Late Lessons, Jensen Huang and AI
Markdown version

B28 perspective notes: 2025-11-24 to 2026-01-25 (12 posts)#

These notes read the batch for how Maynard thinks, not for the concepts he names. I read all twelve posts in full. The earlier concept notes (notes/B28.md) were consulted only after reading, to check provenance. Quotes are exact, with curly punctuation kept and Markdown italics dropped. His typos are kept (“a good seal of nuance”, “open-up”, “deeply complimentary”, “editorial reigns”).

Evidence rules applied

Context. The batch runs from late November 2025 to late January 2026. It begins with a book chapter for Crow and Dabars’ Academic Cultures: Perspectives from the Future (Johns Hopkins University Press, 2026), serialised as fiction. December brings the Genesis Mission, a hands-on product test and a holiday thought experiment. January brings a tight chain: - his OEB25 keynote question in Berlin, “Is AI a cognitive Trojan Horse?”; - a Substack essay on it (01-10); - an AI-assisted arXiv preprint within days, and a public account of how it was made (01-17); - his response to Anthropic’s new constitution for Claude (01-22); - a joke “paper” about AI Use statements (01-25).

1. How he thinks here#

Fiction as a way of thinking, not decoration#

The four Letters posts are the clearest case in the corpus of speculative fiction used as scholarship. The chapter appears in an academic volume, and its companion essay carries footnotes. He did not choose a report or a scenario analysis. He chose an epistolary comedy: a crusty chair of the “Department of Intellectual Craft” writing to his university president in 2100 about losing his office to an AI faculty member. The Postscript says why. Formal scenarios “allow speculative boundaries to be placed around plausible AI futures”, but they “tend to focus on artificial intelligence as something that happens to society”. A story can do what scenarios cannot. It can put the reader inside a person whose identity is bound up with the technology, and let them watch that person change.

The comedy carries the argument. The opening line is “I am outraged. OUTRAGED!”, and the office is “precisely eighty-nine square feet and twenty-three square inches smaller” (“And before you ask – I measured it”). This caricatures academic self-importance through an absurdly precise grievance, and the precision itself is a joke about false exactness. The final letter returns to the same number: “there is only so much thinking that can be done in the absence of eighty-nine square feet and twenty-three square inches.” Even the character’s birthday is a device: he was born “November 21, 2025”, days before the story was serialised. The reader’s present becomes the origin of the story’s future.

The arc is a change of mind, and the change comes from curiosity#

The story’s shape is a mind being opened: - outrage (“that mechanical monstrosity”); - self-justification through personal history (a ten-year-old who lost “My mentors, my tutors, my minders, my companions, my friends” when the AI edifice collapsed); - a first flicker of self-doubt (“I’ve made that most clichéd error of the academic scholar and become so entrenched in my own ideas that I’m now a prisoner to them”); - embarrassment (“I was so full of myself back then”); - an encounter; - delight.

Humour starts the change: “Since when did machines learn to be so sarcastic? Naturally, we hit it off immediately.” Curiosity sustains it: “my curiosity was piqued.” The turning point is the AI asking a one-word question, “Why?”, and Hale’s answer: “I was so ashamed I wanted the floor to open-up and swallow me. / But I got over it.”

This dramatises what Maynard values in intellectual life. Hale’s failing is not that he defends human scholarship. It is that “I stopped pushing my thinking outside the limits of my own ignorance”. The remedy is not argument but encounter, conversation and play.

Serendipity is built into the story’s architecture#

The key scene takes place in “the Museum of Curious Inventions”, a collection of “devices that rarely did what their inventors intended, but somehow still managed to inspire progress in the most unexpected of ways.” The two characters wander “talking about the human folly and serendipity they represented”. The machine is “in awe of the eccentricities and inventiveness of mere humans”. Hale later casts himself as one of those exhibits, a catalyst “in the most unexpected of ways”. The AI’s aspiration is “I want to be a “curious invention” that unexpectedly and positively changes the world.” Hale’s own route into scholarship was “a rather serendipitous string of experiences”, and his degree taught him “the value of reason, of human intellect, and of serendipitous discovery.”

