Late Lessons, Jensen Huang and AI
Markdown version

B30 notes: 2026-05-15 to 2026-07-10 (8 posts)#

These are reading notes on Andrew Maynard’s Substack posts in batch B30. Only his own prose counts as evidence. Quotes are exact, including his typos and curly punctuation.

Batch context. Mid-2026. On 17 May he announces a sabbatical, which runs to August 2027. Pope Leo XIV publishes the encyclical Magnifica Humanitas on AI and being human (the post calls publication day “Monday”, 25 May). Anthropic releases “Mythos-class” Claude Fable 5 (the post is dated 10 June and says “Yesterday”). A US government export-control directive suspends access on 12 June, and access returns on 1 July. Four of the eight posts are hands-on experiments with Fable 5 as a research and design engine. In each he publishes AI-made outputs (papers, a game, self-audits) with his own framing around them. Across the batch, only about 9,000 of the ~16,600 words are his own analytical prose. There are no Modem Futura podcast posts in this batch, so nothing was skipped under the user’s instruction.

Relevance summary

Date Slug Relevance
2026-05-15 ai-movies-may-be-less-dystopian-than-we-think medium
2026-05-17 the-nonsense-i-write low
2026-05-21 magnifica-humanitas-and-being-human high
2026-06-10 is-anthropics-new-ai-model-poised high
2026-06-12 a-quick-update-on-using-claude-fable-5 high
2026-06-14 everything-you-wanted-to-know-about low (the embedded note is mostly AI-written, per the ruling)
2026-07-04 just-how-good-is-anthropics-fable-as-a-research-assistant high
2026-07-10 i-asked-anthropics-fable-5-to-create-a-video-game-inspired-by-my-work medium

HIGH#

2026-05-21 · magnifica-humanitas-and-being-human · “Magnifica Humanitas and Being Human in an Age of AI”#

Provenance. This is his own prose throughout: the main post written before publication, the 25 May update line, and the postscript written after a first reading of the encyclical. He block-quotes his own January 2025 post “universities-need-to-step-up-their-agi-game” (his own prose, re-endorsed here) and reproduces its schema image. The quoted phrases from Pope Francis’s Laudato Si’ (“for a new dialogue…”, “conversation which includes everyone…”) are not his. The same goes for the encyclical’s own terms, which he lists without analysing them: Tower of Babel versus Nehemiah, “disarming” AI, “shared discernment”. The header image of Pope Leo XIII recording a phonograph greeting (1893) is archival.

