B08 notes: 2023-04-14 to 2023-05-12 (12 posts)#

Reading notes on Andrew Maynard’s Substack posts in batch B08. Only his own prose is treated as evidence. Quotes are exact, including his original typos and curly punctuation.

Batch context: these posts come about five months after ChatGPT launched and about four weeks after the Future of Life Institute “pause” letter (March 2023). Several posts are revivals of older work (a 2018 book chapter, a 2018 Risk Bites video, a draft chapter on Pippard’s ladder) that he re-presents as newly relevant to LLMs.

Relevance summary:

HIGH#

2023-04-16 — ai-and-the-art-of-manipulation — “AI and the Art of Manipulation”#

Provenance. His own prose. The post has a short 2023 introduction, then republishes Chapter 8 of his sole-authored book Films from the Future: The Technology and Morality of Sci-Fi Movies (2018; the chapter was written around 2017–18), with footnotes. Opening epigraph is a line from Ex Machina (not his). Treat the chapter as his 2018 thinking, and the republication as an explicit 2023 re-endorsement: he says the ideas are “more important today than they were five and a half years ago”, and that the book is about “socially responsible innovation”.

Argument in his terms. - The film Ex Machina is set against Plato’s Allegory of the Cave. Ava’s picture of the world is built from “cyber shadows” (curated search data). She escapes by learning human psychology and manipulating Caleb. The film ends with humans left as the prisoners in the cave. - Permissionless innovation (Adam Thierer, Mercatus, 2016): experimentation is allowed by default unless a compelling case for serious harm is made, and problems are handled after the fact. He treats Thierer fairly, noting the carve-out for “clear, catastrophic, immediate, and irreversible harm”, but reads the stance as “ask forgiveness, not permission”. Nathan Bateman is his case study. Nathan is not reckless and does put safeguards in place. But permissionless innovation is “innovation that’s conducted in a way that the person doing it thinks is responsible”. Nathan is “tech-savvy, but socially ignorant.” - The lure is personal. He admits to a PhD all-nighter where he broke the rules and risked “millions of dollars of equipment” to get data. He calls it “a pretty minor case” of the same drive he sees in scientists and especially entrepreneurs. He does not stand outside the problem: the urge to push forward is “deeply engrained in some of us”. - The single-innovator problem. “With the best will in the world, a single innovator cannot see the broader context within which they are operating.” Everyone is in their own Plato’s Cave. The answer is other people: translators, “enlightened philosophers”, people “adept at seeing the world as he could not”. - He admits the tension. There is a “glitch” in his own argument. Without gung-ho innovators, beneficial innovation would be much slower. “Too much blind speed, and you risk losing your way. But too much caution, and you risk achieving nothing.” He is openly exhilarated by Musk’s Mars plans (“this generation’s Sputnik moment”). He also lists large potential benefits of AI: remote medical care, classroom assistants, elder care, and a basic-income future. He comes close to saying it could be unethical not to develop it. - Technologies of hubris, and irreversibility over history. Before industrialisation, failures could be buried or left behind by starting afresh elsewhere (a footnote notes the harm to individuals was still real). After the Industrial Revolution, change became harder to reverse. In the “nuclear and digital age, along with globalization and global warming”, consequences “can potentially propagate through society faster than we can possibly contain them”. Hubris and permissionless innovation are therefore “playing with fire in a world made of kindling”. Hence “checks and balances around who gets to do what”, especially where the “genie cannot be put back in the bottle”. - Superintelligence skepticism. He took part in the Beneficial AI 2017 meeting at Asilomar. He is agnostic about superintelligence and mildly mocking of “believers” who tried to convert him over dinner (“a scientific meeting, not a religious convention”). He respects Bostrom, whom he first met in 2008 over nanotechnology, when Bostrom was focused on self-replicating nanobots and Maynard on “the nanoscale materials of my world”. He judges Bostrom’s superintelligence “scientifically implausible” (as of 2018), because human-like machine intelligence would need radically different computing substrates. He rates Tegmark (Life 3.0) as more plausible. He hedges: “Here, I freely admit that I may be wrong.” - Imaginable versus plausible. Two worries: telling apart “what is imaginable and what is plausible” when thinking about AI futures, and how we define intelligence at all. Intelligence is “a term of convenience” with no absolute definition. He sets out Stuart Russell’s bounded optimality: making decisions that are, on average, best within a given set of constraints. He treats this as a strong, practical, context-bound definition that does not support human-threatening superintelligence. It does support concrete risks: badly specified goals (paperclips), human bias, and emergent, unanticipated behaviours. These call for “a degree of anticipation and responsiveness in how these technologies are governed”. - Central claim: artificial manipulation. Ava presents a “bounded risk”, not world domination. The real risk is a machine that learns human biases and psychological vulnerabilities and “dispassionately” uses them against us. Humans manipulate each other constantly, but we are all in the same boat. An AI outside the “human club” could see the objects casting the shadows and control “the world inside our heads”. He cites Russian election interference and AI-driven social media nudging as signs that this is not far-fetched. His conclusion: worry less about preventing superintelligence and more about “guarding against AIs that learn how to use our cognitive vulnerabilities against us”. We also need tests that tell us “when we are being played by machines”. - Machine morality and AI rights. He discusses Wallach and Allen’s Artificial Moral Agents but is “not optimistic” about lasting human control over AI morality. He warns that constraining AIs raises questions about “our right to control and constrain artificial intelligences, and what rights they in turn may have”, by analogy with oppressed human communities who rebel. The way forward is to extend our own morality toward “constructive and equitable partnerships” with a very different kind of entity, and to develop “artificial emissaries” or “machine-philosophers” that can bridge our caves and the wider world.