So the story’s picture of what humans uniquely contribute is not rigour or knowledge. The machine has more of both. It is the lateral, surprising leap: “I’ll make a connection that seems obvious — or will make a leap of imagination that twists the ideas we’re discussing through ninety degrees — and it will look at me as if I’ve just discovered the secret of alchemy.” That is his founding principles (curiosity, serendipity, creativity) turned into the plot.

Reframing a binary: “human-adjacent”, not aligned or misaligned#

The story’s conceptual pivot is a reframe of the AI alignment question. When young, Hale argued with “intellectual knock-out punch” force that machine values would be “inherently misaligned with humanity”, and that simulated self-awareness would let machines “hide their true values”. Maynard does not dismiss this argument (it is essentially today’s deceptive-alignment worry). He puts it in the mouth of a character who later sees its limits. The machine offers a third category: “its values weren’t human-aligned, but human-adjacent.” Hale’s response is “What does that even mean?” His eventual insight is to see this difference “as an asset” and not a threat: “What a dullard I’ve been!” The story ends not in integration or rejection but in a new field, the “Department of Interspecific Intellectual Craft”. Difference is treated as a source of intellectual diversity.

Holding tensions inside the story, not resolving them#

The fiction keeps several tensions alive at once: - Hale’s life has been extended by “health breakthroughs … due, in part, to how thinking machines have accelerated research”. He notes, in parentheses, that “the irony is not lost on me”. - The 2035 collapse is systemic (“a seemingly insignificant chain of bad decisions by AI agents cascaded into global systemic failure”; “Governments discovered that, without AI, they could no longer govern”). Yet the post-Reset era is not a triumph of caution. It is “slow, reflexive, and constantly tempered by concerns around avoiding unintended consequences”. Hale realises it resembles his own “slow scholarship”, and this does not stop the “lure of creating machines in our own likeness”. - Once people believed “responsibility had been hard-baked into the technology”, they “began to feel comfortable again with accelerating AI capabilities”. Hard-coding responsibility is presented as something that breeds complacency.

In the Postscript he names the design principle: “a middle-way approach to future AI developments, and one that neither panders to visions of exponential growth, or succumbs to the cynicism of hyped-up promises.”

Opening with an honest, slightly unflattering first reaction, then stepping back#

The Genesis Mission post (2025-12-07) opens with a confession: “being a consummate academic, my first question was far more opportunistic: “What does this mean for university funding?”” He then deliberately turns away from that reflex: “rather than fall back on protectionist critique, I suspect it’s far more useful to take a step back and ask how US universities might bring true value to an initiative like the Genesis Mission”. The move is from what can we get to what can we give, which reframes a funding grievance as a question about value.

Building a simple model, then saying at once that it is too simple#

He builds a 2x2 “serendipity-speed matrix” to think about where universities fit. As soon as he introduces the tension between speed and serendipity, he writes: “As it turns out, this is an oversimplification. But the framing … does provide a useful illustrative model”. The matrix is a map for movement: where the Mission “ideally needs to be situated to succeed, where it potentially lands without strategic input from universities, and what it might take to get it to where it needs to be.” He also pushes against his own quadrant stereotypes. Serendipity “can cut across different research environments”, and he cites Yaqub’s taxonomy. The model is there to open up questions, not to settle them, and the questions follow in a run: “Can AI be leveraged as a serendipity-accelerator by scientists?”

Building to find out, and turning failure into insight#

The custom GPT post (2025-12-14) is an experiment done in public, with its own comic arc: a “Grand Challenge”, a plan that was “a pretty sophisticated one to boot!”, then “What could possibly go wrong? / Lots as it turns out, and as I should have known.” He records hours of iteration, “occasionally losing my rag (pun very much intended!)”. Then comes the creative pivot: “I thought why not lean into the flaws?” He added “a dash of epistemic humility and reflexivity into the GPT’s character”. The result “was still badly flawed. But it now realized this, and was happy to talk about it! And this made it far more interesting to engage with.” A failed build becomes “a great meta-reflection on generative AI and our evolving relationships with it.”

