# Transcript — Draft Discussion, 31 March 2026 Cleaned transcript of a discussion between Nick Young and Enrico Terrone about the most recent draft of the generating philosophy paper. `[unclear]` marks places where the source transcript was too garbled to reconstruct confidently. `[?]` after a word marks a best guess. --- ENRICO: Maybe presenting it — whether it would be more effective presenting three challenges to this idea that LLMs can do philosophy first. The challenge from authorship, say — and that's the result, to the text, or the end of Section 1, more or less. That's Section 1. Yeah, exactly. In a sense, presenting Section 1 already as an introduction — say, oh, we have this nice example from science fiction, then we have reality examples of science, then can we do philosophy? Tonight [?] — or, that can be true? We are considering three main objections. The first objection is that philosophy is in the philosopher, and we can in a sense address this objection or this challenge by distinguishing between text-oriented and non-text-oriented conceptions. And so, instead of — because now it's more introductive in this way, that would be already the first part. Yeah. And then — so, make the paper more — and then we — Also, the titles are nice, but probably calling them in a separate way gives more structure. Yeah. Yeah. The challenge from authorship, the challenge from abduction, and the challenge from experience or phenomenology. NICK: OK. Could I just — stick with the — so, tell me exactly how this first section — how is that a problem, the challenge from authorship specifically? So I just want to make sure I understand it before I go. Yeah. ENRICO: So the challenge is basically: you need a person to do the philosophy. Is that the idea? Or — well, what's the specific issue? Yeah. Yeah — man, I wonder — maybe my question is, does it make sense to call that a challenge? That's basically the point — but it seems — the kind of — also considering discussions with others and what we have in the [paper], where we are presented so far — this seems like a sort of resistance to [the idea of] de-centring the philosopher, because you see philosophy as something that is in the making of philosophy. In a sense, this can be a first challenge. And that makes it work, because if we've already introduced the sort of scientific advances, that's maybe an interesting contrast case. Exactly. In science, people will be happy to say: yeah, it's possible. Yeah, science — we don't care about who, basically. We don't study the — in philosophy you also have this weird thing that history of philosophy is considered part of philosophy, whereas nobody thinks that history of science is making science. So one reason for that may be that in philosophy you have such a strong connection between philosophers and ideas that you have to study the philosophers historically to better understand the ideas. I think we can [deal with this]. It's from authorship, and it's also easy to deal with, because you say: OK, maybe you have two accounts of philosophy, one person-based, the other text-based, but we think that the text-based [account] is robust enough. And there's the peer review argument, which seems to show that we indeed rely on this text-based conception. NICK: OK, good. That's clear. And I just wanted to make sure I understood the idea of it. That makes — yeah, I think I can make that work. And second — now, we are looking at the [draft]. I still — even though you say you wrote to say that's not the point because you just changed Sections 2 and 3, but Section 1 — I still prefer to read it all the way through. And this seems a bit beside the point — probably it will become more [relevant] later, when we — but at this point, I wonder whether we really need this distinction, also because it seems in tension with Dellsen, because as far as I remember, Dellsen is thinking that philosophical progress is just like scientific progress. So this attempt at the very beginning to draw a distinction between science and philosophy — in which science involves discoveries whereas philosophy doesn't involve discovery — I wonder whether we really need that at that point, because it comes back in Section 3. ENRICO: Yeah, it comes back when we discuss [Pigliucci], and probably there it is more relevant. But at that stage also, because one may object that maybe also Putnam is discovering something about linguistic behaviours — linguistic behaviours are there in the world, they just are — just, maybe not [the right place] at this stage to engage with this. Yeah. NICK: This way is