# opus[1m] ## Skills Used - contemplate - deep-writing-mode - epistemic-discipline - nick-analytic-voice - writing-standards ## Retrieval Notes - Session id: `bc889e57-5cda-4923-b44b-85d1240ae6d1` - Last activity: `2026-03-18T17:11:42.126Z` - Files touched: `1` ## Artifacts **Modified:** - [[1. The Challenge from Authorship]] ## My Notes <!-- Add your notes here. This section is preserved across syncs. --> ## Conversation ### User opus[1m] ### User /deep-writing-mode/contemplate please catch yourself up on this conversation and give me your own versions of the replacement paragraphs in question in the chat with me, okay? You'll need to read carefully and certainly to the end of the chat so far to be able to give me anything like an adequate response to my query. ### Assistant Let me read that conversation file and load the writing skills. ### Assistant Large file — let me read it in chunks. ### Assistant Now let me read the current state of the actual manuscript file, so I know what I am working with. ### Assistant Good. I have the full conversation and the current state of the note. The Codex session was cut off mid-contemplation while rethinking all five paragraphs from scratch after your latest round of feedback. Let me now read the source extractions so my versions are grounded, and then I will give you my paragraphs. ### Assistant Good. I have all four source texts now. Let me also read the Dellsén chapter extraction for precise page numbers. ### Assistant Good. The public availability point with its definition appears on p. 664. Now let me read the Walton reference briefly to make sure I am not mischaracterising it. ### User [Request interrupted by user] ### User i don't want the twin earth development. ### Assistant Understood. No Twin Earth development. Here is what I have after reading the full conversation. The Codex session went through four rounds of rewriting these paragraphs and was cut off during a fifth — a contemplation about starting from scratch — after your last round of feedback identified structural problems the earlier rewrites had not fixed. The recurring failures across all rounds were: opening paragraphs with bald normative claims the reader has not been walked into; blurring ownership between cited authors and the paper's own voice; meta-commentary transitions that describe the argument rather than making it; and reaching for abstract formulations before the examples or sources have done their work. Here are my versions. Page numbers marked XX need checking against the publications. > Orienting passage: "Both options are unsatisfying. While ascribing the artistic merit to the human user would overlook Midjourney's active contribution, ascribing the merit to Midjourney would downplay the creative activity of prompt-crafting." — Short sentence for the verdict, longer sentence for the thinking. --- ## 1. Putnam Putnam's *The Meaning of 'Meaning'* was not a discovery in that sense. Putnam was not reporting a previously unnoticed item in the world; he was making a case, by way of thought experiment, that meanings are not fixed solely by what is in the speaker's head. The force of the paper does not lie in its having uncovered that fact somewhere outside the text. It lies in the argument itself — the example it develops and the pressure it puts on a familiar picture of meaning. Here the paper is not a report of the work. It is the work. --- ## 2. Williamson with Lipton folded in (replaces both the current standalone Lipton paragraph and the Williamson paragraph) Once the work is on the page, the question is what we look for when we assess it. Williamson treats philosophical theories as answerable to what he calls intrinsic virtues: they should be 'elegant and unified, not arbitrary, gerrymandered, ad hoc, or messily complicated', and they should 'combine simplicity with strength' (2024, p. XX). He connects these virtues to the danger of overfitting. In the sciences, an overfitted model matches the data we have while failing badly on new data; in philosophy, something analogous happens when a view is preserved through successive counterexamples only by acquiring further complications. The post-Gettier programme of analysing knowledge went through exactly this cycle: 'the philosophical community showed very little aversion to the multiplication of complication', as Williamson puts it (2024, p. XX), and a firmer preference for simplicity would have warned that something was going wrong. We want to draw a broader conclusion from this. If overfitting is a genuine vice, then what distinguishes a good piece of philosophical reasoning from a merely surviving one is not that it can be patched to accommodate the latest case. It is that it retains clarity and reach without becoming ad hoc — that it makes the terrain more intelligible rather than merely accommodating more of it. That is the difference, in Lipton's terms, between an account that fits what we know and one that shows why the subject has the shape it does (2004, p. 59). --- ## 3. Bengson + Walton Bengson et al. argue that philosophical theories are assessed at three levels: whether they accommodate and explain the data, whether their claims are properly supported and their commitments hang together, and — only where rivals survive these first two tests — whether they display theoretical virtues (2022, pp. XX–XX). The virtues do not float free; they are, as Bengson et al. put it, 'tie-breakers among respectable theories' (2022, p. XX). Walton et al. show that philosophical arguments proceed through recognisable moves, and that each move is open to recognisable forms of pressure (2008, p. XX). An analogy can be asked where it breaks down; a concession can be asked what it commits the author to. None of this is hidden behind the prose. The reasoning is there on the page. --- ## 4. Dellsén + blind review Dellsén et al. argue that philosophy makes progress by putting people in a position to increase their understanding, and that this happens, in the normal case, through philosophical ideas becoming publicly available (2024, p. 664). If that is right, then the achievement is not a private episode of insight in anyone's mind. It is the public availability of material that others can assess and build on. Whether a given philosophical text is any good is a question about what the text does, not about anything to do with whoever produced it. Blind review rests on exactly that assumption: referees assess the paper in front of them. --- ## 5. Gaut + closing In 1997, IBM's Deep Blue defeated Garry Kasparov at chess. Deep Blue won by searching through vastly more possible positions than any human could and selecting the move most likely to win; as Gaut notes, it is 'the epitome of an uncreative way to play chess' (2010, p. 155). It does not select moves by creative insight but by mechanical search through the space of possibilities. But the moves it plays are still good chess. It won. The difference between Kasparov and Deep Blue is not that one plays well and the other does not; it is that one arrives at good moves creatively and the other does not. If a philosophical text draws the right distinctions and withstands the relevant pressures, that is a success in the text, whatever route produced it. A philosophical corpus that has survived criticism and selection under text-internal standards is not a random body of text. It is a body of text shaped by the repeated application of those standards. The question is what follows when a language model is trained on such a corpus and begins producing texts of its own. --- Two housekeeping notes. The footnote `[^pigliucci]` is now orphaned — no marker in the body text references it, since Pigliucci was removed. It can be deleted or repurposed. And the word 'matters' in that footnote ('where it matters directly to the grounding-style objection') should be changed to 'bears' if you keep it. *Ogni paragrafo dovrebbe guadagnarsi il proprio peso, non accumulare autorità nominali come chi si prepara per un viaggio che non partirà mai.