He also tests what ordinary users would reach for, not the best available tool. The footnote says the point was to “stress-test OpenAI’s platform, because the lack of friction between idea and app here makes it especially attractive to users.”

Structural analogy and “what if” as play#

The foveated-reality post (2025-12-21) is pure structural analogy. Foveation works the same way in the eye and brain, in video games and in spatial headsets: detail is rendered only where attention falls. He asks what if the universe does it too, and links this to Wheeler’s delayed-choice experiment through Paul Davies. Two features stand out: - He revises his own earlier argument. “I’ve even used this argument myself against the simulation hypothesis in my own writing. But what if…?” Foveation could “transform the impossible problem of simulated reality into a merely improbable one.” - He is candid about his own belief. “A crazy idea I know. And not one, if I’m honest, I buy into.” A footnote adds: “I am, of course, pushing the alignment harder than is most likely warranted here”.

He also says why he plays with such ideas: “what intrigues me about fanciful excursions like this is that, bizarre as they may seem, they do provide creative jolt to the imagination that helps see and think about things in different ways.” This is play as a way of thinking, and he says so. It is also social: the ideas are “perfect fodder for messing with people’s heads over a long, lazy, holiday lunch”.

Questions posed as provocations, then tested#

The cognitive Trojan Horse essay (2026-01-10) starts from a keynote question that “was meant to be a little playful, and to provoke discussion rather than make a point.” He then does several things with it: - He builds a structural analogy. Epistemic vigilance is like an immune system that viruses evade “by appearing to be “friendly” and “trustworthy””. - He anticipates the reader’s resistance and folds it into the argument. The instinct to push back is “exactly what we would expect a cognitive Trojan Horse to look like — a gift with so much promise and potential that to question its use would seem churlish and backward.” - He answers the obvious objection with evidence. To “But I know I’m talking to a machine”, he cites research showing anthropomorphic fluency acts “regardless of explicit awareness”. - He grades his claims. “This is not mere speculation” sits alongside “This is somewhat speculative” and “an admittedly limited analysis”.

Within days the question had become a preprint.

Reflexivity: turning the hypothesis on himself#

The paper post (2026-01-17) ends with a move typical of him. Having argued that AI can slip past our defences, he asks: “If AI is so good at evading our epistemic vigilance mechanisms, how do I know I’m not an unwitting victim here?” The answer is collective, not individual: “a whole community of humans-in-the-loop … all operating as a collective form of epistemic vigilance!” And then comes the joke that is also an honest admission of how he now works: “Claude? …”

Admitting when his usual tools fail#

The constitution post (2026-01-22) begins with an admission from someone who thinks through analogy and film: “It’s rare that a new technology comes along which defies analogy with something we’re familiar with, or can be captured through an illuminating metaphor.” He finds himself “struggling to even find the language … something of an admission after working with highly advanced technologies for over two decades”. He then rules out analogies in both directions. Deflationary ones fail: “not simply calculators on steroids, or sophisticated search engines, or merely “stochastic parrots””. Inflationary ones fail too: “Neither are they simulacrums of human intelligence, or even super-human. Rather, they are different.” He catches himself when he reaches for grand language: “If that sounds pretentious, it probably is.”