correct, apparently? It's correct? It's OK? Yeah, yeah, yeah, yeah. I did look this up, if you check. Yeah. ENRICO: So, starting somewhere around the Dellsen paragraph — yeah. We may — maybe then, if we decide to turn this into a first challenge, we can have a — maybe yeah, get something from the first section. Here, I already told you, but you haven't changed that — your problem — instead of "it's abrupt" — [the draft] has [something like] "Williamson calls this overfitting" — [what's needed] is something like: "On the other hand, there is no significant progress when likeliness prevails at the expense of loveliness. Williamson calls this overfitting." There's the need for a sentence that — yeah, bridges the paragraphs. NICK: Yeah. ENRICO: Well, [that's] a small thing. [I] want just to note here: I would put a line break here — like, paragraph break. Yes. The distinction between product and process. NICK: Got it. ENRICO: And then this is — I'm noticing more and more — that's clearly a hallmark of LLMs, and something that LLMs tend to do often: always say "this is not X, this is Y." Yeah. Usually we just say "this is Y" — we don't have the [negative formulation]. In the corpus there is a lot of this way of speaking, so it's incorporated. It's true that we also do that, but LLMs do that — it's a real habit. With the other one — and I actually have a problem on the way I'm using the LLM at the moment, which is these triplet examples — X, Y, and Z — so many. NICK: But I have a specific tool that I run now that says: remove every single [triplet]. ENRICO: [unclear — possibly: "Sure, I repeat the judgement about whether it's a genuine understanding of the subject"] — yeah, got it. And probably at the beginning — in sum, a philosophical [claim] — because it's a bit, again, abrupt. The — this is not clear how [it] connects to the previous [paragraph]. So I would have a new paragraph and maybe say something like "In sum" — OK. NICK: Yeah. One — by the way, one more thing about that "it's not X but Y." I think that's also possibly a quirk that has been developed through training, because if you think that every word is determining the word that comes next — yeah, I wonder if you sort of train it to always use these sorts of phrases, you're going to get more analysis following on. You see what I mean? Because if you just say "it's X" — yeah, you'll get less stuff. But if you say "it's not X but Y," you get more stuff. So — yeah, [that might be] why it's doing that. ENRICO: Yeah. Here — because we are using — otherwise it seems a bit repetitive. So — is it Section 3? Which section — you know, we are still exactly towards the very end, but since we already have — yeah, here it seems that we never mentioned Lipton before, but we have — "but as seen above, Lipton distinguishes" — Here, also, I will make it a bit more explicit. So: "Would a system confined to [language] — that [produces text] — would [it] not have access to those features?" NICK: Yes. ENRICO: So, a bit quick. Yeah. I would say — add a sentence, something like: "In a similar vein, [one] might [argue] that philosophy would not [be possible] without access to the proper evaluative criteria. But philosophy is different in this respect." I think it's a way of making the argument a bit easier to follow. NICK: Yep. ENRICO: And again, a mere rhetorical condition — it's [an em dash] instead of using — this is another LLM typical thing: I find a lot of — oh, the em dash. Yeah. Yes. But not — yeah. In general, the — inciso [Italian: parenthetical] — I don't know, it's putting a sentence into — yeah, like an interjection — and, yeah. But I think in this case the answer is that [specific content]. [Discussion of punctuation] ENRICO: But what's the point? [An apostrophe?] Now — it's not a dot. I was calling — when you — the bottom [mark] — like, comma — is this — yeah, this is a full stop. Full stop in British English, period in America. Period, right? Period. And then: "This answer, however, grants what matters for our purposes" — instead of having the — oh yeah. NICK: Yeah, I'd say "their answer" — yeah, that — without the em dash, or whatever. ENRICO: So, [comma], and then — yeah, exactly, that. Yes. Or even better: "This answer [is that...]" In this way, we just have a complete sentence, and then we can put a period after the quotation, after page 12 also. In this way we have two sentences. I think it's easier to read. "This answer..." Yeah, exactly. OK, great. Now we can finally go to the real new section. So, Section 3. I wonder whether we can just cut, or maybe — because the beginning, it's a bit confusing. And "the text-internal reply from Section 2 should [be] the [starting] point, where available." Oh, I — I am [of] the impression that we can also start with "Philosophy proceeds by abduction from the armchair, Williamson argues." But maybe — if you think this first sentence is important, [it would be] better to make it more [intelligible?]