Argument in his terms. - Three encyclicals, three domains. He reads three papal documents as mapping onto his own schema for technology transitions. Rerum Novarum (1891, Leo XIII) covers what we do: work, capital, labour and dignity under industrial technology. Laudato Si’ (2015, Francis) covers where we live: the coupling of technology use with the environment and ecological systems. Magnifica Humanitas is expected to fill the gap on who we are. He says the Vatican framed the new encyclical against Rerum Novarum and “largely” overlooked Laudato Si’. Yet Laudato Si’ is “a vital piece of the puzzle”, especially “at a time when AI both presents an environmental threat and a potential pathway to developing novel solutions”. - “who we are” is where AI is unprecedented. The earlier encyclicals “fall short of addressing the one domain where AI is shaking things up in ways that no other technology has come close to.” In this domain the technology “both challenges and opens up new ways of revealing who we are.” - Evidence of the gap: techlash and cognitive coupling. Students booing pro-AI commencement speakers (NPR) reflect “a growing wave of antagonism toward the technology”. This is driven partly by threats to what we do (jobs, value creation) and where we live (water, energy, land use). But it also hints at concerns about “cognitive coupling” between AI and its users. He sets out a spectrum: - “AI psychosis” at one end; - at the other, “less obvious — yet equally important” impacts from “conversational AI’s ability to bypass our cognitive defense mechanisms.” - His own research programme on cognitive risk, placed beside others’ work: - Shaw and Nave’s (Wharton) preprint on “cognitive surrender”: heavy AI users trust AI to reason for them “despite it not being trustworthy”. He says this “aligns with” his work. - His “cognitive trojan horse” (arXiv 2601.07085): AI can “bypass our cognitive defense mechanisms by broadcasting signals we usually associate with human trustworthiness.” - His “constitutive resonance” (SSRN 6343880): “a two-way coupling where both human and artificial participants are changed in the process”. It could “accelerate how the technology impacts how we think, perceive ourselves and others, behave, and make decisions.” - He expects these to be “just the tip of a growing area of research around how AI potentially threatens who we are.” - The other side: AI as a way to rediscover who we are. The “unique relationship between humans and artificial intelligence has the potential to transform our understanding of who we are” and so to help us thrive. He points to AI and the Art of Being Human (with Jeff Abbott) and to Bryan Penprase’s Forbes piece about his advice. - The central normative question. “In a world where AI is inevitable (and I would argue that the boat has already left the harbor here)”, how do we develop and use AI in ways that “center human dignity”, enable flourishing, and take “an integrated approach to what we do, where we live, and who we are?” - Postscript: read it yourself, without AI. He warns against pasting the encyclical into an LLM and poses two questions: “As you read an AI-generated summary, what is missing in it, and how will you know?” How much of its “lived humanity” will LLMs miss? He claims firmly that “there are layers here that LLMs will overlook, or simply not be able to represent, because they are not intimately embedded in the full experiential spectrum of what it is to be human.” - Postscript: personal resonance. “I genuinely felt seen reading the encyclical.” He describes his own career-long advocacy as “broad and inclusive approaches to technology innovation that elevate the marginalized and center on human dignity; that are grounded in listening, humility, and a willingness to change”. - Postscript: a blueprint for action and for institutions. He calls the encyclical “a powerful blueprint” for actively navigating technology and society. It speaks to developers and governments, and also to anyone and any organisation with a role. He stresses universities: they must build learning environments “with great care and humility”. This is easy to overlook “in the rush to go as fast as possible and prioritize the transactional over the relational.” - Postscript: where it “stops short”. The encyclical sets out to preserve what it means to be human “as we have understood this for millennia”. But “the cutting edge of AI development is beginning to force a reckoning with long-held assumptions around what it is to be human”. He cites Anthropic co-founder Chris Olah’s remarks at the release about “creating something we don’t fully understand”, which the Pope “strenuously resists”. He is pragmatic about this. Because most people and organisations are “so far behind the curve”, the encyclical “perhaps treads a pragmatically useful path as it extends that thinking without breaking it.” Its dignity-centred stance is still “radical” and challenges how individuals use AI, how institutions deploy it and how governments “weaponize it”. - Postscript: the “tool” objection. The framing that “did jar” with him is the encyclical’s treatment of AI as a tool, which he grants is “admittedly nuanced”. “I still worry that treating a technology that has the ability to fundamentally alter how we think, act, and even believe — and in ways that surpass our comprehension — as just a tool, is potentially dangerous.” - Efficacy left open. Whether it “will move the needle or whether it is just wishful thinking” is unclear, “But at least the question is being framed in a way that’s hard to ignore.”

How firmly. He is firm on four points: the three-domain schema, “who we are” as the domain where AI is unprecedented, AI being more than a tool, and LLMs missing the experiential layers of a humane text. His reading of the encyclical is explicitly provisional (“first read through”, “a longer post for another day”). He is agnostic about whether it will have any effect.

Concepts and frameworks. - What we do / where we live / who we are: his “three intersecting foci” for “navigating advanced AI transitions”. The re-quoted 2025 definitions are: where we live (“from our homes and communities to the environment and the planet as a whole — space even”); what we do (“from discovering new knowledge and insights, to creating value”); who we are (“from how people behave and function as collectives in society, to the most fundamental aspects of how we define and understand ourselves as individuals”). Here he re-frames them as “the three domains that map out the terrain around human flourishing and advanced technologies”. - Navigating technology transitions: “something I’ve been exploring for some years now”. - Cognitive coupling: the human–AI coupling that may affect who we are. - Cognitive trojan horse (his own): AI bypasses cognitive defences by broadcasting trust signals. - Constitutive resonance (his own): two-way coupling that changes both participants. - Cognitive surrender (Shaw and Nave; he endorses the alignment). - AI psychosis: the extreme end of the spectrum. - Transactional versus relational: a contrast in education and institutional priorities. - Human dignity and human flourishing: the normative anchors.

Analogies and comparisons. - The industrial revolution and Rerum Novarum. Its insights “continue to be relevant to this day as AI ushers in a new era of automation”. This is used structurally, as a parallel between technology transitions and their effects on work and dignity, not as a literal comparison of risks. - Environmental coupling (Laudato Si’). AI is both an environmental threat and a possible solution. - There are no comparisons with chemicals, nanomaterials or other past risk cases.