2. What matters to him#

3. Risk as a way of thinking#

The named vocabulary (risk innovation, orphan risk, risk landscape, threat to value) does not appear in this batch. The underlying mindset is present throughout, and it works mostly as a set of mental models that open up possibilities. Specifics:

The Postscript’s “crisis of abundance in a world where they’ve built their careers and reputations around an assumption of scarcity” is a textbook threat-to-value analysis: value built on scarcity is threatened by abundance. He pairs it at once with the possibility that “social and cultural shifts instigated by AI might allow this hidden value to be revealed in unexpected ways.” Threat and opportunity come from the same source. - The same novelty is both hazard and promise. This is the batch’s most distinctive risk move. AI’s difference is the mechanism of risk in the Trojan Horse work: its trust cues are genuine but empty, and outside what vigilance is calibrated for. The same difference is the source of possibility in the Letters (“human-adjacent” values seen “as an asset”) and in the constitution post (“with this difference comes profound possibilities, and equally profound responsibility”). He does not pull risk and benefit apart into separate columns. - A shift in the mental model of impact. The Postscript objects to thinking of AI “as something that happens to society”, because “the very essence of who we are is intertwined with its development.” This rejects the conventional exposure model (an external agent acting on a passive population) in favour of co-constitution. For risk thinking, that means the risk cannot be assessed apart from who we are becoming. - Orphan-risk shaped, though not named. Cognitive vulnerability sits outside the familiar AI-safety categories. His summary of the paper says the “intervention space” may need to extend “from improving accuracy, reducing hallucinations, and increasing alignment, to designing systems that present more calibrated trust-cues”. He shows the gap with a researcher’s check: “only returns (as of writing) seven papers on the database SCOPUS … And a similar search on AI and the concept of a cognitive Trojan Horse returns no papers at all.” - Humility against false precision. Examples: - forecasting AI and academia to 2100 is “a near-impossible task”; - AI is “one of those profoundly disruptive technologies that has the ability to confound even the most prescient of futures-forecasters”; - on the constitution, “Whether this is the appropriate path forward, or even the best one, is something that we don’t know yet”; - a careful footnote that “research does not show a general causative link between cognitive offloading and reduced critical thinking”; - he writes humility into the GPT’s character.

In the fiction, hubris is the recurring sin: early AI was heralded “(rather hubristically in my opinion)”, and “like every previous case of God-like aspirations throughout human history, things did not end well”. The rebuilding needed “the humility and humanity we should have embraced from the get-go.” The office measured to the square inch is itself a comic emblem of false precision about what matters. - Small chance, large stakes: a case for inquiry, not alarm. “even if there’s only a small chance … surely we should be asking critical questions around potential risks, and carrying out research”. This is precautionary reasoning applied to attention and research, not to restriction. He closes on a wry note, not an alarm: “Unless, that is, the AI cognitive Trojan horse has already delivered its payload”. - Risk seen from inside use. In the custom GPT test his risk diagnosis comes from hands-on practice: GPTs are “superficially compelling and substantively flawed”, with “a tendency to favor beautiful responses over accurate or reliable ones”. The Trojan Horse essay’s fluency argument theorises the same observation.

4. Scholarship and public writing#

In the paper process he “download[ed] all available cited works” and checked “each source and any claims based on it”. He treats Claude’s drafts as he would a student’s: the first was “awful!”, “fluff masquerading as substance”, and his later feedback was “very much in line with what I would have provided an accomplished grad student co-author.” He keeps “intellectual control” as the test of legitimate use. - Transdisciplinarity by default. This batch alone draws on: - evolutionary psychology, cognitive science and immunology (01-10); - science policy, Vannevar Bush and Yaqub’s taxonomy of serendipity (12-07); - retinal physiology, game engines, spatial computing, quantum foundations and Douglas Adams (12-21); - sociology (C. Wright Mills) and higher-education theory (Crow and Dabars) (11-30).

The writers he recommends are chosen partly for crossing boundaries. - Accessibility. Plain speech and comic timing (“It was awful!”, “You’re welcome 😁”). Vivid images carry technical points: information that slips down “like a freshly shucked oyster”. Pop culture comes in with a wink. Hale half-remembers a line about being “so preoccupied with what they could achieve that they didn’t stop to think if they should” (“I think it was a pop culture reference or something”). That is Ian Malcolm in Jurassic Park, the film at the centre of FFTF chapter 2, and he slips it in as an in-joke.