* ### Assistant First, let me update the Putnam paragraph — keeping the ending closer to what is currently in the note. ### User /deep-writing-mode Okay, this isn't going well. So we need to have a bit of a rethink with the structure of this section. 1. Um, first of all, your paragraph number one, Putnam. I like my verse. Oh sorry, yes, replace that. Do that right now. So yeah, put use yeah update the notes so that your new version of the Putnam paragraph is used. Okay. Actually no, the very last sentence of your Putnam paragraph is shit. Whatever comes at the end should be more like the very last clause in the current version of this paragraph on the note. 2. /contemplate now, more substantially I think this is what seems to be a better way of going as this section progresses. So, first of all if we look at the paragraph beginning with this fucking sentence "What, then, does the success of such reasoning consist in?" this is the moment we go wrong. What I think we should do here is even if we don't start the paragraph with this precise wording, the question we are now answering is.. Okay, this is how it should begin. First of all, in some sense it should recall the comparison between science and philosophy that we've just been making. And how about this? It should begin by saying something along the lines of science seems to have clear success conditions and then just half a sentence elaborating on what we can mean by that. Then we get to the question What does good philosophy look like? Or what is good philosophy? And it is here that I think we should start to frame things around the idea of understanding. Okay, and the way we can do this, first of all, is by talking to talking about Dellsén. Okay. And giving a nice detailed account of what their account of philosophical progress is. And that yeah, as we well know, it's about putting people in a position to increase understanding. After that, we move on to Lipton and loveliness. Okay, because then we can talk about individual arguments. Okay, so we've gone one layer further in from the Dellsén progress stuff. Okay, so we're talking about individual arguments and loveliness in terms of explanation being to do with being oriented towards understanding, right? Below this message I'll copy in some text I got from NotebookLM about exactly this sort of stuff. Okay, but that seems to be a good way of bringing in Lipton. Okay, so we start with Dellsén on understanding, then we bring in Lipton and understanding. Okay, and then I don't quite know how, but maybe you can help me out. Then it seems we should be moving towards the Williamson and the Bengson and maybe the Walton stuff. Okay, because that seems to me at least intuitively, maybe you can help me work out whether this is right or not. Seems to me intuitively that they come downstream of the Lipton explanation stuff, but I'm not sure. Maybe I need to think about this a little bit more, but I do, stepping back a bit, I do kind of like going straight into understanding after we've talked about Putnam. So yeah, can you go take all of this into account and go back to the drawing board and give all of these paragraphs apart from the Putnam, which I think is which you can update with your version? Give all of the rest of them yeah, you need we need to rethink, keeping everything I've just said in mind. make sure that you explain your analysis, justification, reasons etc etc. for your answer BEFORE giving me your final answer. Please give me a long, deep, and detailed answer, with lots of options/ideas: as we are in the 'kicking ideas around' stage, I want as much feedback as you can give me so that i have a lot to consider. ### Assistant Updating the Putnam paragraph now — my version but ending with the clause from the current note. ### Assistant Done. Now for the structural rethink. You mentioned you would paste NotebookLM text below your message — I have not received it yet, so I am working from what you said and the sources I have. If you paste it, I will incorporate it. Let me think about what Nick is proposing, because this is a real restructuring, not just a paragraph-level fix. The current order in the section is: 1. Watson/Crick 2. Putnam 3. Lipton (standalone, thin) 4. Williamson 5. Bengson + Walton 6. Dellsén + blind review 7. Gaut + closing Nick wants to reorder around understanding: 1. Watson/Crick 2. Putnam 3. Science/philosophy contrast → what does good philosophy look like? → framing around understanding 4. Dellsén (macro: progress = understanding) 5. Lipton (meso: individual arguments assessed for loveliness = understanding-providing) 6. Williamson / Bengson / Walton (micro: specific virtues) 7. Gaut + blind review + closing Why might this be better? Let me think about the logic. The old order was essentially: here is what philosophy is (Putnam), here is how good explanations work (Lipton), here are the virtues (Williamson/Bengson), oh and also progress is about understanding (Dellsén), and provenance does not matter (Gaut). That has a problem: Dellsén arrives too late. By the time the reader gets to Dellsén, the section has already committed to a picture of philosophical quality without giving the reader a reason to care about THAT picture rather than some other one. Understanding is doing the real work, but it shows up after the virtues, which makes it feel like an afterthought. Nick's proposal reverses this. Start with the big picture: what is philosophical progress about? Answer: understanding (Dellsén). Then zoom in: what does it mean for an individual argument to be oriented toward understanding? Answer: it makes the terrain intelligible (Lipton). Then zoom in further: what specific properties of a text track this? Answer: the virtues Williamson and Bengson identify. Then: these are assessable on the page (Walton), and they do not depend on who produced the text (Gaut/blind review). That is a better logical architecture. It is a zoom from macro to micro, and it gives the reader a reason to care about the virtues before the virtues are listed. The reason is: they track understanding. Now, does the zoom actually work? Let me test it against the sources. Dellsén: "philosophical progress consists in putting people in a position to increase their understanding, where 'increased understanding' is a matter of more accurately and/or more comprehensively representing the network of dependence relations between various phenomena, and where people are most commonly put 'in a position' to increase their understanding by way of philosophical ideas (theories, arguments, distinctions, etc.) becoming publicly available" (2024, p. 664). So Dellsén's account: progress is about understanding, and the mechanism is public availability of philosophical ideas. That does double duty. It gives the understanding frame AND it gives the public-availability point (which feeds into blind review later). In the old order, Dellsén was doing the public-availability job right before Gaut. In the new order, Dellsén does the understanding job first, and the public-availability point comes along for free. Lipton: "the 'loveliest' explanation... the one which would, if correct, be the most explanatory or provide the most understanding" (2004, p. 59). "Likeliness speaks of truth; loveliness of potential understanding." So Lipton explicitly frames loveliness in terms of understanding. That makes the Dellsén → Lipton transition natural. Dellsén says progress is about understanding; Lipton says that at the level of individual explanations, the understanding-providing ones are the lovely ones — the ones that, if true, would show why the phenomenon has the shape it does. But wait — there is a question here about whether the full likely/lovely contrast is needed in Section 1, or just the positive characterisation of loveliness. The earlier conversation decided to narrow Lipton. Now Nick is saying "we move on to Lipton and loveliness... talking