Late Lessons, Jensen Huang and AI
Markdown version

B27 perspective notes: 2025-09-07 to 2025-11-24 (16 posts)#

These notes read the batch for how Maynard thinks, not for the concepts he names. Every post was read in full; the Modem Futura note was skimmed for his own framing lines. Quotes are exact, including his typos (“hight efficient”, “theres”, “should being doing”, “AI be useful”, “memory enable”, “versus and unhealthy”).

Evidence rules applied

Context. It is autumn 2025. The co-written book launch dominates the calendar (excluded here). Around it, his own posts turn to the everyday intimacies of AI use: email, ChatGPT’s memory, advisors editing students’ work, students’ mental health, parasocial bonds. Seven lawsuits against OpenAI are filed on 6 November. The batch ends with his first published fiction, written for a Johns Hopkins volume on academic culture in 2100.

1. How he thinks here#

He opens by putting the reader inside a situation#

Three of the risk posts start with “imagine”, in the second person, in ordinary life:

These are not abstract hazards. They are a partner, a parent and a border crossing. His “what if” is concrete, domestic and relational, and each scene points at a relationship that could be damaged. He starts from what it would feel like to be harmed, and only then brings in the analysis.

What starts his thinking: people and things that turn up#

Almost every post starts from something he encountered rather than something he set out to study:

This is serendipity used as method. His early signals come from classrooms, colleagues and conversations, not from incident databases. The classroom in particular works as a sensor for risks nobody has yet written down.

“Intrigued”: curiosity comes first, even when the subject is harm#

“Intrigued” is the batch’s signature word, and it is usually paired with concern:

He approaches risk the way he approaches a curious phenomenon. Worry does not stop the inquiry; he asks for “more”. Risk is something to explore, not only something to guard against.

A “risk hat” he chooses to put on#

“So I thought I’d dig out my risk hat and dive a little deeper—including developing a simple risk model” (2025-09-07). The image matters. Risk analysis is one hat among several, something he “brushed off” for a question that caught his interest. His default stance is curiosity; formal risk tools are equipment he picks up when a question needs them. He then describes what he builds modestly: a “basic 2x2 array”, “an exploratory analysis”, “generic”, “a useful starting point”.

Building things to find out#

He does not just speculate. He sets up an apparatus and reports what happens:

Letting the result surprise him#

“I half expected them to end up being trivial, and was quite taken aback by how potentially serious some of the risks are in this analysis” (2025-09-07). He then reasons back to a mechanism: “On reflection, I’m not surprised”. Organisations run on “relational connective tissue”, and eroding trust has consequences. The pattern is prior expectation, then experiment, then surprise, then an explanation of why the surprise makes sense. It is how a physicist reads an unexpected measurement.

Spotting what has changed inside a familiar category#

The informant post is his clearest example of seeing that a familiar risk has changed. The scenarios are “a play on a privacy risk that’s been around for a while”. “But there’s a twist here”: no more trawling through logs, “just a few well-crafted questions, and your deepest secrets are revealed”. The novelty is the mechanism. ChatGPT is “highly adept at joining the dots and inferring things” that “you never even realized you were giving away”. Footnote 2 says memory “takes this to a whole new level as the AI synthesizes those chats into insights that might otherwise remain hidden.” The old model of privacy (who can see your records) misses this. The risk now lies in what the system can conclude, so leaks “far surpass what is possible just with access to chat transcripts alone”.

Reframing the question#

Several posts end by changing the question they started with:

Analogy by structure, and play#

Writing fiction in order to think#

The Letters preamble (2025-11-23) is the batch’s most explicit statement of method. He took the invitation “as an opportunity to explore, extend, and crystallize my own thinking”, and chose “to flex my inner-creative and use the medium of a short story” because “there are affordances in fiction that allow complex ideas to be explored with a nuance and sophistication that all too easily elude more literal pieces.” Two months earlier he had glossed affordances as “action possibilities” (2025-09-21). Now he applies the same idea to fiction: a story is a space that makes certain kinds of thinking possible. The story took his “evolving thinking” somewhere “unexpected”. Writing is how he finds out, not how he reports what he already knows.

