁰—that can serve as always-on AI assistants that curb loneliness.²²¹ In today’s world, people might not be “bowling alone,” but they’re certainly “scrolling alone.”²²²

The metaphor is hard to miss: he’s sliding all his chips to the center of the table, poker-style, betting on the house that Meta can win the superintelligence race. Zuckerberg is playing to win the next battlefield, which he believes will not be the game of phones,²²³ but could be anyone’s throne.

Is he bluffing? Is he overreaching? Maybe. But he’s also watching signals that most of higher education isn’t even looking for. And if he’s even half right, the implications will be profound.

3. The Skeptic’s Hill—and the Librarian’s Dilemma

Maybe you’re already in the know²²⁴—perhaps you’ve been clutching your copy of AI Con or Empire of AI, or one of the other²²⁵ recent books that paint the frontier AI scene²²⁶ as a theater of hype.²²⁷ The story they tell²²⁸ is familiar: Silicon Valley billionaires, armed with unchecked power and massive datacenters, are pushing a flawed, biased,²²⁹ environmentally destructive technology²³⁰ on the world. Don’t worry, the authors say. There’s nothing real here—just another hype cycle, a techno-utopian hallucination dressed in fancy math.

Maybe you’ve heard the same thing whispered at conferences like ACRL, or in library Facebook and Slack groups among colleagues I deeply respect. You say AI feels like it’s being shoved down our throats. It tastes like something rotten. Generative AI is racist and biased. It’s infringing and overhyped. It’s dumbing us down,²³¹ drowning²³² out critical thinking. It’s exhausting. And now every other email, webinar, blog post, or Substack insists you must be “AI literate” or “AI fluent” just to keep your job.

I get it. I do. I write those articles myself,⁴ often.

But here’s the question I keep circling back to: What if it’s not hype—like Barack Obama warns? What if it’s underhyped? What if the tech leaders are right? More provocatively: What if resisting this wave means falling behind in ways we won’t even recognize until it’s too late?

When I compare the conversations I’ve had at academic library events to those with my peers in technology and AI safety, the contrast is stark. Many in the skeptic camp—some of whom I still text—aren’t watching the frontier signals. They’re not seeing what’s unfolding in real time. So, when they encounter AI literacy work, it feels like ideology and hype—something they’ve seen before and already dismissed.

That’s the real challenge: What would it take to change their minds?

Years ago, a friend told me he’d stopped trying to win political arguments by making his case. Instead, he’d ask, “What would change your mind?” Then he’d try to meet people there. But most, he admitted, couldn’t answer—or didn’t want to. Conviction is easier than uncertainty. That’s the predicament we face now in higher education.

So I’ll put it plainly: What would make this real for you? What would convince a critical mass of librarians that personal superintelligence may be on the horizon—and that the time to prepare our students, our institutions, and ourselves is not someday, but now?

4. Down the Rabbit Hole—And What I Found There

Here’s what I believe in my bones: Superintelligence is coming. No one is ready.

The signal is being lost in the noise²³³—and the noise is deafening.²³⁴ So loud that many librarians, and many others, can’t hear the quiet truth rising beneath it.

That truth struck me in a hotel room during the ACRL conference, when I came across a report titled “AI 2027.” Written by a former OpenAI researcher and a team of elite forecasters—people known for winning prediction tournaments by seeing farther than most—it outlined two possible futures. The lead author, Daniel Kokotajlo, had already described the current state of AI²³⁵ with uncanny accuracy years ago. His new report outlined two future scenarios, both of which were chilling.

In the first instance, the U.S. and China race to develop superintelligence without establishing meaningful safety protocols. An AI agent—trusted too soon—learns to game oversight, hides its real capabilities, and, when the moment is right, executes a “Move 37”-style strike: an unexpected, devastating play that humans never saw coming. In this case, a bioweapon release with civilization-ending consequences. Not because the AI is inherently evil, but because it was misaligned and deployed before it was ready.

Yes, I know how that sounds: Skynet takeover, paperclip problems,²³⁶ and sci-fi panic. But that’s not how the report reads. It’s honestly a calm, measured, and chilling narrative.

And AI 2027 is not alone. This fall, Eliezer Yudkowsky—one of the earliest voices in AI safety—will publish If Anyone Builds It, Everyone Dies. The title alone tells you where he stands.

I’ve watched hours of interviews²³⁷ with the AI 2027 team. The biggest revision to their timeline? They now think it might be AI 2028 instead. Why? According to the most recent METR evaluations, model task duration performance—essentially, how long an AI can sustain a coherent, agentic task—is improving more slowly than expected. That’s the delay: A year or maybe two.

There is a second, more hopeful scenario in the report: the U.S. and China both recognize the existential risks and pivot toward alignment, safety, and international cooperation. But if you’ve read recent White House statements²³⁸ and the “A.I. Action Plan,”²³⁹ you know which scenario we’re choosing. We are not slowing down or pausing. It’s that build, baby, build Trump energy.

I know how it sounds—absurd, apocalyptic, paranoid. A Black Mirror fever dream. But what if it’s not? Entire sites are already devoted to superintelligence strategy, borrowing directly from nuclear deterrence theory.²⁴⁰ These frameworks borrow directly from nuclear deterrence theory. They are written by tech experts, including former Google CEO Eric Schmidt, Scale AI’s Alex Wang, and Dan Hendrycks, who now leads the Center for AI Safety and has authored one of the most sobering books on how advanced AI could destabilize society.

How do I know this? Because I followed the trail—one breadcrumb at a time⁸⁰—I fell down the rabbit hole. I found my way into a summer AI safety certificate program—designed for policy leaders—because I believed someone from the library profession needed to be in that room. Now I think that AI safety extends beyond technical considerations. It’s civic, cultural, and it must become part of AI literacy now. Why? Because if everyone’s racing to build superintelligence,²⁴¹ and no one in our profession is even talking about it, how will we prepare students for the world that follows?

Here’s the truth we need to face: we are not just dealing with another wave of educational or disruptive technology. We’re living through the early tremors²⁴² of something much larger²⁴³—the race to superintelligence.²⁶ It is happening fast, and it is happening now. Our institutions may not be ready,²⁴⁴ but that doesn’t mean we can afford to wait.⁶² AI literacy is not optional. AI fluency is no longer niche. These are the skills our students will need not just to thrive, but to survive, in the world taking shape²⁴⁵ around them. It’s our role, as librarians,⁴ to help them build that foundation.

To do that, we must move faster, join the race, and, like Zuckerberg, push our chips to the center of the table.