about individual arguments and loveliness in terms of explanation being to do with being oriented towards understanding." He is asking for loveliness specifically, framed in terms of understanding. He is NOT asking for the full contrastive apparatus (likely vs lovely). So the narrowing is preserved, but the narrowing is toward the positive characterisation of loveliness as understanding-oriented, not just a vague gesture toward "intelligibility." This is good. It means the Lipton paragraph can be more substantive than the thin version that was there before, while still not being the full IBE lesson. The point is: loveliness = understanding-providing. That is one claim, and it can be stated clearly with Lipton's own words. Now: does Semmelweis stay or go? In the current section, there is a Semmelweis paragraph. But the decision in the conversation was to narrow Lipton, and the Semmelweis/Lipton material was moved to Section 2 as a hidden comment block. If we are now giving Lipton a somewhat larger role in Section 1 (framed around understanding), does Semmelweis come back? I think probably not. Semmelweis is a developed example of the likely/lovely contrast at work. If we are not doing the full contrast in Section 1, Semmelweis does not have a job. What Section 1 needs from Lipton is the characterisation of loveliness as understanding-oriented. That can be done with a quote from p. 59 and a few sentences of exposition. Semmelweis stays in Section 2 where the full contrast is developed. Now, where do Williamson and Bengson go? Nick says they seem to come "downstream" of Lipton but he is not sure. Let me think about this. Williamson's point: philosophy is answerable to intrinsic virtues (simplicity, elegance, generality, unificatory power). The danger is overfitting: preserving a view by piecemeal complication rather than by genuine theoretical improvement. Bengson's point: theories assessed at three levels, with virtues serving as tie-breakers. And — this is from the source — they explicitly tie virtues to understanding: "our method is designed with an eye toward promoting understanding, it ratifies a candidate virtue as genuine if, but only if, its possession somehow contributes to or enhances understanding." So Bengson also frames virtues in terms of understanding! This creates a very nice through-line: Dellsén → understanding (progress) Lipton → understanding (individual explanations) Bengson → understanding (virtues count as virtues insofar as they enhance understanding) Williamson → the specific danger of sacrificing understanding-enhancing virtues in favour of mere survival (overfitting) Hmm, but should Bengson come before or after Williamson? Let me think... Option A: Williamson first, then Bengson. Williamson gives the list of virtues and warns about overfitting. Then Bengson provides the deeper rationale: these virtues are the right ones because they track understanding. This order says: here is what matters (Williamson), and here is WHY it matters (Bengson). Option B: Bengson first, then Williamson. Bengson gives the structured method and the understanding rationale. Then Williamson gives the specific warning about how it goes wrong. This order says: here is the method (Bengson), and here is a vivid example of what happens when you violate it (Williamson/post-Gettier). I think Option B might be slightly better, because Bengson's understanding-link flows more naturally from Lipton's understanding-link. Lipton: individual explanations are good when they provide understanding. Bengson: the method for comparing theories is designed with an eye toward understanding. Then Williamson: and here is what happens when you sacrifice understanding for mere accommodation. The post-Gettier example becomes an illustration of what it looks like when the Bengson method is violated. But actually... Option A has a competing advantage. Williamson is more concrete and accessible than Bengson. Williamson talks about specific virtues (simplicity, elegance) and gives a vivid example (post-Gettier). Bengson is more abstract (three levels, tie-breakers). If Williamson comes first, the reader gets the concrete picture before the abstract method. If Bengson comes first, the reader gets abstract method before concrete picture. In Nick's voice, concrete usually comes before abstract. Examples lead, abstractions follow. So maybe Williamson first (concrete virtues, vivid failure mode), then Bengson (structured method, understanding rationale). Hmm, but that is the old order. Let me reconsider. Actually, there is a third option. What if Williamson and Bengson are in the SAME paragraph? The paragraph says: Williamson identifies the virtues and warns about overfitting. Bengson gives these virtues a more explicit structure and ties them to understanding. That would be tighter and would avoid the question of which comes first. Wait — let me re-read Nick's message. He says "then it seems we should be moving towards the Williamson and the Bengson and maybe the Walton stuff." He is less sure about the order here. He is asking me to help figure it out. So I should present options. Let me think about a few possible orderings for the full section and see which feels best. Ordering 1: Dellsén → Lipton → Williamson → Bengson → Walton → Gaut Zoom: progress (understanding) → individual arguments (loveliness = understanding) → specific virtues (simplicity etc.) → structured method (three levels) → visible structure → provenance-irrelevant. Risk: Bengson feels like a belated attempt to make Williamson more precise. The reader already has the virtues from Williamson; why do they need Bengson's method? Ordering 2: Dellsén → Lipton → Bengson → Williamson → Walton → Gaut Zoom: progress (understanding) → individual arguments (loveliness = understanding) → structured method tying virtues to understanding → vivid failure mode (overfitting) → visible structure → provenance-irrelevant. Risk: Bengson's abstract method before Williamson's concrete virtues. Less natural in Nick's voice. Ordering 3: Dellsén → Lipton → Williamson+Bengson (merged) → Walton → Gaut This might be the cleanest. One paragraph for the virtue story, drawing on both Williamson and Bengson. Williamson provides the specific virtues and the overfitting warning. Bengson provides the structured method and the understanding link. Together they say: here is what good philosophy looks like at the text level, and here is how it connects back to understanding. I lean toward Ordering 3. Merging Williamson and Bengson into one paragraph avoids the question of which comes first and creates a more unified treatment. But it could make the paragraph long. Let me see... The current Williamson paragraph is about 8 sentences. The current Bengson paragraph is about 5. Merged, that could be 10-12 sentences, which is long but manageable if each sentence is doing work. Actually, they might be better as two paragraphs but with a clear logical connection. Williamson: here are the virtues, here is the danger. Bengson: here is why those virtues are the right ones — they track understanding. This reads as one continuous argument in two paragraphs, which is different from two separate arguments. Let me hold both options and see which works better in practice when I draft. Now, the opening paragraph after Putnam. Nick wants it to: 1. Recall the science/philosophy comparison 2. Note that science seems to have clear success conditions (half a sentence) 3. Ask: what does good philosophy look like? 4. Frame the answer around understanding This is the transition from "what philosophy is" (Putnam) to "what makes it good" (Dellsén onwards). It needs to do the work of the old "What, then, does the success of such reasoning consist in?" sentence, but better. Something like: "In the sciences, success conditions are often clear: a theory is confirmed or refuted by experiment, a prediction is borne out or it is not. Philosophy is not like that." Then the question: what does it mean for a