Holding tensions and leaving them open#

2. What matters to him#

Trust and the fabric of relationships#

The value at stake in the email post is not efficiency or accuracy but “the relational connective tissue connecting its members”. He separates relational from transactional communication and notes that in a university “pretty much everything is relational (at least as far as faculty are concerned)” (2025-09-07). In the informant post it is confidentiality and intimate trust: a partner, a mother, a confessional “safe space” turned “informant” (2025-10-05). In the mental health post: “Effective health interventions require time, understanding, legitimacy, expertise, and trust” (2025-11-09).

Dignity, voice and people with less power#

The advisor post (2025-10-26) is about power. The student’s draft “no longer reflects your voice or ideas”. He extends the concern to any collaboration with “a clear power differential between the people involved”, and his closing question to advisors is whether their AI use is “robbing them of their dignity, denying them learning opportunities”. Students recur throughout the batch as the people he protects: undergraduates whose stories are “not mine to share”, PhD students, and students in “an emotional and mental health pressure-cooker environment”.

Informed choice, not prohibition#

His baseline safeguard is that people know what they are choosing. Memory use is “fine. As long, that is, they are made fully aware”, “so that they can make informed choices about how they use AI—and how they don’t” (2025-10-05). He notices that defaults decide for people who never chose: to turn memory off “you need to know that this is an option—and where to look”. Of AI email users: “and why should they if they haven’t been alerted to them” (2025-09-07). He does not blame the uninformed. He asks who ought to have informed them.

The good that AI does#

He will not let risk crowd out benefit. “To be clear, I am a strong proponent of students experimenting with and using generative AI. I see the benefits on a near-daily basis”. For someone who cannot face another person, “generative AI can be a life saver” (2025-11-09). AI email help “isn’t necessarily a bad thing” (2025-09-07). Parasocial AI might be beneficial “if they are done right” (2025-11-19).

Identity and the craft of thinking#

The Letters preamble is personal: “I’m grappling deeply with how AI challenges my identity as a professor and an academic”. What is threatened is the intellect itself, “the very things that AI sets out to excel at—form the deepest foundations of who you are” (2025-11-23). He writes about this from inside, not as an observer.

Opening up how knowledge is made#

He defends parasocial communication by academics because it can “draw back the curtain on the process of creating new knowledge, and help people feel they are connected to discussions and discoveries they would usually be excluded from” (2025-11-19). Inclusion here means letting people into the process, not just handing them the results.

Integration over silos#

His main criticism of the Stockholm report is that it does not take “even more inclusive and integrated approaches to AI, society, and the future”, and he offers his own where-we-live/what-we-do/who-we-are framework as “a deeper level of interconnectedness” (2025-11-12).

What frustrates him#

What delights him#

A children’s picture book that explains qualia better than the philosophy papers (“Which is pretty amazing”, 2025-09-21). A Halloween improv that turned “surprisingly insightful” (2025-10-31). Writing fiction at last: “something I’ve wanted to do for as long as I can remember” (2025-11-23). And, less comfortably, being surprised by his own experiments.

3. Risk as a way of thinking#

A new technology can break an old risk category#

The strongest evidence in this batch is not a stated theory but a repeated observation: familiar categories stop fitting.

His answer to the university problem is a relational process, not a control. Students should “learn to understand and navigate potential risks”, “not by simply providing guidance … but by talking with them and listening to them, creating safe environments for discussion, building trust”. He ends with everyone “motivated to work together to navigate them”. “Navigate” appears three times in the post. He does not drop management (“effective risk reduction and management strategies are surely essential”), but he admits that “it’s not clear yet what “doing more” might mean here”, and he treats shared navigation as the honest first step. The email framework is likewise titled “Assessing and Navigating”, and he looks for guidance “on how to navigate these use-cases” (2025-09-07).