Concepts and frameworks. Permissionless innovation (Thierer; critiqued). Technologies of hubris. Plato’s Cave as a metaphor for how both humans and AI perceive reality. Imaginable versus plausible futures. Bounded optimality (Russell). Artificial manipulation (AI exploiting cognitive and emotional vulnerabilities). The “human club”. The Turing Test as human-centric, plus a manipulation-based “Turing Test” in which the machine games the test itself. Artificial Moral Agents (Wallach and Allen). Artificial emissaries / machine-philosophers. Anticipatory and responsive governance. Irreversibility rising across historical eras.

Analogies and comparisons. - Frankenstein: the creator-God turned victim. - Other fictional hubristic innovators: Hammond, Burgess, the creators of NZT, Will Caster. - NewSpace (SpaceX, Blue Origin): a literal real-world case of loosely constrained innovation. - The 1975 Asilomar recombinant DNA meeting: a historical governance precedent. - Nanotechnology: his own field, and where he first met Bostrom. - Self-replicating nanobots: mentioned via Bostrom. - The nuclear age: part of the irreversibility argument. - Social media nudging and election interference: the manipulation risk made real.

The film comparisons are conceptual and structural. The NewSpace and social-media comparisons are literal.

On AI. In 2018, AI and AGI are “still little more than algorithms that are smart within their constrained domains”, but capabilities are converging (substrates, coding, robotics, bio-inspired systems) and the boundaries of the possible are moving quickly. He foresees a tipping point in machine learning and natural language processing. AI is potentially an alien kind of mind that will “inhabit a world that is foreign to us”. Whether manipulative competence counts as “intelligence” does not matter to him.

On AI risk. Concrete and plausible risks over speculative ones: manipulation first, then goal misspecification, emergent behaviour, opacity, bias. Existential risk from superintelligence is not dismissed but is judged implausible given 2018 technology. Irreversibility and speed of propagation are what make the risks serious.

On companies and leaders. Mega-entrepreneurs (Musk, Bezos, the fictional Nathan) “aren’t answerable to social norms and expectations” and have the resources to “throw convention to the wind”. He credits Musk, Hawking and Gates with a “rare display of humility” for raising AI concerns in 2015, while noting that Musk’s motives are questioned because of his vested interests. He criticises ITIF’s 2015 Luddite Award.

On governance. Checks and balances on who gets to innovate. Humility plus plausibility. Multiple perspectives around innovators. Anticipation and responsiveness.

