B18 notes: 2024-09-28 (1 post)#

Reading notes on Andrew Maynard’s Substack post in batch B18. Only his own prose counts as evidence. Quotes are exact, including his original typos and curly punctuation. Quotes from his prompts to ChatGPT keep the straight quotes and double hyphens they have in the post.

Batch context: late September 2024. The week before, he had written about Google’s NotebookLM and its AI-generated “podcast” feature (2024-09-22). This post came out of that work. He needed a piece of original fiction to feed into NotebookLM, asked ChatGPT (GPT-4o) for one, and ended up in a long author–editor exchange with it. This is Part 1 of a two-part series. Part 2 (2024-09-29, the conversation with ChatGPT about its process) falls in the next batch and is not covered here. The NotebookLM “podcast” mentioned here is AI-generated audio, not the Modem Futura podcast, so the instruction to skip Modem Futura podcasts does not apply to this post.

Relevance summary:

MEDIUM#

2024-09-28 — chatgpt-as-author-part-1-the-story — “ChatGPT as Author Part 1: The Story”#

Provenance. Mixed. About 88% of the post’s ~15,900 words is not his. - His own prose (about 1,450 words): the subtitle, the italic series note, the framing essay (about 1,270 words), the closing note and two footnotes. - His own words written as prompts to ChatGPT (about 420 words): the original prompt and two rounds of editorial feedback, all quoted in the post. These are his own text and are used below as evidence of his judgements as an editor. - Not his (about 14,000 words): the whole novella The Memory Capsule, 10 chapters, the third and final draft, written by ChatGPT (GPT-4o). It is published as an experimental artefact, not as his writing. None of its content (the NeuroDyne “Capsule”, the “Free-Willers”, manipulation of people’s futures, the virtue of uncertainty) is evidence of his views. ChatGPT named the protagonist “Andrew”. He says he has “no idea why” and kept it despite finding it “embarrassing”. The header image is from Midjourney. - What he chose to publish: he did not choose the theme of AI-manipulated futures. His brief to ChatGPT was a Wyndham-style story about “the personal and social consequences of a new technology that’s developed”, seen “through the social waves it produces as seen through the lives of his primary characters”. The theme was ChatGPT’s invention. He does not comment on it in Part 1. He presents the novella as context for Part 2, which he calls “where it gets interesting”.

Argument in his terms. - He revises his view of generative AI as a writer of fiction. He opens with his prior view that ChatGPT’s fiction ranged “from bad, to kill-me-now excruciating”, and adds “ChatGPT, it turns out, is not a very good creative writer.” Then comes the turn: “Or so I though.” The subtitle frames the post the same way: “A week ago I would have said no way.” After three drafts he concludes that the result was “enough to make me revise my ideas of what might be possible with generative AI.” - The quality came from an iterative author–editor process, not from a single prompt. He describes “slipping into the role of human editor to my AI author”. The process ran from a 10,000-word Wyndham-style prompt, to an outline, to ChatGPT’s own plan for staying consistent (“tracking character arcs, narrative themes and motifs”), to Draft 1 (19,500 words, “The Quiet Invention”). Then came critical feedback, Draft 2 (14,400 words, retitled “The Memory Capsule”), more feedback, and Draft 3. He says his feedback was “on form and style, but not on substance”. Footnote 2 credits ChatGPT’s large context window and a conversation “over 172,000 words” long. He thinks “a couple more editorial iterations” would have produced “something substantively better”. - He judges the result honestly and without hype. “It’s still not great writing.” He lists faults: an “obsession with metaphors using body parts to capture emotion” (chest, heart, throat, hands), derivative and predictable passages, a plot device that repeats, and gaps in coherence. Yet: “I’ve read plenty of published humans who I would rank as worse than this”. - It was co-creation, not autonomous machine creativity, though even that is unsettled. “The important point here is that this was a process of co-creation.” ChatGPT alone “would have struggled”. He then hedges: his follow-on conversation with ChatGPT (Part 2) made him “question even this”. - Authorship, ownership and the creative force are now open questions. This is the post’s main conceptual move. He makes a structural comparison with a human writer: if he gave a real person “initial direction and iterative feedback on form and style, but not on substance”, the novella “would unequivocally belong to them” (their byline, their copyright). So with ChatGPT as author: “who owns the work? Who’s byline should appear on it? Who or what was the creative force behind it?” He states the conventional answer, that it is all his and ChatGPT is only a tool, and says that is what he “would have said a week ago”. Now he is “not so sure”, because he “was not responsible for the plot line, the overarching themes, or any of the specific details.” - Part 2 is the real turning point. He calls the follow-on conversation “the game change” and “an honest conversation with ChatGPT as an LLM”. He does not say here what changed his mind. (Part 2 is in B19.)