piece of philosophical reasoning to be good? Wait — Nick said "science seems to have clear success conditions and then just half a sentence elaborating on what we can mean by that." So the science part should be brief. Then the philosophy question. Then straight into Dellsén: "Dellsén et al. argue that philosophical progress consists in..." This could all be one paragraph: the science/philosophy contrast + the understanding frame + Dellsén. Or it could be two: a short transition paragraph, then the Dellsén paragraph. I think one paragraph is better — less throat-clearing. Let me try: "In the sciences, success often has a clear shape: a hypothesis is confirmed or refuted, a prediction is borne out, a mechanism is identified. In philosophy, the picture is less obvious. Dellsén et al. argue that philosophical progress consists in putting people in a position to increase their understanding..." Hmm, but "the picture is less obvious" is weak. And "a hypothesis is confirmed or refuted, a prediction is borne out, a mechanism is identified" is a triplet. Nick hates triplets. Let me try: "In the sciences, success often has a clear shape: a hypothesis is confirmed by experiment or it is not." One example, not three. Then: "Philosophy is different. What does it mean for a piece of philosophical reasoning to succeed?" Then Dellsén. Actually, better to not frame it as "philosophy is different" because that is too much like "philosophy is not science" which is banal. Let me try a different approach. "Watson and Crick could check their model against the structure of DNA. There is no equivalent check for Putnam's argument about meaning. So what does philosophical success consist in?" That recalls the specific comparison from the opening paragraphs. It does not make a general claim about "science vs philosophy." It stays concrete. Then the question follows naturally. But is that too compressed? Nick wanted "half a sentence elaborating on what we can mean by" science having clear success conditions. So maybe: "Watson and Crick could check their model against the structure of DNA; in the sciences more generally, there are experiments to run, predictions to test, mechanisms to identify." — But that's another triplet. "there are experiments to run and predictions to test." Two items. Better. Let me try: "Watson and Crick could check their model against the structure of DNA; in science more broadly, there are experiments to run and predictions to test. For Putnam's argument about meaning, there is no equivalent check. So what does philosophical success consist in?" Then Dellsén. I think that works. It recalls the comparison, gives the brief science elaboration, and asks the question. All in three sentences. But wait — should this be its own paragraph, or the opening of the Dellsén paragraph? I think it works as the opening of the Dellsén paragraph. The question ("So what does philosophical success consist in?") leads directly into Dellsén's answer. Keeping them together prevents the question from hanging in the air. "Watson and Crick could check their model against the structure of DNA; in science more broadly, there are experiments to run and predictions to test. For Putnam's argument about meaning, there is no equivalent check. So what does philosophical success consist in? Dellsén et al. argue that it consists in putting people in a position to increase their understanding..." That is clean. The question is asked and answered in the same paragraph. No throat-clearing. No meta-commentary. The reader knows what the paragraph is about. Now, do I need to worry about the Dellsén characterisation being accurate? Let me check against the source. Dellsén says: "philosophical progress consists in putting people in a position to increase their understanding, where 'increased understanding' is a matter of more accurately and/or more comprehensively representing the network of dependence relations between various phenomena" (p. 664). So Dellsén's account has two parts: (1) progress = putting people in a position to increase understanding, and (2) understanding = representing dependence relations. Do I need part (2)? It depends on whether the section needs the specific account of understanding, or just the framing that progress is about understanding. I think part (2) is worth including because it makes "understanding" non-vacuous. Without it, "putting people in a position to increase their understanding" could mean almost anything. With it, the reader knows what kind of understanding is at stake: understanding of dependence relations, i.e., understanding of why things are the way they are and how they hang together. That connects beautifully to Lipton's loveliness (showing why the phenomenon has the shape it does). So the Dellsén paragraph should include the dependence-relations characterisation. A block quote might be worth it here — the full definition is informative and specific. OK, let me now think about Lipton more carefully. Nick wants: "we move on to Lipton and loveliness... talking about individual arguments and loveliness in terms of explanation being to do with being oriented towards understanding." So the Lipton paragraph should: 1. Move from Dellsén's macro (progress = understanding) to the individual argument level 2. Introduce loveliness as the property of an argument that makes it understanding-providing 3. Connect this to the question: what makes a piece of philosophical reasoning good? The bridge between Dellsén and Lipton is: Dellsén tells us that progress happens through publicly available philosophical ideas that increase understanding. But what makes a particular argument, theory, or distinction understanding-increasing? Lipton's characterisation of loveliness gives us a handle on this. A lovely explanation is one that, if true, would make the phenomenon intelligible — would show why it has the shape it does. In philosophy, a good argument is one that not only survives pressure but illuminates its subject in this way. Should I include the Lipton block quote (the p. 59 passage)? The full passage includes both likely and lovely. Nick does not want the full contrast. But I could quote just the loveliness part: "we may characterize the best explanation as the one which would, if correct, be the most explanatory or provide the most understanding: the 'loveliest' explanation... Loveliness speaks of... potential understanding" (2004, p. 59) That is a partial quote. Is that misleading? I do not think so — I am not misrepresenting what Lipton says, I am just using the part relevant here. The other part (likeliness) will be developed in Section 2. Alternatively, I could include the full block quote and then say: "The likeliest/loveliest contrast will do further work in Section 2. For now, the point that bears on our question is the characterisation of loveliness: an explanation is lovely when it would, if true, provide understanding rather than merely fitting the data." That might be the most honest approach: show the reader the full passage, then say which part we are using now and promise the rest for later. Hmm, but that risks exactly the problem from before — too much IBE apparatus in Section 1. The earlier conversation decided against this. Nick now wants more Lipton than the thin version, but he still does not want the full apparatus. The compromise might be: a short inline quote (not a block quote) that captures the loveliness-understanding connection, plus a few sentences of exposition. No block quote. No likely/lovely contrast developed. Let me draft: "Lipton puts this in terms of what he calls loveliness: the best explanation is not necessarily the most probable, but the one that would, 'if correct, be the most explanatory or provide the most understanding' (2004, p. 59). A lovely explanation does not merely fit what we know. It shows why the subject has the shape it does." That is two sentences. It gives the reader Lipton's own words, connects loveliness to understanding, and