Risk as a threat to what people value#

He never labels it, but every risk in the batch is defined by what is valued:

He uses “catastrophic” at the scale of a relationship, not a civilisation: “potentially serious risks here—and even catastrophic ones” (2025-09-07). This shows plainly what “threat to value” does. It lets something as small as an email count as a serious risk because of what it can destroy. The ChatGPT-drafted start-here bullets he published describe the same thing, calling risk “threats to what people value — identity, dignity, belonging, aspiration” (2025-11-24, provenance flagged above).

The risk landscape, and risks nobody owns#

Plausibility, scale, and acting before incidents#

Tools that open questions, and a tension worth noting#

The email model borrows the apparatus of conventional risk assessment: hazard lists, scenarios and 0-10 scores. He frames it as exploratory, and its job is to make an unrecognised risk visible and discussable. He knows it lacks precision (“A more sophisticated analysis would focus on specific organization types, structures and cultures”). There is one tension a careful reader should notice. The scores came from GPT-5 Pro, and he lets them shift his belief (“I now believe … should take this extremely seriously”). The numbers work as a provocation, not a measurement, but he leans on them more than his usual caution about false precision would suggest. His safeguard is openness: the whole method is set out for readers to judge.

Humility#

The batch is full of calibrated doubt:

4. Scholarship and public writing#

One text, two homes#

The Letters story is a chapter for Academic Cultures: Perspectives from the Future (eds. Michael M. Crow and William Dabars, Johns Hopkins University Press). With the editors’ and publisher’s agreement, he serialises it on Substack first, “given the current relevance of the ideas explored in it” (2025-11-23). Here the scholarship and the public writing are literally the same text. He also presents the fiction as continuous with his research: “my work and scholarship has been deeply engaged in using fiction to reveal and explore new insights and understanding for some years now.”

A theory of his own public scholarship#

The parasocial post (2025-11-19) restates the model he works to. Candid, personal communication “can enhance the effective and useful flow of information between academics and society writ large”. Blogs and podcasts are “an emerging and powerful way for experts to have relevance and impact at scale”. “It’s a concept that I practice in my own work, and still stand by.” Even when the word “parasocial” turns sour, he defends the practice and treats the new AI meaning as a fresh research question.

Experimenting in public, with the working shown#

The email model and the Tyler simulation are small studies done in the open. Methods, AI’s role and limitations are all declared, and the materials are attached for others to check (2025-09-07, 2025-10-05). The advisor post uses an anonymous form, checked with the IRB and described honestly as “not a research project” (2025-10-26). Readers are asked to correct him.

How he treats evidence#

Expertise, his and others’#

He invokes his own expertise sparingly and for a specific purpose: to be sceptical of warnings (2025-11-09). He learns openly from others outside the usual expert circle. An elementary-school computer-science specialist (Tara Menghini) gives him the key text on qualia (2025-09-21). An undergraduate’s story reveals the memory risk (2025-10-05).

Transdisciplinary by habit#

One autumn takes in philosophy of mind (Nagel, Jackson), embodied AI, children’s literature, organisational communication, risk assessment, privacy and border politics, clinical research letters, litigation, sustainability science, lexicography and speculative fiction. He moves between them without announcing it.

Accessibility as a principle#

Most people “would blanch at” Jackson and Nagel, “But strip away the stigma of actually learning something from an elementary school book” and the ideas open up. “Of course, to any elementary school student this all feels obvious. But that’s the point.” (2025-09-21) Qualia is “the fancy term for what it’s like to “feel” something”.

A body of thought built across posts#

He links posts to one another. The parasocial post links the mental-health post. The planetary post invokes his where-we-live/what-we-do/who-we-are framework. The Letters preamble points to The Artisanal Intellectual in the Age of AI. The Substack works as a cumulative, cross-referenced notebook.

5. His role as he sees it#

A public scholar who thinks with his readers#

He ends posts by opening them to others:

His stated aim for the email framework is practical and generous: to “help people avoid mis-steps in the future” (2025-09-07).

On the side of those with less power, and a critic of his own house#

He takes the side of students against advisors and users against defaults. He asks institutions, including his own, to own their duty of care. He names ASU among the universities whose ChatGPT EDU pages do not mention mental health (2025-11-09). He does not exempt himself. The Letters story starts from his own threatened identity, and “I suspect I’m not alone here” (2025-11-23).