How firmly. The post is exploratory and experiential, reported as a personal experiment. He is firm that his view of AI’s creative capability has changed, and firm that the quality depended on the editor role. He is openly unsure about authorship and ownership: he raises the questions and does not answer them. Footnote 1 is modest: others have probably gone “down similar rabbit holes”, though he had found no example of “a long and iterative author-editor approach”.

Concepts and frameworks. - Co-creation: human editorial direction plus AI generation. He treats it as the mechanism behind the quality. - Human editor / AI author: a role reversal. The human steps back from content to form and style, and the AI supplies plot, themes and details. - Tool vs collaborator vs partner: he names the three statuses and questions the default (“tool”). - Byline, copyright, “creative force”: his criteria for authorship. Responsibility for plot, themes and details is the test he applies. - He does not develop a named framework. The concepts are raised as questions.

Analogies and comparisons. - The human author–editor relationship, used structurally. What would be true of a human author with the same editorial input is used to test whether ChatGPT is only a tool. - Teaching: a comparison in passing. He notes that his harsh feedback to ChatGPT is “not the sort of feedback I give my students”. His editorial role resembles, but deliberately departs from, his role as a teacher. - Published human authors, as the benchmark for quality. - John Wyndham, whom he has been “re-reading … over the summer”, as the model for fiction about “the personal and social consequences of a new technology”. This choice fits his long-standing use of science fiction to explore technology’s social effects. - No comparisons with past technologies (chemicals, nanomaterials and so on).

Views on AI. - What kind of thing it is: a system that can produce passable long-form fiction when given iterative direction. It can analyse its own writing, plan revisions and keep track of narrative elements across a long context. His second-round prompt voices the earlier criticism: “There’s no human soul to the story – it feels like you are mimicking what a human would write, but with no comprehension of what it means to be a person”. It frames AI writing as mimicry without understanding of personhood. His later prompt says Draft 2 had added “maturity and soul”. The post implies that the gap narrowed through editing, without saying whether the mimicry critique still holds. - Unreliable self-report: asked for 50,000 words, ChatGPT kept “telling me it was thinking about things, and then admitting that it wasn’t actually doing anything”. This is a small observation that ChatGPT’s claims about its own process are unreliable. - What is new: capability that emerges through long human–AI interaction and a large context window, rather than from one-shot output.

Views on AI risk. None stated directly. The authorship and copyright questions touch on governance, but he does not frame them as risks.

AI companies and leaders. No direct comment. OpenAI’s ChatGPT (GPT-4o), Google’s NotebookLM and Midjourney are used as tools.

Governance and who decides. He implicitly asks who holds rights and credit (byline, copyright) in AI-authored work, and does not answer. There is no call for regulation.

Cognition, language and human formation. Only indirect. What makes fiction good is layered, non-linear storytelling, maturity and “human soul”. His first feedback prompt says ChatGPT writes “like an 8 year old that has a 30 year old’s grasp of grammar and language”. Being human is the implicit yardstick. The post does not discuss how AI shapes human thinking.

What he criticises and who he engages. He criticises his own earlier assumptions and the “conventional wisdom” that AI is only a tool, questioning both rather than rejecting them outright. He engages ChatGPT (directly, as author), Wyndham (as model), readers (asking for comments on “both the process of creation here and the resulting novella”) and, implicitly, anyone who has run similar experiments.

Change of view signalled. Yes, explicitly and twice: on whether generative AI can write passable fiction (“Or so I though”; “A week ago I would have said no way”) and on whether AI in such work is only a tool (“This is what I would have said a week ago. Now I’m not so sure”). This picks up a thread from his 2023-01-31 annual-review experiment (see B07). There, working with ChatGPT “felt like working with a kind and able colleague”, and it “slightly worries me” that he was “already thinking of ChatGPT as a colleague and a collaborator”. The 2023 post described how collaboration felt in professional writing. This post asks who gets the credit for creative authorship. He now admits that his formal default had still been “tool”, and he is no longer sure it holds.

Quotes (his own prose). 1. “the quality of the novella was enough to make me revise my ideas of what might be possible with generative AI.” 2. “Of course, the important point here is that this was a process of co-creation.” 3. “Conventional wisdom says that this is all mine — helped by ChatGPT of course, but only in the role of tool, not a collaborator or partner.” 4. “Now I’m not so sure though as I was not responsible for the plot line, the overarching themes, or any of the specific details.”

Supplementary (his prompt text): “it feels like you are mimicking what a human would write, but with no comprehension of what it means to be a person”

LOW / NONE#

None. The batch has only the one post above.