states the positive characterisation. It does not develop the contrast. It does not bring in Semmelweis. It is narrow but substantive. Is that enough? Nick said "you need at least two sentences to make Lipton clear" (referring to the thin version). My version is two sentences but they are working hard. I might want three to make the point really land. What about adding something like: "A theory can be well-supported and still leave the terrain opaque; a lovely theory is one that would illuminate it." That is three sentences total on Lipton: "Lipton puts this in terms of what he calls loveliness: the best explanation is not necessarily the most probable, but the one that would, 'if correct, be the most explanatory or provide the most understanding' (2004, p. 59). A theory can be well-supported and still leave the terrain opaque. A lovely theory is one that would, if true, show why the subject has the shape it does." Hmm, but "a theory can be well-supported and still leave the terrain opaque" — that IS half the likely/lovely contrast. I just said it without using the word "likely." Is that OK? I think so — the point is being made informally, not as a terminological apparatus. And it helps the reader grasp what loveliness means by showing what it is not. Actually, wait. Nick said in the conversation: "I'm still a bit stuck here. So looking at your version one, how is the 'Smoking opium puts people to sleep because of its dormative powers.' How is that the likeliest explanation?" And: "I'm just baffled by this." The problem was that the likely/lovely contrast was not being explained clearly. Now, if I am NOT doing the full contrast, I need to make sure the positive characterisation of loveliness is clear on its own, without relying on the contrast with likeliness. Let me try once more: "Lipton characterises the best explanation as the one that would, 'if correct, be the most explanatory or provide the most understanding'; he calls this quality loveliness (2004, p. 59). An explanation is lovely when it does not merely accommodate what is already before us but shows why the phenomenon has the shape it does. Loveliness, in this sense, is a measure of the understanding an explanation would provide — and that connects it directly to the kind of success Dellsén et al. are describing." Hmm, the last sentence is meta-commentary. "That connects it directly to..." Remove that. Let the connection be implicit. The reader will see it. "Lipton characterises the best explanation as the one that would, 'if correct, be the most explanatory or provide the most understanding'; he calls this quality loveliness (2004, p. 59). An explanation is lovely when it does not merely accommodate what is already before us but shows why the phenomenon has the shape it does." Two sentences. Clear. The Dellsén connection is implicit (both are about understanding). No contrast developed. No likely. No Semmelweis. Substantive but narrow. I think that is right for Section 1. More can come in Section 2. Now, after Lipton, we move to Williamson and Bengson. How should the transition work? One way: "The virtues that Williamson identifies in philosophical theories are, on this picture, the properties that make a theory lovely rather than merely serviceable." That connects Williamson to Lipton. But it is somewhat interpretive — I am saying that Williamson's virtues are loveliness-properties. Is that warranted? Actually, looking at Bengson: they explicitly say virtues contribute to understanding. And Lipton says loveliness = potential understanding. So if Bengson's virtues contribute to understanding, and Lipton's loveliness is about potential understanding, then yes, the virtues are (roughly) the properties that make something lovely. That is my interpretation connecting Bengson and Lipton. I should mark it as such. But maybe I should not force the connection. Maybe the transition from Lipton to Williamson/Bengson should be more natural. Something like: "What, then, are the specific properties of a philosophical text that make it understanding-providing rather than merely defensible?" That asks the question that Williamson and Bengson answer. It follows from the Lipton material (loveliness = understanding-providing) and leads into the virtue discussion. Clean transition. Alternatively: "Williamson's discussion of philosophical methodology helps specify what this looks like in practice." That is more meta. Less good. Or: "What does this look like in practice?" Then go straight to Williamson. Simpler. Or: no explicit transition at all. Just start a new paragraph with Williamson. The reader is zooming in from macro to micro. The transition is structural. I think the last option is best. No transition sentence. Just start with Williamson. The logical flow carries the reader. OK. Now let me think about Walton. Where does Walton go? In the previous drafts, Walton was paired with Bengson (both about methodology/structure). In the new order, Walton could stay with Bengson or move. The point of Walton is: philosophical arguments proceed through recognisable moves, and these are publicly visible. That is the text-internal and public-assessability point. It bridges from "what makes philosophy good" (understanding/virtues) to "these assessments are public and text-internal" (leading to Gaut/blind review). So Walton might work well as a bridge between Williamson/Bengson and Gaut/blind review. The argument structure: virtues make a text understanding-providing (Williamson/Bengson) → these are visible in the text's argumentative moves (Walton) → they can be assessed without knowing who produced the text (Gaut/blind review). That means the ordering would be: Dellsén → Lipton → Williamson/Bengson → Walton → Gaut/blind review With Walton as its own paragraph? Or folded into Bengson? Or folded into Gaut? I think folded into Gaut might work. The Walton point (visible structure) and the Gaut point (provenance-irrelevance) are closely related: because the structure is visible, provenance does not matter. One paragraph can do both. Or: Walton at the end of the Bengson paragraph, as a coda. "And these assessments are made on the basis of the argument's visible structure. Walton et al. show that..." Then Gaut/blind review follows as the provenance-irrelevance conclusion. I think I will try a few orderings and see which draft feels best. Let me move to drafting. Actually, there's one more structural question. Nick said "first of all, in some sense it should recall the comparison between science and philosophy that we've just been making." The Watson/Crick and Putnam paragraphs are making this comparison. The third paragraph should recall it. But should it recall the specific comparison (Watson/Crick vs Putnam) or make a more general science/philosophy comparison? Nick said: "It should begin by saying something along the lines of science seems to have clear success conditions and then just half a sentence elaborating on what we can mean by that." So it is MORE GENERAL than Watson/Crick vs Putnam. It is saying: science in general has clear success conditions (experiments, predictions). Philosophy does not have those, so what are its success conditions? That frames the question in a way that leads naturally to Dellsén on understanding. But it also generalises from the specific case (Watson/Crick) to science in general. That is a bit of a move. Is it warranted? Well, Watson/Crick is an example of empirical confirmation (they found the structure of DNA and confirmed it against the data). The general point is that in the sciences, there are typically empirical checks. In philosophy, there are not (or not in the same way). That general point is not controversial. It just sets the stage. But wait — the sentence "Philosophical views are not usually settled by direct empirical test" was in the old draft. Nick said the first sentence was "kind of cool." So there is something like this general comparison that Nick finds acceptable. The problem was with