This page shows how The Eyepiece was written, in dialogue between Laurentius Paulus and a language model (Claude), on the morning of 12 October 2026. The prompts were written in Finnish; they are given here in English translation, as close to the original wording as English allows. The model’s answers are summarised, not quoted in full; where the model proposed something that was rejected, changed or corrected, that is said. Times are Helsinki time.
Background
The entry grew out of work on a pamphlet the previous day (11 October), where the model was asked to look more closely at Plato’s Phaedrus and Galileo’s Sidereus Nuncius as two older suspicions of new instruments.
The dialogue
1 · 06:45
“So let’s write about suspicion towards machines: writing, the telescope, the LLM.”
Model: proposed a Journal entry in seven phases (Phaedrus → telescope → language model) and asked what had prompted the idea that morning.
2 · 06:47
“The frame I had in mind: I have sent I don’t know how many proposals to editors and experts, the latest ones for example, and no one answers, not a word, complete silence. I think of Galileo’s telescope: suddenly there is a machine that seems to overturn the whole picture of reality. The suspicion is not directed at what we believed before, but at the device that is sensed to distort reality. From here we get to the question of the prompt.”
Model: accepted the frame and the analogy; added three cautions: the comparison concerns instruments, not persons; the reason for the silence cannot be known; publication should wait until open deadlines have passed. Proposed a three-act outline: The Silence / The Instrument / The Prompt, with Kepler as the turn.
3 · 06:49
“Maybe there is also this: I have had fourteen years to let this mature. The conclusions and A Tale of a New Era were born as a lived life. Now someone gets the results straight in front of them. That is quite different; it comes out of nowhere, even though the AI hype says something is happening.”
Model: added the asymmetry (fourteen lived years against results received at once) to the Pathos, and proposed that the Laurentius Paulus journal itself is the missing “path”.
4 · 06:50
“It may be that the texts I offer need a little digesting. Yes.”
Model: added this to the opening and began writing prose for the first act.
5 · 06:50
“Let’s put this essay on the desktop in three acts.” / “An outline — don’t write finished text.”
Correction: the model had started writing finished prose. It set the prose aside and produced an outline only.
6 · 06:53
“This doesn’t need to start with ‘Anyone who has…’. We can forget the structure of the previous essay (general experience – my experience – conclusion). This entry is about suspicion of technology and how to get past it. That essay — was it Copernicus’s? — that explains how the lens works is a good analogy for explaining how the LLM works.”
Model: the book that explained the lens was Kepler’s Dioptrice (1611), not by Copernicus. Proposed that the third act becomes “a Dioptrice for the language model”.
7 · 06:55
“We can also bring in the old experience of research as slow work: go to the library, find the source, copy it. The research I do breaks everything ‘sacred’ in research, because I don’t make the effort; the knowledge just drops when asked, because I have thousands of years of data at my fingertips. Of course a scholar whose life rests on seeking and testing knowledge looks at what I do with the language model and concludes that it can’t work like that.”
Model: added this; proposed that the effort was where the testing happened, and that the search has become easier but the testing has not.
8 · 06:56
“Someone spends their whole life producing research that I summarise and use in a minute. That doesn’t feel fair. And someone working with an LLM may look suspicious in itself.”
Model: added fairness and the suspect figure; proposed “making the debt visible” by naming sources and scholars.
9 · 06:57
“Let’s map the web of suspicion aimed at this project. I want to show that the author of the Tale sees his position also as it looks from outside.”
Model: produced a map of eleven strands of suspicion, each with what is justified in it and an answer. Two (religion, personal crisis) were left out of the entry.
10 · 06:58
“And Kepler’s essay on how the lens works is our climax, chapter 4, because it answers the content of this chapter.”
Model: restructured the outline with Dioptrice as the climax and the web of suspicion as the Pathos.
11 · 06:59
“How much has the media talked about how an LLM works? It feels like a bit of a mystery to people. Is that where the suspicion comes from? When you know how something works, you also know what it can and cannot do, don’t you?”
Model: searched and found: Searchlight Institute & Tavern Research (August 2025, 2,301 US adults: 28% know the model predicts the next words, 45% think it looks answers up in a database, 51% know it can make things up); Reuters Institute, Generative AI and News Report 2025 (weekly use 18% → 34%); Tully et al., Journal of Marketing 2025 (lower AI literacy, greater receptivity). It found no reliable measure of how much the media has explained the mechanism.
12 · 07:03
“Dioptrice — yes. Let’s do something similar for the language model in this entry, bringing in the suspicions and the research results. If it needs a longer essay, we can change the chapters from three paragraphs to five.”
Model: checked the contents of Dioptrice (written August–September 1610, printed in Augsburg 1611; Galileo had promised a theory of the instrument and never published it; the book explains image formation and spherical aberration) and wrote a seven-chapter outline with five paragraphs each.
13 · 07:13
“Words, concepts and NUMBERS — many don’t notice at once that numbers are also related to each other; alone they are nothing.”
14 · 07:14
“Doesn’t the LLM translate words into numbers precisely so that it can model them?”
Model: confirmed and explained the steps: text into pieces (tokens), pieces into rows of numbers, relations computed, probabilities for the next piece, back into words.
15 · 07:17
“Let’s sketch the visual models on the desktop in the right places right away.”
Model: drew four sketches (Sky and Glass, Two Sides of Not Knowing, A Dioptrice for the Language Model, The Eyepiece).