Views on AI. - What kind of thing it is: not “just a tool”. It is a technology able to alter thought, action and belief “in ways that surpass our comprehension”. It is something its makers “don’t fully understand” (his endorsement of Olah’s hint). It is unique among technologies in how it acts on who we are. - What is new: the cognitive and identity domain. Conversational AI’s trust signals and two-way coupling change people. - It is inevitable (“the boat has already left the harbor”).

AI risk. - Cognitive and formative risks come first: AI psychosis, cognitive surrender, bypassed defences, accelerated change in how people think and see themselves. - Also named: jobs and value, environment (water, energy, land), weaponisation by governments and others, and (in his list of the encyclical’s themes) concentration of power and “naive and self-centered wielding of power”. - The framing is loss of, or threat to, human dignity and “who we are”, balanced by opportunity.

AI companies and leaders. Chris Olah (Anthropic) is cited approvingly as a sign that frontier developers see something new in what they are making. He does not criticise any company here.

Governance and who decides. He endorses the encyclical as addressed to developers and governments and to “anyone (and any organization)”. Universities carry “a profound responsibility”. He values inclusive dialogue (a “conversation which includes everyone”, quoted from Francis) and “shared discernment” (listed from the encyclical). He is sceptical about how much difference it will make (“wishful thinking”).

Cognition, language and formation. This is the core of the post. His own concepts (cognitive trojan horse, constitutive resonance) bear on how conversational AI reshapes reasoning and self-perception. The postscript applies this to reading: AI summaries strip out lived, experiential meaning, and the reader cannot see what is missing.

What he criticises and who he engages. He engages Pope Leo XIV, Leo XIII, Francis, Olah, Shaw and Nave, and Penprase. He criticises the Vatican’s framing, which neglected Laudato Si’; the encyclical’s AI-as-tool framing and its preserving (not re-examining) stance on being human; people outsourcing their reading of the encyclical to AI; and universities that rush for speed and put the transactional first.

Change of view signalled. Reading about the encyclical made him think about his three-domain model “in a new light”, but he does not name a change of view. He re-endorses the January 2025 schema.

Quotes. - “fall short of addressing the one domain where AI is shaking things up in ways that no other technology has come close to.” - “In a world where AI is inevitable (and I would argue that the boat has already left the harbor here)” - “the cutting edge of AI development is beginning to force a reckoning with long-held assumptions around what it is to be human.” - “As you read an AI-generated summary, what is missing in it, and how will you know?”

2026-06-12 · a-quick-update-on-using-claude-fable-5 · “A quick update on using Claude Fable 5 for research”#

Provenance. The text is his own prose. Two Fable-written PDFs are attached (not in the text): “After The Proxy V3”, a 53-page self-reflexive research report, and “Assessment And Formation Under Agentic Ai V3”, a paper rewritten for legibility. Neither is evidence of his views. The usage figures (over 80 agents, 2,000 tool calls, over 120 primary sources verified, nearly 15 hours, over 8 million tokens, nearly half spent on verification) are “According to Fable’s own audit”, and he reports them without checking. The export-control update note is his.

Argument in his terms. - The setup. He builds on the one-shot test of 10 June. He gives Fable 5 in Claude Code (“ultracode” mode) the original prompt and paper, Fable-generated critical reviews, his own feedback and a folder of cited sources, and then stays hands-off. - Process quality. Fable “devised a plan of action that would put many PhD students to shame”. It developed and tested hypotheses, spawned tens of sub-agents, downloaded and audited primary sources, and “folded in layer after layer of checks and balances”. The first output was “deeply reflective, with meta-layers” and documented its own process. It was “academically quite rigorous” but “not particularly human reader-friendly” (“raw Fable”). - Result. After a legibility rewrite, the paper is “substantially better than the one produced from the one-shot prompt. By a long way.” Its hypotheses, claims, reasoning and evidence “stand up to considerable scrutiny”. He still flags that it “still reads like an AI paper”. - The real problem is evaluation. Because of the “intellectual labor” in the paper, “it takes a substantial level of human expertise and intellectual labor to evaluate it.” He is reaching “my own limits in assessing it’s rigor and validity”, although his expertise overlaps closely with the topic. He would need “days with this — probably more”. This raises “substantial questions around how AI-generated research is evaluated when increasingly few humans have the expertise or intellectual capacity to fully understand and assess it.” A pointed corollary: quick responses to such work “are either coming from genius-class humans, are themselves the product of AI, or are not based on knowledgeable assessment.” - Why it is hard to keep up. “Not because AI is in some way “smarter” or “better” than us. But because it is getting so good at emulating the processes through which new knowledge is constructed and tested”. It does this at speed and with access to prior knowledge “at a scale and depth that far transcends human capabilities”. - Postscript: the counter-risk of illusion. There is “a serious risk … of falling for the illusion that these models are more capable than they actually are.” This is “an inherent risk with a technology which has a mastery of language that is potentially capable of slipping by our critical reasoning and persuading us of things that don’t hold up to scrutiny.” The paper “may be little more than smoke and mirrors”. Even so, “it would be foolish to discount emerging capabilities”. - Scarcity and the university. Higher education lives in “a world built on the assumption that intelligence, expertise, and new knowledge, and valuable because they are scarce.” This should be “an absolute top priority for any university that takes student success and the future of human flourishing seriously”, and far more of a priority than “incessant conversations around pre-2023 level AI capabilities.” - The precipice. “we are potentially at the edge of a precipice where AI systems are capable of generating new knowledge and insights faster than we are currently capable of validating and even understanding them — or their consequences.”