. Yeah, yeah. OK. Then here I will just add: "from a corpus" — then: "a system that produces text from a corpus with the right quality properties." Then here I would add "according to Zahavy," because otherwise others may think that Newtonian mechanics faced an empirical crisis. So it's just — so, "according to [Zahavy]," just to be a bit more careful. [It's] missing — "according to Zahavy." Sorry — "according to" — oh sorry, it's "in accordance [with]." NICK: Yeah. OK. Yeah. In this way, we are not committed to big claims in the history of science. ENRICO: This is another thing with LLMs as well, which is worth watching: they're really bad at [distinguishing] — blending the author's view with the quotations. NICK: Yeah, you've got to be really careful about that. ENRICO: Yeah. Here, it's a bit — not super clear through this space, at least to me. "What [Einstein] did was imagine being inside an elevator uniformly accelerated through deep space" — do we really need to be in deep space? Probably yes. But anyway, my [concern] is: "enclosed [or: released] objects would appear to fall with identical acceleration, regardless of composition." Composition means the matter there. Yeah. It's not very clear — was it [clear to you]? [Nick acknowledges] ENRICO: OK. Again, here, I would just say: "as what Pigliucci calls the discipline's starting points." Period. And then something like: "He characterises those [starting points] as" [quotation], etc. So — OK, because [otherwise it's a] super long sentence. So, but: "The world enters philosophy as what Pigliucci calls the discipline's starting points." Yeah, he calls [them] the discipline's starting points as — and then the [relevant] quotation. NICK: Have you managed to read the Pigliucci yet? ENRICO: Yes. I like it a lot. NICK: It's interesting, right? ENRICO: It's a very good paper. Because also [the claims he] makes can actually [apply to] something completely [different] as well. Yeah. Yeah. NICK: [It] started with Alexander's [?] — getting his philosophy thing [going]. ENRICO: Have you read [it]? NICK: I haven't yet. I tried — it's not good. It's — no, it's good. It's not super convincing, but yeah. I have half of that [read]. With us, to the point — we can incorporate something. It seems like a relevant thing to — ENRICO: OK. I can really go back to that, finish that, and then think [about how to use it]. Yeah, yeah. I mean, I put it in the same folder in which I have all this AI-and-philosophy stuff. Yeah. Here — yeah, I would say, just: "Assuming that the philosophical contribution is something [that the text does]" — without "without exception, one argues" — I think we don't need [that]. We can just [say]: "Assuming that the philosophical contribution is something that [the] text does, the question then is..." This is where things get hard, buddy. Yeah. Wait, wait, wait. But I like it a lot. I think it's almost — it's always going in the right direction. It's — this, here, I think it's too early. We don't need to show that Einstein — [or the] Einstein [thought experiment] — to do with [how we] organise sensory experience. I think it's misleading, because we have to make a point about philosophy, so that's a further thing. But it's — oh, I wonder whether this can go maybe later, when we talk about — after the philosophical case is more [established], or in a footnote, or even — NICK: OK. But surely not at that point? ENRICO: I find [that] at that point it's abrupt and it looks like something that's beside the point. NICK: OK. Interesting, because this is around here — ENRICO: OK, no, carry on. Maybe I'll think about it. The more you say — just because — yeah, it's interesting. It's — yeah. NICK: I found this part just tough — to get the ordering of information right. Because this clearly connects well to [what follows] — because the sentence is: "The question then is whether a corpus of ordinary language preserves the everyday experience that [Pigliucci] identifies as philosophy's starting points." And so: Moore looks [at] coins; the same for Putnam. So it seems that if we are talking about philosophy, inserting that piece on Einstein — it's even — sure, we'll say "oh, even Einstein" — but before going to the limit case, maybe yeah, [we