16 · 07:24 – 08:31
Drafts chapter by chapter: “Let’s write out the climax — first draft”, “chapter 5 draft”, “let’s write chapter 3”, “chapter 2”, “then chapter 6”, “then chapter 7”, “and the first chapter too — and the post right away”.
Correction made by the model while drafting: it had dated Kepler’s own observation of Jupiter’s moons to 1611; it corrected this to 1610.
17 · 08:06
“Isn’t language an order in which signs get their meaning? Then mathematics is language.” / “Yes, this numbers-and-words thing has to be articulated.”
Model: agreed and added the distinction between a formal language (mathematics) and ordinary language; the model translates between them. It suggested Galileo’s remark that the book of nature is written in the language of mathematics (Il Saggiatore, 1623) as a link.
18 · 08:43
“Should chapter 7 tell this essay’s prompts and how the LLM was used in articulating the historical analogy — the reference itself came from me.” / “This essay’s prompts, answers and re-prompting go on one separate page, as a link.”
Result: this page.
19 · 08:52 – 08:54
“Yes — that’s good. Let’s update it once the journal entry is finished.” / “Yes — finish the post with its headings on the website.” / “And let’s publish right away — add a link to the prompts at the end. English, i.e. the translations, is enough for us; you can say that separately somewhere.”
Model: checked the remaining facts before publication (see below), corrected one sentence, added category, tags and a header image, and published the entry together with this page.
20 · After publication
“Then I would like to send this essay for publication as well — would it be The New Atlantis, or what?” / “Let’s put an email draft with the pitch into the Mail program.”
Model: suggested The New Atlantis as the closest fit, and drafted a pitch for a longer essay of about 3,500 words that would set the personal frame aside and develop the history and the explanation, with links to this entry and to this page. The author reviewed the draft and sent it from his own address on 12 October 2026. In the author’s words, he would not have sent it had the entry not become presentable through this dialogue.
21 · 09:41 (after publication)
“In that essay Plato stays at the level of a mention — is there anything more than a reference to the text?” / “The pharaoh’s suspicion would be good to show.”
Model: checked the published entry and found that Plato was not in it at all; the idea of writing as the first suspected instrument had dropped out between the outline and the drafts. It added the scene from the Phaedrus (274c–275b), in which King Thamus of Egypt refuses Theuth’s gift of writing, to the last paragraph of chapter I, and added the dialogue to the sources.
22 · 09:43 – 09:47
“An interim question: what would be the Dioptrice of writing — is there one? It is a technique whose distortion is still being caught today.” / “So hermeneutics arose to explain the distortion.” / “Any suggestion how the journal entry could be extended without breaking the structure — raise the question of a Dioptrice for writing, not settle it yet.”
Model: answered that writing had no single Dioptrice; trust in it grew from practices (textual criticism, from Lorenzo Valla to Karl Lachmann, and hermeneutics, from Schleiermacher on), while a theory of writing as a technology came late, closest in Walter Ong’s Orality and Literacy (1982), which compares Plato’s objections to writing with those raised against computers. It proposed one sentence-level change to the first paragraph of the Pathos section, which raises the question without settling it.
23 · 09:47
“Now that we have changed a detail, it is right that you go through the whole text once more — paragraphs to one length and the structure in order. Update the prompt page too.”
Model: added the Ong passage to the Pathos section and Ong to the sources, then shortened the four paragraphs that had grown long (the Thamus paragraph, the first paragraph of Pathos, the paragraph on formal and ordinary language, and the paragraph on who brought what) so that every paragraph is again about eighty words. The structure, five paragraphs to each phase, was left unchanged.
Who brought what
- The author: the frame (the silence), the analogy (the telescope and the suspicion aimed at the device), the idea of the lens-explaining book as the climax, the fourteen years, the library and fairness, the suspect figure, numbers alongside words, language as an order that includes mathematics; after publication, the suspicion of King Thamus and the question of a Dioptrice for writing.
- The model: historical details (Horky, the Bologna demonstration, the contents of Dioptrice, Galileo’s unpublished theory, spherical aberration), the survey results, the map of suspicions, structural proposals and first drafts of the prose; after publication, the details of the Phaedrus scene, Walter Ong and the draft of the pitch.
- Corrections: the model corrected the author (Kepler, not Copernicus); the author corrected the model’s approach twice (outline instead of finished prose; no borrowed paragraph structure); the model corrected its own date for Kepler’s observation, and removed one unconfirmed quotation after checking.
Checked before publication
- The Bologna demonstration in April 1610: confirmed. Magini and Caprara could not see the moons of Jupiter.
- Horky: confirmed that he called the images an optical illusion and published Brevissima peregrinatio contra Nuncium Sidereum (1610). A sentence attributed to him in the first draft (“wonders on earth, deceives in the heavens”) could not be confirmed from the sources found, and was removed.
- Galileo’s letter to Kepler of 19 August 1610 about learned men refusing to look: confirmed.
- Kepler’s own observations of Jupiter’s moons, 30 August – 9 September 1610, with a telescope lent by the Elector Ernest of Cologne, published as the Narratio: confirmed.
- Galileo’s passage on the book of nature written in the language of mathematics, Il Saggiatore, Rome 1623: confirmed; used in paraphrase.
- Added after publication: Theuth and Thamus in Plato’s Phaedrus (274c–275b), and Socrates on the written text that cannot answer questions (275d–e): confirmed. Walter J. Ong, Orality and Literacy (1982), chapter 4, comparing Plato’s objections to writing with objections to computers: confirmed.