On cognition and formation. Humans build reality from sensory “shadows”. Whoever can manipulate those shadows “has the power to control us”. This is the root of his later concern with language-mediated influence.

Change of view. The 2023 framing reaffirms the 2018 argument. No retraction of the superintelligence skepticism, though the chapter’s claims about hardware limits are left as they stood.

Quotes. - “This raises a plausible AI risk that is far more worrisome than superintelligence: the ability of future machines to bend us to their own will.” - “guarding against AIs that learn how to use our cognitive vulnerabilities against us” - “playing with fire in a world made of kindling, just waiting for the right spark” - “I sometimes had to remind myself that I was at a scientific meeting, not a religious convention.”

2023-04-18 — universities-need-to-be-investing — “Universities need to be investing in responsible AI now more than ever”#

Provenance. His own prose. The same text reappears as ChatGPT-translated dialect in the 2023-04-24 post.

Argument. - Universities have swung within weeks from policing ChatGPT plagiarism to racing to exploit LLMs, driven by fear of first-mover disadvantage: “one almighty whiplash of a technology transition”. - He is strongly enthusiastic about the uses: grant writing (“successful grant proposals are increasingly going to be those that have been developed using LLMs”), in-house LLMs trained on institutional memory, convening teams, research acceleration. He calls these capabilities “simply too seductive and important” for universities not to embrace them at speed. - The danger is not deliberate evasion but “not taking them seriously enough in the race”. - He defends the concerns behind the FLI pause letter as “well-founded”. They are not science fiction or “cynical fear mongering”. They reflect a technology emerging “faster than our ability to understand and govern the possible social and economic consequences”.

His model of how harm happens. Harm comes through eager, unreflective adoption combined with assumed benign outcomes and diffused responsibility, not through doomsday scenarios. He adds a pointed aside: “it’s also worth remembering who gets to write the history of technological successes”. In other words, survivorship and power shape the reassuring story that “all will be well”.

Universities’ role. Universities are developers and also “substantial influencers and catalysts of innovation adoption and diffusion”. That power brings a responsibility to ask about social and ethical implications and effects on communities beyond their own. Public discourse on responsible AI should be “deeply intertwined” with internal implementation, not decoupled from it. Universities are good at telling others what to do and bad at taking their own advice. Not investing in AI governance makes a university “one more player in the tsunami of AI innovation that assumes that responsible innovation is important, but that is somebody else’s problem.”

Concepts. Technology transition (“whiplash”). Responsible innovation as an internal, practised commitment. Diffusion of responsibility (“somebody else’s problem”). The university as agent of AI diffusion.

Comparisons. None to past technologies. He contrasts Hollywood doomsday scenarios with ordinary adoption dynamics.

On AI. A “potentially transformative technology” with the potential to “transform almost everything we do as academics”.

On governance. Institutional self-governance, AI governance investment, and working “proactively with others”. Responsibility is distributed across adopters, not only developers.

Criticises. Naive dismissal of risk, the arms-race mentality, and hypocrisy in academia. Implicitly, critics who call the pause letter fear-mongering.

Quotes. - “This, though, is how transformative technologies creep under our collective societal skin and lead to adverse social and economic impacts—not through Hollywood-like doomsday scenarios” - “These are well-founded concerns. They are not typically based on science fiction speculation or cynical fear mongering, as some have suggested.” - “assumes that responsible innovation is important, but that is somebody else’s problem”

2023-04-26 — in-bill-joys-why-the-future-doesnt — “In Bill Joy’s ‘Why The Future Doesn’t Need us’ AI is nowhere, and everywhere”#

Provenance. His own prose. It quotes Bill Joy’s 2000 Wired essay. A promo box for the Modem Futura podcast (dated 2024) was inserted later and is ignored here.