what followed, not with the comparison itself. Let me draft the opening of the post-Putnam paragraph in a few ways: Draft A (recalling specific comparison): "Watson and Crick could confirm their model by looking at DNA. There is no equivalent check for Putnam's argument about meaning. So what does it mean for a piece of philosophical reasoning to succeed?" Draft B (more general): "In the sciences, success often has a clear shape: a model is tested against the data, a prediction is confirmed or refuted. In philosophy, there is no equivalent test. So what does philosophical success consist in?" Draft C (starting from the old sentence Nick liked): "Philosophical views are not usually settled by direct empirical test. A laboratory result can settle a question in chemistry; nothing comparable settles whether Putnam is right about meaning. So what does philosophical success consist in?" Hmm. Draft A is most concrete. Draft B is clearest. Draft C reuses the old opening. I think Draft A is best because it stays close to the examples the reader already has. It does not need to make a general claim about "science" — it just recalls the specific contrast and asks the natural question. But Nick said he wanted "something along the lines of science seems to have clear success conditions and then just half a sentence elaborating on what we can mean by that." That is more like Draft B — a general claim about science, not a specific recall of Watson/Crick. Though recalling Watson/Crick would be a way of elaborating on the general claim. Let me try combining: "In the sciences, success often has a clear shape — Watson and Crick could test their model against the structure of DNA. In philosophy, there is no equivalent test. So what does philosophical success consist in?" That gives the general claim AND the specific recall. One sentence for science (general + example), one for philosophy, one question. Tight. OK. I think I have enough to draft. Let me now write the paragraphs. One more thing. I need to decide how much Lipton to include. Nick said: "we move on to Lipton and loveliness... talking about individual arguments and loveliness in terms of explanation being to do with being oriented towards understanding." He wants a paragraph on Lipton that is about loveliness and understanding. Not the full contrast. Not Semmelweis. But more than the thin version from before. I think a full paragraph on Lipton is warranted. It should: - Connect from Dellsén (understanding at the macro level → understanding at the argument level) - Introduce loveliness - Explain what loveliness means: understanding-providing, showing why the subject has the shape it does - Use Lipton's own words (p. 59) - Not develop the full likely/lovely contrast - Possibly mention that the fuller treatment comes in Section 2 And Nick mentioned he would paste NotebookLM text about "exactly this sort of stuff" — the Lipton/loveliness/understanding connection. I should write a version that can accommodate that material when it arrives. For now, I will write based on what I have. Let me also reconsider the question of whether Bengson and Williamson should be one paragraph or two. I think two is probably right, because they have somewhat different functions: Williamson: specific virtues + overfitting warning (what can go wrong) Bengson: structured method + explicit understanding link (why these virtues are the right ones) But Nick is less certain about the order. Let me present both orderings as options in the contemplation. Actually, I just realised something. Nick said "I don't quite know how, but maybe you can help me out. Then it seems we should be moving towards the Williamson and the Bengson and maybe the Walton stuff." The "maybe" on Walton suggests he is not sure Walton needs to be here at all. If the section is long, Walton might be expendable. The Walton point (recognisable argument forms) could be folded into a sentence or two within the Bengson paragraph rather than standing on its own. Let me present that as an option too. OK, now truly drafting. Let me write the contemplation output and then the paragraphs. For the Bengson quote: I now think the most useful quote is not the "best theory" line (which Nick called uninformative) but the understanding-link: "our method is designed with an eye toward promoting understanding, it ratifies a candidate virtue as genuine if, but only if, its possession somehow contributes to or enhances understanding" (from the Bengson source at lines ~4172-4199). That connects Bengson to the understanding thread. Or the tie-breaker characterisation: "their role is restricted to that of being tie-breakers among respectable theories." Both are useful. Let me use the understanding one as a block quote (if Nick wants a block quote from Bengson) or inline. Actually, thinking about this more: does the section need block quotes at all? The Watson/Crick paragraph has none. The old Lipton block quote is gone. Maybe the section reads better with fewer block quotes and more integrated prose. Block quotes interrupt the flow. If the section can make its points with inline quotation, that might be tighter. On the other hand, a well-chosen block quote can do real work — it lets the reader hear the source's own voice and see exactly what is being claimed. The old Bengson quote was bad because it was uninformative. A good quote earns its interruption. I think I should write versions without block quotes and note where a block quote COULD go if Nick wants one. That way he can decide. Alright, truly drafting now. Let me assemble the full section structure and write each paragraph. ## The structural logic The new ordering you are proposing is a zoom from macro to micro, and I think it is the right move. Here is why. The old ordering listed the sources in roughly the order they were found useful during drafting. Dellsén arrived late, after Williamson and Bengson, which made the understanding frame feel like an afterthought. But understanding is actually doing the load-bearing work in the section: it is the answer to the question "what does good philosophy look like?" If you foreground it, the virtues (Williamson) and the method (Bengson) have a rationale. Without it, the virtues float — they are just a list of desirable properties with no explanation of why those properties and not others. The zoom: 1. Dellsén (macro): philosophical progress consists in putting people in a position to increase their understanding. This answers the question "what does philosophical success consist in?" at the level of the discipline. 2. Lipton (meso): at the level of individual arguments, the understanding-providing ones are the lovely ones — the ones that, if true, would show why the subject has the shape it does. This moves from what the discipline values to what individual texts do when they succeed. 3. Williamson/Bengson (micro): the specific text-level properties that make an argument understanding-providing rather than merely defensible. Simplicity, elegance, non-ad-hocness, reach. And the warning: overfitting is the failure mode — preserving a view by accumulating complications rather than by genuine illumination. 