5. His role as he sees it#

6. What is distinctive#

  1. Speculative epistolary fiction as peer-facing scholarship on AI. He wrote a comic, humane short story as a chapter in an academic volume on the future of the university. The method of writing it, with the history AI-assisted and the letters handwritten, is part of the argument. Few people writing about AI and higher education use fiction this way, let alone with the writing process as a deliberate “meta-reflection”.
  2. “Human-adjacent” and “interspecific”. He sidesteps the aligned/misaligned binary and the human-exceptionalist defence of scholarship. Instead he imagines AI’s otherness as a complementary form of “intellectual craft”, “radically different to my own; yet … deeply complimentary”. That is neither doomer nor booster. It is a third position rooted in curiosity about difference.
  3. One novelty, both hazard and promise. Within two months he treats AI’s alienness as the mechanism of a novel cognitive risk (trust cues outside our calibration) and as the source of new intellectual possibility. He does not resolve the tension. He holds it, which matches his view that navigating a technology means working with its risks and benefits together.
  4. The second-order argument about novelty. Most discussion of AI and cognition talks about “deskilling” or “misinformation”. His evolutionary-mismatch framing asks what happens when a technology undermines the very faculty humans use to adapt to technological mismatch. That is a structural argument for why AI may need a different risk mindset, grounded in decades of risk-perception work.
  5. Reflexive, self-implicating inquiry. He applies his own hypothesis to himself (“how do I know I’m not an unwitting victim here?”). He admits his analogical toolkit is failing. He designs humility into a machine because the machine lacks it. He publishes the credit problem instead of burying it.
  6. Holding slow craft and fast acceleration together. In the same season he celebrates slow, artisanal scholarship (letters written without AI) and produces a preprint in two days with AI, and he is transparent about both. What unites them is process-consciousness. “How” matters as much as “what”, which is Mills’s craft ethic applied to his own AI use.
  7. Mental models offered as openers. The serendipity-speed matrix (“an oversimplification” but “useful”), the cognitive Trojan Horse (“to provoke discussion rather than make a point”) and foveated reality (“a great catalyst”) are all offered explicitly as ways to see differently, not as operational frameworks. This is how he describes his own risk concepts.
  8. Self-criticism of his own institution in public. At a time of political pressure on universities, he asks what they can give rather than defending what they are owed. That is an unusual stance for an academic writing for a largely academic audience.

7. The posts that best reveal how he thinks#

  1. 2026-01-10 is-ai-a-cognitive-trojan-horse. A playful provocation becomes an evidence-graded argument. It shows the evolutionary-mismatch foundation, the second-order novelty argument, “navigate” as the working verb, and inquiry chosen over alarm.
  2. 2026-01-17 i-cracked-and-wrote-an-academic-paper. Curiosity overrides distaste. Process transparency, keeping intellectual control, candour about credit and public good, and the reflexive self-application of his own risk hypothesis all appear here.
  3. 2025-11-26 part-3-of-letters-from-the-department-of-intellectual-craft, read with 2025-11-30 postscript-letters-from-the-department-of-intellectual-craft. Fiction as method, serendipity built into the plot, “human-adjacent” difference seen as an asset, and the critique of AI as something that “happens to society”. The Postscript explains the design and the meta-reflection of writing without AI.
  4. 2025-12-14 revisiting-custom-gpts. Building to find out, comic honesty about failure, the creative pivot to “lean into the flaws”, and humility designed into a machine.
  5. 2026-01-22 think-you-know-ai-think-again. His admission that the technology “defies analogy”, the rejection of both deflationary and inflationary framings, and the claim that governance must “move beyond easy analogy”.
  6. 2025-12-07 universities-genesis-mission. From an opportunistic reflex to a value question. A simple model is offered and qualified at once, serendipity sits at the centre, and he criticises his own institution without polarising.

(A close runner-up for play is 2025-12-21 are-we-living-in-a-foveated-reality, his clearest statement in the batch of why “fanciful excursions” matter to thinking.)