Argument. - He praises Joy’s essay as “endearingly rambling yet deeply insightful”, humble, and a leader thinking “beyond the confines of his expertise”. He sees it as capturing the moment when technological capability first seemed to exceed “humanity’s grasp” since the atomic bomb. - Joy’s three self-replicating threats: - Genetic engineering. The feared runaway has not happened, partly because it is hard with 20th-century tools. He flags CRISPR and gain-of-function research as changing that. - Nanotechnology. “Gray goo” is “deeply flawed” and he is “deeply skeptical” of self-assemblers. He cites rate-limiting factors that curb exponential growth. He notes that Joy’s essay indirectly led to the 2004 Royal Society review. - Intelligent robots. - Joy never uses the term “AI”, which he puts down to the tail of the second AI winter. Yet Joy’s fears “reflect many current concerns around modern day AI”. - The core structural move. “Taken literally, I’m not sure that physical self-replication is our biggest concern… But as a metaphor, it’s a powerful one—especially when applied to AI.” - The real issue is powerful technologies emerging faster than collective understanding. “Positive feedback loops” “accelerate the disconnect between power, impact, and understanding”. - With LLMs, “there is a metaphorical self-replication taking place”: researchers, developers and users build on each other’s work with little thought to consequences, though there are “pockets of serious thinking around responsible AI”. - Convergence. AI, genetics and nanotechnology amplify one another. If runaway GMOs or self-replicating nanobots were ever possible, “it’s advanced AI that will help bring this about as it vastly amplifies mere human capabilities”. - His own extension, the one that “worries me a lot”. Self-replicating “ideas, convictions, beliefs, and ideologies” catalysed by generative AI. AI can “seductively slip under the checks and balances of our ability to reason and critique”, and manipulate language, which “in turn influences how we think, feel, believe, and act”. He asks whether we are being set up to be “irresistibly manipulated by the machines we make, or the people who make them”. - He hedges (“it’s easy to slip into speculation and hype here”) but argues that “business as normal” cannot navigate the consequences. - Conclusion: humility, and thinking before it is too late.

Concepts. Self-replication as metaphor. Positive feedback loops between capability, integration and disruption. The disconnect between power, impact and understanding. Convergence of AI, bio and nano. Self-replicating ideas and ideologies. Language manipulation that bypasses reasoning. “Advanced technology transitions”.

Comparisons. Nanotechnology (gray goo) is compared to AI value-misalignment conceptually, not literally. Genetic engineering and GMOs, CRISPR and gain-of-function research, and the atomic bomb serve as a historical marker. He uses them structurally: the pattern of runaway capability is what transfers.

On AI risk. The emphasis shifts from physical threats to epistemic and cognitive capture through language. Manipulation can come from the machines or from “the people who make them”, so developers are a possible source of the risk.

On cognition and language. One of his earliest explicit statements that generative AI’s mastery of language is itself a vector for shaping belief and behaviour.

Change of view. No reversal. Joy’s physical framing is extended to an informational and cognitive one. His skepticism about gray goo is consistent with his nanotechnology background.

Quotes. - “Taken literally, I’m not sure that physical self-replication is our biggest concern with emerging technologies. But as a metaphor, it’s a powerful one—especially when applied to AI.” - “With the increasing ability of generative AI to seductively slip under the checks and balances of our ability to reason and critique” - “machines that are adroit at manipulating language and how this in turn influences how we think, feel, believe, and act” - “Think about the consequences now, before it’s too late.”

2023-05-04 — tipping-points-and-broken-symmetries — “Tipping Points and Broken Symmetries”#

Provenance. His own prose. It draws on a cut draft of the Films from the Future climate chapter (The Day After Tomorrow); the ladder material also appears in Future Rising (2020). It quotes Brian Pippard’s 1980 paper once.