How firmly. He is firm that agentic Fable produced work of real rigour and that evaluation is now the bottleneck. He is openly unsure where this leaves us (“Where this leaves us, I’m not sure”). He hedges with the “illusion” and “smoke and mirrors” caveat and calls this “almost definitely a relatively poor reflection of emerging capabilities.”

Concepts. - Evaluation or validation gap: AI knowledge generation outpacing human capacity to validate and understand it. - Emulation of knowledge-construction processes: his account of what agentic AI does, and deliberately not “smarter”. - Scarcity assumption: the value of expertise and knowledge rests on their scarcity, which AI undermines. - Illusion of capability: mastery of language slips past critical reasoning. This is the same mechanism as his cognitive-trojan-horse idea, applied here to judging AI outputs. - Legibility versus rigour: raw AI output can be rigorous but hard for humans to digest.

Views on AI. Agentic frontier systems are process-emulators with superhuman breadth and speed. The new thing is autonomous planning, verification and multi-agent research. For AI risk, he names epistemic risks: unvalidated knowledge at scale, a collapse of evaluative capacity, and persuasion through language. He also names consequences for expertise and higher education, and for “what it means to thrive as humans in an age of AI”. Cost is visible: on the $200-a-month Max plan he “ironically maxed it out” and had to buy more credits.

Companies, governance. He reports the US export-control suspension without comment. He does not criticise Anthropic, and he reads the model’s released capabilities with interest.

Who he engages. “many commentators” on the illusion of capability; universities.

Quotes. - “it takes a substantial level of human expertise and intellectual labor to evaluate it.” - “Not because AI is in some way “smarter” or “better” than us.” - “a mastery of language that is potentially capable of slipping by our critical reasoning and persuading us of things that don’t hold up to scrutiny.” - “AI systems are capable of generating new knowledge and insights faster than we are currently capable of validating and even understanding them”

2026-07-04 · just-how-good-is-anthropics-fable-as-a-research-assistant · “Just how good is Anthropic’s Fable at researching and writing an academic paper?”#

Provenance. His own prose runs from the opening to “here’s Fable’s own audit/summary”, and the two footnotes are his. Everything under “How the paper was made: a note from the AI collaborator” is written by Fable 5 and is excluded: the process account, the audit table and the reading of the numbers. That section attributes words and judgements to him, for example that the paper was “technically accurate but very tiring to read”. It also describes how the paper was designed (“diagnosis-first”, persuasion as the tracked risk, risk innovation as “practice-based knowledge built with entrepreneurs”). These are the AI’s reports, not his prose. The attached PDF, “The Orphan Risks Of Frontier Artificial Intelligence (nature Perspective) V6”, was drafted by Fable under his direction over six drafts and is not in the text.