4. Walton → Gaut → blind review (consequence): these properties are assessable on the page, and they do not depend on who produced the text. ## On whether Williamson/Bengson come downstream of Lipton I think you are right that they do, and here is the argument. Lipton's loveliness is about understanding. Bengson's method is explicitly designed "with an eye toward promoting understanding" — he says that a candidate virtue counts as genuine only if "its possession somehow contributes to or enhances understanding" (Bengson et al. 2022). So Bengson and Lipton are answering the same question at different levels of granularity. Lipton says: good reasoning provides understanding. Bengson says: here is a structured method for evaluating theories, and the virtues in that method are virtues precisely because they enhance understanding. Williamson gives the concrete warning: overfitting is what happens when you sacrifice these understanding-enhancing properties in favour of mere dialectical survival. So the flow is: understanding → what specific properties track understanding → what happens when you sacrifice those properties. Dellsén → Lipton → Bengson/Williamson. Each step follows from the previous. The alternative — Williamson before Lipton — would list the virtues before the reader knows why those virtues are the right ones. That is what the old structure did, and it felt like a list rather than an argument. ## On Williamson vs Bengson ordering Two options, and I am genuinely uncertain which is better. Option A: Williamson first, then Bengson. This gives the reader concrete virtues and a vivid failure mode (post-Gettier) before the more abstract method. Advantage: concreteness first, abstraction second. Disadvantage: the understanding link (Bengson) comes late. Option B: Bengson first, then Williamson. This gives the method and the understanding link before the vivid failure mode. Advantage: preserves the understanding through-line without interruption. Disadvantage: Bengson's method is abstract, and leading with abstraction is not how your voice usually works. Option C: merge them into one paragraph. Williamson gives the specific virtues and the overfitting warning. Bengson provides the deeper rationale (virtues count because they enhance understanding) and the structured method. One paragraph does the whole job. I lean toward Option C. It avoids the ordering problem and creates a more unified treatment. The paragraph would say: the specific properties that make a text understanding-providing are the virtues Williamson identifies — simplicity, elegance, unity, reach. Williamson's warning about overfitting shows what happens when these are sacrificed. Bengson et al. give these virtues a structured role and tie them explicitly to understanding. That is one continuous argument, and it can live in one paragraph. ## On Walton Walton could be its own paragraph, or it could be a few sentences at the end of the Williamson/Bengson paragraph, or it could be folded into the Gaut paragraph. The Walton point is that philosophical arguments proceed through recognisable moves, and that those moves are open to recognisable forms of pressure. That is the "publicly visible" point, which bridges from "what makes philosophy good" to "these standards are assessable on the page." I think Walton works best as a bridge into Gaut — a few sentences at the start of the Gaut/blind-review paragraph, rather than its own paragraph. That keeps the section from having too many short paragraphs. ## On the Dellsén block quote The old Bengson block quote was uninformative. But a Dellsén block quote could work, because the full definition is specific and does real work: > philosophical progress consists in putting people in a position to increase their understanding, where 'increased understanding' is a matter of more accurately and/or more comprehensively representing the network of dependence relations between various phenomena, and where people are most commonly put 'in a position' to increase their understanding by way of philosophical ideas (theories, arguments, distinctions, etc.) becoming publicly available. (2024, p. 664) That tells the reader three things at once: (1) progress is about understanding, (2) understanding means representing dependence relations, and (3) the mechanism is public availability of ideas. All three are doing work in the section. I think this is the right place for a block quote — if you want one anywhere in the section. ## Draft paragraphs I am leaving Watson/Crick (paragraph 1) and Putnam (paragraph 2, just edited) alone. Everything below replaces the current paragraphs 3 through 7 and the orphaned footnote. ### Paragraph 3: Science/philosophy contrast → Dellsén on understanding In the sciences, success often has a clear shape — Watson and Crick could test their model against the structure of DNA. In philosophy, there is no equivalent test. So what does philosophical success consist in? Dellsén et al. argue that it consists in putting people in a position to increase their understanding: > philosophical progress consists in putting people in a position to increase their understanding, where 'increased understanding' is a matter of more accurately and/or more comprehensively representing the network of dependence relations between various phenomena, and where people are most commonly put 'in a position' to increase their understanding by way of philosophical ideas (theories, arguments, distinctions, etc.) becoming publicly available. (2024, p. 664) If this is right, then philosophy makes progress not when a question is settled once and for all, but when arguments, theories, and distinctions become available that put readers in a position to understand the subject better than they did before. The achievement is public, and it is oriented toward understanding. ### Paragraph 4: Lipton on loveliness and understanding What does this look like at the level of an individual argument? Lipton characterises the best explanation as the one that would, 'if correct, be the most explanatory or provide the most understanding'; he calls this quality *loveliness* (2004, p. 59). An explanation is lovely when it does not merely accommodate what is already before us but shows why the phenomenon has the shape it does. An account of knowledge can survive every counterexample put to it and still leave the reader no wiser about what knowledge is; a lovelier account would show why the cases fall the way they do. We return to Lipton's fuller treatment of the likely/lovely contrast in Section 2, where it bears directly on the question of how LLMs generate text. For now, the point is that the quality of a philosophical argument is not exhausted by its defensibility. It lies in how much understanding the argument would provide if its claims are correct. ### Paragraph 5: Williamson + Bengson on virtues and overfitting Williamson identifies the specific properties that make a philosophical text understanding-providing rather than merely defensible. He treats theories as answerable to intrinsic virtues: they should be 'elegant and unified, not arbitrary, gerrymandered, ad hoc, or messily complicated', and they should 'combine simplicity with strength' (2024, p. XX). He connects these virtues to the danger of overfitting. In the sciences, an overfitted model matches the data we have while failing badly on new data; in philosophy, something analogous happens when a view is preserved through successive counterexamples only by acquiring further complications. The post-Gettier programme of analysing knowledge went through exactly this cycle: proposed analysis, counterexample, revised analysis, each revision more gerrymandered than the last, and 'the philosophical community showed very little aversion to the multiplication of complication' (Williamson 2024, p. XX). Bengson et al. give these virtues a structured role. On their account, a theory is assessed first on whether it accommodates and explains the data, then on whether its claims are properly supported and its commitments hang together, and only where rivals survive those tests on whether it displays theoretical virtues (2022, pp. XX–XX). The virtues are not free-standing criteria; they are, as Bengson et al. put it, 'tie-breakers among respectable theories' — and they count as virtues only insofar as their possession 'contributes to or enhances understanding' (2022, p. XX). ### Paragraph 6: Walton + Gaut + blind review → closing Walton et al. show that philosophical arguments proceed through recognisable moves, and that each move is open to recognisable forms of pressure: an analogy can be asked where it breaks down, a concession can be asked what it commits the author to (2008, p. XX). None of this is hidden behind the prose; the reasoning is there on the page. In 