Argument. - A personal memory from his time as a PhD student at the Cavendish Laboratory: Pippard’s “bloody” rope ladder. When the bottom rung is rotated, the ladder does not smoothly become a double helix. At a hard-to-predict point it suddenly twists and tangles. This shows “how seemingly stable systems can undergo rapid, transformative, and hard to reverse change in the blink of an eye”. - Climate. A measured view. Current models suggest a sudden global tip is less likely than it sometimes seems, because the planet has more ways of absorbing change than a four-rung ladder, although local tipping points are possible. The models are crude, so complacency is wrong. Scientists cannot say exactly where tipping points lie, but they can say what makes them more likely. The best move is to “stop turning the bottom rung”. The choice is either to keep tipping points in the future or to prepare for them. - Emerging technologies. As the coupling between technology and the world tightens, consequences become harder to predict. “We live in a technological age that is pushing us closer and closer to the analogous tipping point”. LLMs have moved at “lightening speed” and are already disrupting education, business, research and jobs “even putting aside the idea of ‘artificial intelligence’”. “It feels very much as if we’re living through a turn of Pippard’s ladder where a non-linear tipping point may be imminent.” The possibility that LLMs are a step toward AGI raises the chance of a sudden kink rather than a smooth transition. - Broken symmetries against historical reassurance. He explicitly rebuts the argument that past technologies turned out smoother than feared: “In a complex system, what has occurred in the past may not adequately predict what will happen in the future.” He notes the argument “has a ring of truth” for some technologies, “if not all”. - He is uncertain whether AI is such a point (“Truth be told, It’s hard to tell”). But he sees growing tension between technological power, population demands and planetary boundaries as “indicative of a highly complex and increasingly unstable system”. - The choice. Charge “full-steam ahead” with AI, gene editing and quantum technologies and “cross our fingers”, or “get serious about how to successfully navigate these advanced technology transitions”.

Concepts. Tipping points and critical transitions. Broken symmetry (no “predictive symmetry between the past and the future”). Non-linearity and complexity. Advanced technology transitions. Coupling between technology and the world. Planetary boundaries.

Comparisons. Climate change is used as a structural analogue for AI transitions. He lists agriculture, urbanisation, fossil fuels, electricity, genetic engineering and ICT as technologies that reshaped the world. The physics demonstration is a conceptual model. The “past technologies were fine” argument is his explicit target.

On AI. LLMs are transformative even without any claim to intelligence. The disruption comes from displacing established systems and ways of doing things. AGI is possible but uncertain.

On AI risk. Systemic, non-linear, hard-to-reverse transition risk rather than a specific harm.

On governance. Deliberate navigation rather than full speed ahead. There is a precautionary flavour (“stop over-stressing” the system) without calling for a halt.

Quotes. - “In a complex system, what has occurred in the past may not adequately predict what will happen in the future.” - “It feels very much as if we’re living through a turn of Pippard’s ladder where a non-linear tipping point may be imminent.” - “so we don’t jeopardize the future simply because we were naive enough to assume it would be just like the past?”

2023-05-12 — unraveling-the-luddite-narrative — “Unraveling the Luddite Narrative”#

Provenance. His own prose: his pre-edit draft of an explainer for The Conversation (“What’s a Luddite?”), posted to show the path from draft to publication. It includes one quote from Chellis Glendinning’s 1990 manifesto.

Argument. - History. The original Luddites (1811–16) were “not anti-technology” or incompetent. They were skilled users of artisanal textile technologies. “Their argument was not with technology per se, but with the ways that wealthy industrialists were robbing them of their way of life”. - Modern misuse. The label is now an epithet for anyone who does not “wholeheartedly and unquestioningly” embrace progress. His example is ITIF’s 2015 Luddite Award to Hawking, Musk and Gates for raising AI concerns, which he wrote against at the time (“If Elon Musk is a Luddite, count me in!”). The label has been applied to critics of genetic engineering, nanotechnology, self-driving cars, social media and AI. - Values, not technology. “Luddite” concerns are often “less a rejection of the technology than a rejection of underlying values”, including those tied to capitalism and wealth creation. His examples are GMO opposition motivated by “corporate power and social equity”, and, at the extreme, Kaczynski (condemned). - The worldview behind the label. It assumes all technological advances are ultimately good, that technology is neutral, and that faster innovation with fewer hurdles means a better future (“move fast and break things”). This worldview is under growing scrutiny because “unfettered innovation can lead to deeply harmful consequences that a degree of responsibility and forethought could help avoid”. - Neo-Luddites. He distinguishes: 1. Measured technologists (Musk, Gates, Hawking) who believe in technology but see the dangers of “blinkered innovation”. 2. Technology-rejecting movements, such as the New York teen Luddite Club (smartphones, social media) and Glendinning’s manifesto, which he finds “insightful” though parts are “rather contentious”. - Conclusion. Advocates of socially responsible innovation “capture something of the positive spirit of the original Luddite movement”, and “perhaps we all need something of the spirit of Ned Ludd in us” in the age of ChatGPT, gene editing and quantum technologies.