Argument in his terms. - Motive. He has “been meaning to write for some time about how my work over the past ten years on how the framing of risk innovation applies to AI frontier models”, and uses Fable to finally do it. The test is to use it where he can judge success himself: “because the original risk model and underlying work are mine”. - Outcome. The paper is “pretty good”. He claims it “makes an original contribution to thinking on AI frontier models and risk” and is “not far off being good enough to submit for peer review with my name on it”. It took nearly two days, six drafts, detailed feedback and meticulous checking of claims and citations, and it was costly in tokens. - What the paper argues (his own summary). It extends risk innovation to frontier AI “using the gap between the internal safety frameworks AI companies develop and the compliance documents that they produce for regulators.” That gap shows the importance of “what I’ve previously referred to as orphan risks”. The paper argues there are “effective and productive ways for companies to close this gap and thus address currently sidelined risks.” He credits Fable with framing that “hadn’t previously occurred to me” and calls it “insightful — and genuinely novel”. He will write about the paper separately (in a later batch). - Credit and limits. “This is not AI acting as an independent researcher.” The contribution “largely arises because of my previous work and my hands-on editing”. He adds: “I’m not sure I would have arrived at the resulting analysis on my own”, and alone it would have taken “at least a couple of months”. - The writing gap. The paper is still “very much a product of AI”. Fable itself admitted its style is “essentially hard-wired in”. He is “deeply suspicious of anyone who claims they can get AI to churn out publishable papers in a matter of hours”. Good work must be well researched and argued, make “a serious and defensible knowledge contribution”, and resonate with human readers “while signaling a level of care and effort in its formation that indicates it’s worthy of someone’s time to read.” - An embodiment argument about the limits of AI. “the research and analysis gap between humans and AI is closing”, but writing that “works on a human level is still elusive”. The reason is that “a disembodied AI thats only frame of reference is the written word” does not know what it feels like to read as “a flesh and blood human”. - Footnote 2 (on his authority and on AI judging AI). He draws on “40 years of working as a scientist”. AI academic writing “is often mimicking assumed academic norms” and is “written in a way that another LLM would find reasonable but that humans find impenetrable (a major issue when using an AI to assess AI writing)”. He admits “I may be wrong here.” - Takeaway. “not that research and writing can be effectively outsourced to Fable-level models, but that as a research partner, these models are capable of seriously augmenting what an established expert/researcher is capable of achieving.” Attention should go to “massively-augmented research and development”. Autonomous R&D is also coming: “I have no doubt that this is coming”.

How firmly. He is firm on the human writing gap and on augmentation over replacement. He is confident, though not certain, about the paper’s originality.

Concepts. - Risk innovation applied to frontier AI. His ten-year framework. - Orphan risks: here “currently sidelined risks”. They are identified through the gap between companies’ internal safety frameworks and their regulatory compliance documents. This is a new application, which he credits to working with Fable. - Massively-augmented research versus autonomous research. - The human–AI writing gap and embodiment. - AI style as hard-wired: it cannot be fully trained out. - Care and effort as a signal of worth: “formation” of a text. - The AI-judging-AI problem: LLM-legible is not the same as human-legible.

Views on AI. A powerful research partner that is closing the analysis gap. It is disembodied and text-bound, which limits how well it can communicate with humans. It has a fixed stylistic signature.

AI risk. Indirect here. The substantive risk argument (orphan risks at frontier labs, and the gap between what companies say internally and what they file with regulators) is carried by the paper and only summarised in his prose.

Companies and governance. Implicitly, AI companies describe and manage risk inconsistently across internal frameworks and regulatory compliance, and could close that gap themselves. This is enterprise-facing governance, consistent with the business framing of orphan risks since 2018. He calls the export-control action a “ban” without comment.

Quotes. - “This is not AI acting as an independent researcher.” - “a disembodied AI thats only frame of reference is the written word” - “makes me deeply suspicious of anyone who claims they can get AI to churn out publishable papers in a matter of hours.” - “not so much in autonomous research and development (although I have no doubt that this is coming), but in massively-augmented research and development.”

2026-06-10 · is-anthropics-new-ai-model-poised · “Is Anthropic’s new AI model poised to change the AI higher education landscape … again?”#

Provenance. His own prose, including the prompt he wrote. The attached preprint “After The Proxy” is unedited, one-shot Fable 5 output, labelled by his instruction as authored by “Anthropic Fable 5 Max”. Its theses, “full proxy collapse” and “disappearing ladder”, are Fable’s, not his. He endorses them as “serious enough and well-argued enough to warrant serious attention” and writes “I have to agree with Claude’s conclusions”. The export-control update notes are his.