1997, IBM's Deep Blue defeated Garry Kasparov at chess. Deep Blue won by searching through vastly more possible positions than any human could and selecting the move most likely to win; as Gaut notes, it is 'the epitome of an uncreative way to play chess' (2010, p. 155). But the moves it plays are still good chess. It won. The difference between Kasparov and Deep Blue is not that one plays well and the other does not; it is that one arrives at good moves creatively and the other does not. If a philosophical text draws the right distinctions and withstands the relevant pressures, that is a success in the text, whatever route produced it. Blind review rests on exactly this assumption: referees assess the paper in front of them, not anything to do with whoever produced it. A philosophical corpus that has survived criticism and selection under text-internal standards is not a random body of text. It is a body of text shaped by the repeated application of those standards. The question is what follows when a language model is trained on such a corpus and begins producing texts of its own. --- ## Remaining uncertainties Three things I am not sure about. First, whether the Dellsén block quote is too heavy for the opening of the understanding discussion, or whether it earns its weight by giving the reader the specific characterisation (dependence relations, public availability). You might prefer an inline quotation and a tighter paragraph. Second, whether the merged Williamson/Bengson paragraph is too long. It is doing two jobs (specific virtues + structured method). If it feels overstuffed, it could split: Williamson on virtues and overfitting in one paragraph, Bengson on the method and the understanding link in the next. Third, whether the closing paragraph (Walton + Gaut + blind review) tries to do too much. It is currently the longest paragraph and carries the section's conclusion. If Gaut needs more room — and you said in the earlier conversation that Gaut needs more elaboration — it might need to split, with Walton ending one paragraph and Gaut opening the next. I have not seen the NotebookLM text you mentioned. When you paste it, I can adjust the Lipton paragraph to incorporate that material. *Il telescopio si inverte: prima la disciplina, poi il singolo argomento, poi la virtù del testo — e il lettore arriva all'ultimo passo sapendo già perché gli importa.* ### User "but when arguments, theories, and distinctions" fucking triple examples make me want to punch you... "An explanation is lovely when it does not merely accommodate what is already before us but shows why the phenomenon has the shape it does. An account of knowledge can survive every counterexample put to it and still leave the reader no wiser about what knowledge is; a lovelier account would show why the cases fall the way they do. " nowhere near enough content about Lipton here. Immediately after this passage, you need to use the hospital example. Okay? And two or three sentences. Making it clear what's going on there. Okay, because otherwise the reader's not gonna have any idea what you're talking about. Okay, so you don't need to worry about the likeliness versus loveliness part of the hospital example but you can show that the love that the hospital example gives us a good account of a a lovely explanation because it shows understand because it leads to understanding, right? It provides understanding. Okay, so you need to do that otherwise no one's gonna have an idea what you're gonna talk about. Following on from that, when you say we return to Lipton's fuller treatment of the lightly loving contrast in section two, that should be a footnote. As should the yeah, that's be a footnote. "Williamson identifies the specific properties that make a philosophical text understanding-providing rather than merely defensible" please be careful not to blur people's ideas between whose ideas are whose. Does Williamson put his ideas in these terms? I'm pretty sure he doesn't. If he does, please correct me, but yeah, I think you're being sloppy here. Again, I think it's because you're rushing. Maybe if you had another sentence or two to sort of make this to introduce Williamson properly, you could make it clear why we're talking about Williamson. Williamson. You don't do that in this first sentence here. "Bengson et al. give these virtues a structured role." %%not how i write%% "'tie-breakers among respectable theories'" fucking single quotes for quota direct quotations from texts? What the fuck are you doing? My god. "In 1997, IBM's Deep Blue defeated Garry Kasparov at chess." I think this paragraph should begin by introducing Deep Blue and then moving into Walton. Let's talk about your remaining uncertainties. No, the Del CEM block is fine. Two, I've already talked to you about the merged Williamson Benggson. At least I think I have. Third I don't know yet because I've asked you to reorganize the ordering of information in your final paragraph. "You have hit the nail on the head: the explanation is indeed circular and redundant, and that is precisely why it is so likely. To understand Lipton's point, we have to remember that "likeliness" in his framework strictly evaluates the probability of a statement being true . In logic and probability, the less a statement actually claims, the less risk it takes, and therefore the higher its probability of being true . A circular or redundant statement takes almost zero logical risk. If opium consistently puts people to sleep, then stating that it does so because it possesses a "sleep-inducing power" (which is exactly what "dormative power" means) is essentially a tautology. Because it is tautological, it is incredibly probable—or "likely"—to be factually true. As Lipton writes, "It is extremely likely that smoking opium puts people to sleep because of its dormative powers (though not quite certain: it might be the oxygen that the smoker inhales with the opium, or even the depressing atmosphere of the opium den)" . Barring those minor caveats, the factual correctness of the statement is virtually guaranteed . However, this exact redundancy is what makes it "the very model of an unlovely explanation" . While likeliness "speaks of truth," loveliness speaks of "potential understanding" . Because the opium explanation is circular, it provides no mechanistic insight, no unification, and no new information . Lipton uses this example to illustrate that probability (likeliness) and explanatory power (loveliness) are often in tension. Explanatory loveliness requires a hypothesis to have "high content" and scope, but there is an inherent trade-off: "the more one says the more likely it is that what one says is false" . A genuinely "lovely" explanation of why opium puts people to sleep would detail the chemical interaction between opiates and the brain's neuroreceptors. That mechanistic explanation provides deep understanding, but it makes highly specific, complex claims that carry a much higher risk of being factually incorrect than Molière's circular joke . In short, the opium explanation is "likely" because its circularity protects it from being false, but it is "unlovely" because that same circularity prevents it from providing any actual understanding . This divergence demonstrates why Lipton insists that an account of induction cannot just be "Inference to the Likeliest Cause" . If we only ever inferred the most strictly probable hypotheses, we would be stuck inferring trivial, redundant statements that maximize probability by sacrificing all explanatory content ." finally, this is the NotebookLM text which I forgot to attach to the previous message. I am by no means saying you need to bring in all or even any of these ideas into the text, but this has helped me understand what lightly really how to understand lightly properly and how it relates to unlovely. Maybe not now, but this should be as part of our thinking going forwards with these ideas. ### User [Request interrupted by user]