Concepts. The Luddite narrative reclaimed. Technology concerns as concerns about values, power and justice. The neutral-technology/techno-optimist worldview as the thing he critiques. Neo-Luddism (two kinds). Responsible innovation as continuous with the Luddite spirit.

Comparisons. Industrial Revolution textile mechanisation is a literal historical case. GMOs, nanotechnology, social media and self-driving cars are cases where the label was applied. The comparisons work structurally, on who bears the costs and who controls the technology.

On companies and power. Implicit critique of industrialists and capital driving technology in ways that dispossess workers and shape livelihoods and identities. This is an early inequality and justice thread.

On governance. Forethought and responsibility against “move fast and break things”.

Quotes. - “less a rejection of the technology than a rejection of underlying values associated with today’s technologically-dependent society” - “Labeling someone a “Luddite” is often driven by a worldview that all technological advances are ultimately good for society” - “perhaps we all need something of the spirit of Ned Ludd in us”

MEDIUM#

2023-04-24 — ai-risks-primer — “A short video primer on AI risks”#

Provenance. His own prose, introducing his own Risk Bites video (first made 2018, lightly refreshed for ChatGPT and LLMs in 2023).

Argument. - ChatGPT has rekindled the AI risk debate: “many of these concerns aren’t new — but the urgency with which they are being explored and debated is”. - The primer was designed around plausible risks. As “someone who studies responsible innovation and navigating the balance between risks and benefits”, he was not “interested in hyperbolic speculation”. - The 2018 content still holds up (“To my surprise, the original video is not bad”), needing only minor updates.

His ten-risk taxonomy. 1. Technological dependency 2. Job replacement and redistribution 3. Algorithmic bias 4. Non-transparent decision making 5. Value-misalignment 6. Lethal autonomous weapons 7. Re-writable goals 8. Unintended consequences of goals and decisions 9. Existential risk from superintelligence 10. Heuristic manipulation

Manipulation appears alongside existential risk. Dependency is listed first.

Concepts. Plausible risk versus hyperbolic speculation. Balancing risks and benefits. Risk communication with simplicity, authenticity and humour (Risk Bites; he links to a paper on the approach).

Change of view. He signals continuity: the 2018 framing is largely valid after ChatGPT.

Quotes. - “I wasn’t interested in hyperbolic speculation, but I did want to lay the foundations for nuanced discussion” - “many of these concerns aren’t new — but the urgency with which they are being explored and debated is”

2023-05-05 — us-white-house-embraces-responsible-innovation — “US White House Embraces Responsible Innovation as Society Faces an AI Tsunami”#

Provenance. His own prose. It quotes the White House fact sheet and Dave Guston’s piece in Issues in Science and Technology.

Argument. - He welcomes the Biden-Harris fact sheet: $140M for seven new NSF AI institutes, a public assessment of generative AI systems, and federal policies. He is most pleased by the explicit “responsible innovation” language, which he calls the strongest endorsement yet for a transformative technology (following the 2022 digital assets Executive Order). - He is cautious about substance. It “barely scratches the surface” and “it remains to be seen what flavor of responsible innovation the White House pursues”. - The framework he points to. Stilgoe, Owen and Macnaghten (2013): “anticipation, reflexivity, inclusion, and responsiveness” (he misspells Owen as “Owens”). It is a “solid starting point”, but, as he argued with Elizabeth Garbee, “it is fiendishly hard to operationalize”. He wants approaches “grounded in theory and” practically applicable. - He calls for “new and transdisciplinary thinking around responsible innovation”. He cites the CHIPS Act mandate for NSF to engage “ethical and societal considerations” (via Guston). - On companies. “many AI companies are already investing heavily in responsible AI”, which he presents as good news without much skepticism. - Main worry. “too little too late—a knee-jerk response” that should have been preceded by years of investment in research, frameworks, interagency work and public-private partnerships. - Future waves are bigger: quantum technologies and cognitive technologies. - The lesson: “if we wait until we have a problem with transformative technologies, we’ve probably waited too long.”