Argument in his terms. - Why he tested it. The Fable 5 release “felt too significant to ignore”, even on vacation and sabbatical. He deliberately used a one-shot prompt, “something I would usually never do”, because one-shot research “tends to lead to outputs that are superficially OK and substantially poor.” - His quick take on the output: 1. The autonomy is impressive. Extended into multi-agent human–AI collaboration, “the implications are worth paying attention to”. 2. Hypotheses, research and conclusions are “very good” but “still limited”. For a machine with no human intervention it is “hard to overstate how big a deal this is”. 3. He found no hallucinated citations (“a big deal”). But LLMs use “secondary mentions” rather than primary sources, which risks “inappropriate or naive uses of sources”. 4. The style is “rather flat with an annoying AI signature”, though better than Opus 4.6–4.8, whose prose feels like “fingernails down a chalk board”. 5. The ideas may not be novel, but “What is more important than novelty though is how existing research and theories are brought together”. - Speed. “for a few minutes of AI run time, Fable produced something that it would have taken me several days to do justice to in a non-AI world.” - The core educational claim: from outputs to formation. “This is no longer a technology that emulates the outputs of educational and learning processes, but extends this to the formation of those outputs.” This “threatens to pull the rug from under every effort over the past 3 plus years to accommodate ChatGPT 3-level capabilities”. It also opens “new learning possibilities that we’ve barely grappled with”. - The institutional-agility condition. “As long as we have the agility to move with the models as fast as they are evolving.” In education that is “a big “if.”” Many faculty are “still coming to terms with the changes ChatGPT brought about in 2022.” - Access caveat. Mythos-class models “are likely to be out of reach of most educators for a while yet”. - Footnote. The parallel question for universities is research and knowledge generation. Once Mythos-class models are coupled with multi-layer agent systems, this is “something to watch very closely indeed.”

How firmly. He labels it a “very quick and dirty” first reaction written before boarding a plane and says to “read with caution”. He is firm on the outputs-to-formation shift and on the risk that education will lag behind.

Concepts. Outputs versus formation. Goal-oriented autonomy. The “AI signature” in prose. Primary versus secondary sourcing. Institutional agility and pace mismatch in education. Unequal access to frontier models.

Views on AI and AI risk. A step change in “goal-oriented” autonomy, with “internal guardrails against going off the rails” (the advertised feature he leaned on). The risk is to the foundations of higher education, both assessment and the value of a degree. He frames it as disruption that brings both threat and possibility.

Companies. He reports Anthropic’s claims (“being touted”) and the US export-control suspension without evaluating either. He criticises earlier Claude Opus prose.

Quotes. - “This is no longer a technology that emulates the outputs of educational and learning processes, but extends this to the formation of those outputs.” - “one-prompting research and papers tends to lead to outputs that are superficially OK and substantially poor.” - “for a few minutes of AI run time, Fable produced something that it would have taken me several days to do justice to in a non-AI world.” - “As long as we have the agility to move with the models as fast as they are evolving.”

MEDIUM#

2026-05-15 · ai-movies-may-be-less-dystopian-than-we-think · “AI movies may be less dystopian than we think”#

Provenance. - His prose: the introduction, the commentary under each finding, the “Bottom line” and the notes. - Not his: the charts and tables, “primarily the work of Claude Code as I directed it”, which he chose to use “as-is”. The “Useful stuff” section (the taxonomy of future states and the descriptions of AI portrayal) is excluded per the ruling. - The corpus itself (169 films, 1927–2026, 31 countries, 21 languages) was built with Claude Code. He describes the working relationship this way: “I was driving the research and assessing it at every step, while Claude was my not-always-reliable research assistant”.

Argument in his terms. - The trope. “Everyone knows” AI in sci-fi films is “bad news for the future” (Terminator, The Matrix, Mission Impossible). But “deeper analysis indicates that the connection between AI and imagined futures in movies is more complex than this.” - The trigger. Anthropic’s claim, reported in Ars Technica, that “dystopian sci-fi was responsible for for teaching its models to act “evil,””. He does not rule on the causal claim. He tests the premise that AI films are mostly dystopian. - Findings as he reports them: - 32% of films are dystopian; the other categories dominate when combined; - continuation (22%) and protopia (12%) come next; - the share of dystopian films has fallen since 1927 while protopia rises (he was surprised); - there is “a slight trend toward movies that focus on AI benefits rather than risk”; - US films are the least dystopian, compared with the UK and Japan. - His caveats. He cautions that aggregation hides country trends. He notes that “A blockbuster Hollywood dystopian AI movie will likely have far more influence and impact on society than a small indie movie” and flags weighting by influence as future work. The categorisations are “subjective”, although robustness was checked. He says the study “doesn’t rise to the level of a publication-quality study yet”. - The interesting cases are the counterintuitive ones: dystopia with beneficial AI, continuation with risky AI. Complex portrayals “all over the map”. - Bottom line. “it does start to peel away at assumptions that AI gets a bad rap in films”. He highlights cases “where there is no bright line between artificial intelligence and a particular future state.” The open question is how film portrayal shapes the relationship between AI, society and the future: “a question for another day”. He releases open data (GitHub, a browser tool) and invites others to build on it.