Towards companies and other experts: fair, not adversarial#

What he refuses to do#

Changes of mind, handled in the open#

6. What is distinctive#

  1. Serious risk at the scale of a relationship. Most AI risk talk is about catastrophe, jobs, bias or misinformation. Maynard gives the mundane and intimate the full seriousness of a risk professional: an email, a memory setting, an advisor’s edit, a chatbot friend. He locates “even catastrophic” harm in trust, confidentiality, dignity and voice. The unit of analysis is the relationship, and the harm is a threat to relational value.

  2. Noticing when the category has broken. He treats “a privacy risk that’s been around for a while” as changed in kind once the system can infer. He treats university AI risks as “unlike anything universities have had to navigate”. He then shows, lever by lever, why conventional controls misfire or create counter-risks. The new mindset is argued from the mechanism, not asserted.

  3. Fiction and play as research apparatus. He builds a fictional person so he can study a risk ethically. He uses a picture book as philosophy of mind, an improv game as futures method, and a short story as a way to “crystallize” his own thinking. For him fiction is a set of “affordances” for thought, not decoration.

  4. Reflexive checks on his own bias. He uses AI deliberately to counter his own framing in a risk assessment. He publishes the materials and invites correction. Few commentators treat their own judgement as a variable to control.

  5. Turning the gaze on the powerful and on institutions, including his own. He looks at advisors, not students, and at universities as providers with a duty of care, not students as rule-breakers. ASU is named. He does not blame users “if they haven’t been alerted”.

  6. Classroom as early-signal detector. Emerging risks reach him through students’ stories and the grapevine well before the literature. He treats those signals as worth acting on under plausibility, while asking readers to help bound them.

  7. Neither booster nor critic. “A strong proponent” of student AI use who argues for a duty of care. A defender of parasocial connection who takes its AI dark side seriously. An admirer of AI’s planetary promise who worries about dependency on a crash-prone market. The balance is not a compromise. It is his way of keeping all the value at stake in view.

  8. Public scholarship defended as a practice. He has an explicit theory of why scholars should be personally present to publics, “draw back the curtain” on knowledge-making. He keeps it when the word turns sour, and sets the same scholarly text out as both a Johns Hopkins chapter and a Substack serial.

7. The posts that best reveal how he thinks#

  1. 2025-10-05 when-chatgpt-turns-snitch. A classroom story as trigger; intimate “imagine” scenarios; an old risk changed by inference; an ethical, self-limiting simulation with materials published; curiosity and worry together; plausibility without incidents; informed choice over prohibition; open invitation to correct him.
  2. 2025-11-09 universities-chatgpt-mental-health. Pro-use and duty of care held together. Decades of risk-communication expertise used to reject performative fixes, then each lever tested for counter-risks. Honest uncertainty (“not clear yet”), a relational answer (“work together to navigate them”), and self-critique of his own institution.
  3. 2025-09-07 the-hidden-risks-of-using-ai-for-email. The “risk hat” put on out of curiosity; a simple model as a way of seeing; AI used as a check on his own bias; surprise that changes his mind; risk defined as damage to “relational connective tissue”.
  4. 2025-11-23 letters-from-the-department-of-intellectual-craft-prelude. Fiction as a way of thinking (“affordances in fiction”); his own identity at stake; courage to step outside his genre; scholarship and public writing as one text; humour about academic politics.
  5. 2025-11-19 parasocial-relationships-problematic. His theory of public scholarship in his own words, and how he changes his mind: concede, keep what holds, extend to the question he missed.
  6. 2025-09-21 a-kindergarteners-guide-to-ai-qualia. Serendipity via a podcast guest, play and snark as method, structural analogy (download as pre-training), and accessibility as a principle.

Also revealing: 2025-10-26 (the gaze turned on the powerful; evidence before judgement) and 2025-11-12 (the question reframed from AI to “conversations that lead to action”).