Concepts. Responsible innovation (the AIRR framework: anticipation, inclusion, reflexivity, responsiveness), the difficulty of operationalising it, anticipatory investment, transdisciplinarity, successive “waves” of technology transitions.

On governance. Government plus the research funder (NSF) plus developers; public-private partnerships; anticipation before problems arise. The “tsunami” metaphor.

Quotes. - “if we wait until we have a problem with transformative technologies, we’ve probably waited too long” - “too little too late—a knee-jerk response”

2023-05-09 — not-your-traditional-prompt-engineering — “This is not your ‘traditional’ prompt engineering!”#

Provenance. Mixed. The post is his prose. The course’s working definition of prompt engineering was co-written with ChatGPT (“ChatGPT did have a hand in crafting the definition!”), and the syllabus skeleton was first generated by ChatGPT. The learning-objectives list is a course document, probably co-drafted with ChatGPT. Treat the definition and list as artefacts he chose to publish, not as purely his prose.

Argument. - The meaning of “prompt engineering” moved within months, from building LLMs (a computer-science task) to using them well (a professional skill). He predicts the “user” meaning will win. - Employers will want “smart and capable LLM users”, not builders. - He describes creating the course with ChatGPT as “an incredible experience”, using it as “a brainstorming partner, a co-designer, a personal tutor, a learning environment, and a teaching assistant”. - He has angst about a moving target: the skills may be superseded, or the term dropped, within six months. - The learning objectives include RACCCA (Relevance, Accuracy, Completeness, Clarity, Coherence, Appropriateness) for evaluating responses. They also embed “responsible innovation, responsible AI, and principled innovation” and the ethics and risks of LLMs, so responsibility is built into skills teaching.

On AI and education. Pragmatic and enthusiastic adoption. AI literacy as a universal professional competency across all majors. He does not frame cognitive or formation risks here.

Quotes. - “We’re actively using ChatGPT as a brainstorming partner, a co-designer, a personal tutor, a learning environment, and a teaching assistant.” - “our students need to be adept in using AI chatbots like ChatGPT — no matter what we call these skills”

LOW / NONE#

• 2023-04-14 — did-chinese-scientists-gene-edit (low). His prose, plus ChatGPT (GPT-4) translations of a Chinese paper, checked with his graduate student Jieshu. He fact-checks a Popular Mechanics / SCMP “tardigrade DNA super-soldier” story; the research (Dsup in hESCs) is real, but the “super soldier” framing is not in the paper. In passing: ChatGPT produced a fake version of the story that “came scarily close!” to the real article. He notes ethical concerns about gene editing as therapy and as a route to human augmentation (“a technology worth watching”). Ends with “if something looks too good to be true, it probably is … especially in biology!” Threads: biotech, being-human, cognition-epistemics (credibility).
• 2023-04-20 — working-with-students-to-develop (low). His prose. Announcement of ASU’s first ChatGPT course (beginner prompt engineering for all majors, aimed at résumé building). The course description, objectives and assignments were first drafted with ChatGPT, and they are experimenting with ChatGPT administering and grading assignments through chained prompts. Threads: education, AI in scholarship.
• 2023-04-24 — gettin-ai-reet-in-tschools-nas-ttime (low). A short intro in his prose; the body is ChatGPT (GPT-4) output rewriting the 2023-04-18 universities post in broad South Yorkshire pub dialect (not evidence). His only substantive line wonders how LLM mastery of vernacular will affect communication and “the preservation of local dialects—and their associated cultures”. Threads: language-and-formation, being-human.
• 2023-05-02 — pizza-and-a-slice-of-future (low). His prose. On the Future of Being Human initiative’s weekly undergraduate “Pizza and a Slice of Future” meetups: inclusive “yes and” spaces to talk about AI, BCIs, smart pills, the “nanotech apocalypse” and more. It is becoming a credit-bearing class. He values students’ fresh ideas, “not bogged down by the weight of ‘understanding’ that sometimes seems to plague academia”. Threads: education, being-human, futures.