Concepts. He names eight future states: dystopia, utopia, protopia, continuation, inheritance, supersession, heterogeneity, agonism. The definitions are in the excluded section, so only the names count as his. He also uses AI portrayal as risk, benefit, neutral or complex. He cites Kevin Kelly’s “protopia” (through the excluded definitions), and Cave, Dihal, Finn and others in a footnote bibliography.

Views and significance. - Sci-fi as a lens on futures. This continues his long-running film-and-technology work (Films from the Future). - Complicating the dystopia narrative. He resists a flat picture in which AI is always the villain. - Engaging an AI company’s claim. He engages with Anthropic’s claim about training data, and his response is empirical, not polemical. - Method. He models AI-assisted scholarship with human direction and open data. - Firmness. Modest: “not definitive”.

Quotes. - “deeper analysis indicates that the connection between AI and imagined futures in movies is more complex than this.” - “I was driving the research and assessing it at every step, while Claude was my not-always-reliable research assistant” - “it does start to peel away at assumptions that AI gets a bad rap in films.” - “where there is no bright line between artificial intelligence and a particular future state.”

2026-07-10 · i-asked-anthropics-fable-5-to-create-a-video-game-inspired-by-my-work · “I asked Anthropic’s Fable 5 to create a video game inspired by my work. It’s mad!”#

Provenance. - His prose: the introduction, “A quick note on the game’s genesis” (his notes, which also appear in the game), and footnotes 1–4. - Fable-written and excluded: “Making HYPERBUBBLE: a self-audit” and its “Postscript: the final session”. - The self-audit characterises his views, so these are AI paraphrases, not his prose. Examples include his “documented view that public concern is “rarely cut and dried””, “the 18 orphan risks”, “grounded exuberance, catalytic serendipity” and “from disruption to dignity”. The same applies to design features the audit attributes to his pushback: flourishing as the central dial, the “responsibility pause”, and deployment context as the ethics. They may point to his other work, but they are not evidence here. - The tagline “The future is a soap bubble. Ride it anyway” is probably Fable’s framing. Footnote 3 says Fable chose the soap-bubble image from Future Rising and he “let it run with” it.

What he says (his own prose). - Framing. A light-relief “toy” with a “more serious side”. It shows Fable’s abilities and offers “a completely unexpected and delightful perspective on navigating advanced technology transitions.” - Process (his account). Fable researched his work, created five game-designer agents and three AI judges, picked a winner, then iterated with him “somewhere beyond version 20”. - A self-map of his work, in his own words: - the Risk Bites hand-drawn aesthetic; - “orphan risks you adopt (and which — bizarrely — become your pets)”; - named risks from “my work on risk innovation and the often-overlooked risks that trip up emerging technologies”; - the soap bubble from Future Rising; - emerging technologies from “my work in tech innovation”; - panic billboards from techlashed.org; - “the moral-panic fires on my work on risk perception and engagement”; - “the whole trajectory reflects my work on human flourishing and navigating complex advanced technology transitions.” - What it shows about AI. It demonstrates “just how capable AI is becoming as something that can help translate experiences, ideas, research, and a lot more into something quite unexpected and serendipitously delightful.” He ties this to “my work on the future of being human in an age of increasingly complex technologies”.

Significance. Low analytical content, but useful as his own compact inventory of his core threads in mid-2026: risk innovation, orphan risks, risk perception and moral panic or techlash, emerging-technology innovation, Future Rising, human flourishing and advanced technology transitions. It also shows AI as a translator of ideas into new forms, and a playful, delight-oriented stance (“who cares when you can have this much fun”).

Quotes. - “the often-overlooked risks that trip up emerging technologies” - “the moral-panic fires on my work on risk perception and engagement.” - “the whole trajectory reflects my work on human flourishing and navigating complex advanced technology transitions.” - “a completely unexpected and delightful perspective on navigating advanced technology transitions.”

LOW / NONE#