should] start with the philosophically [straightforward] cases. ENRICO: You make it sound simple. Sorry, I've driven myself crazy trying to work this out. NICK: Now you say it, it sounds obvious. Yeah, yeah. Because it's just a paragraph. It doesn't really [have] a big effect — just where to move it, or whether we really need [it]. Because in a sense our point is that philosophy can manage with [what's in the corpus]. Then this applies also — we don't — At least, there is room for manoeuvre — for resisting the [objection]. [That] makes it [work]. Yeah, please. ENRICO: Then this is our [main] point. Then, the science issue — we can just have some consideration that depending on [whether] one thinks there is continuity between philosophy and science, or science is something special — we can have a section, maybe on [that]. But I think at this stage it's very good to have just the two philosophical cases. OK? NICK: The Moore is OK. ENRICO: The Putnam — I wonder where the knowledge [fits] — in the middle of the Putnam case, [in terms of] knowledge, I wonder whether this [is] knowledge or more something like intuition, in [the way we discuss] science. We can — NICK: I'll be back in 15 minutes, OK? What's happening? ENRICO: I have a student — a meeting with the student. NICK: OK. Do you want me to go? I'll go [out] there? Yeah, give me a second. --- [Break — approximately 15 minutes] --- NICK: So, remember this thing with — I think it's Dewey, and when he talks about certain forms of art: so painting is like the crystallisation of visual experience, music is the crystallisation of hearing. Yeah. Maybe we can say thought experiments are like the crystallisation of — yeah — issues, or problems, or something? ENRICO: Yeah. You have to [make it] — yeah, they just make more salient situations that [already exist]. So maybe this [crystallisation framing] can be [useful]. NICK: Then — so, that's — I think [that] is the real theoretical problem. I think [the question is how to handle] the [main] case, which we — at the end — we considered that maybe — and also the Merleau-Ponty case — but whether the — it's because, as it stands, [the argument is]: oh, a lot — most of the [thought] experiments, most intuitions philosophy relies on, are already there in ordinary [language] uses. And so only exceptional cases like Mary and — [or things that] get close to the zombie [case] — but I wonder whether there is more — maybe many philosophical intuitions are more like the Mary case. But anyway, that's something I think we can still [work out]. Then there is [the] Einstein [bit], that — probably this is where [it belongs]. Where you think? ENRICO: Yeah, exactly. We can reuse the previous bit. But yeah, that's probably where more elaboration is still needed, like this. NICK: I wonder — I was also wondering whether maybe literature, in the sense of artistic literature — can also, or even not artistic, just diaries, journalism, memoir, or blogs — can yield these intuitions and phenomenological descriptions that may — yeah, that may support this sort of reliance on experience. So this sort of secondhand experience — maybe, [even] when [descriptions] are not sedimented in philosophical texts, they can be sedimented in other texts. ENRICO: I — probably — sorry, I get it. Yeah, it just clicked. Yeah, what you're saying — yeah, and in other — another sort of factual, non-fictional writing, right? It's kind of — yeah. Or even fictional, because often in fiction, people — it's based on [experience]. [Thought] experiments often are already there in fiction, or just more fine-grained. And an LLM can just extract from that the relevant [material]. NICK: This notion of secondhand experience — [that] phrase got left [out], because you used it before and it was meant to be in here. So — you use the phrase — I'm like: "a repository of secondhand experiences." Yeah, which is a phrase [I had]. I was going to put [it] back in. ENRICO: In this way, probably we can have a more — now it's a bit made too sharp — [as if we] say, oh, we have the Putnam and Moore cases that we can easily deal with, and then there are the difficult cases: Mary, and [what you] call also [the] Einstein elevator [case]. Maybe [it's better] just to [present it as a] continuum, and, depending on how much secondhand experience [is] codified in language — we have [it] even from different sources, like literature — and the more [of that] we [have], the more we can also deal with these cases. NICK: That seems more interesting. Yeah, because that — if we were to do another section about prompting — the — yeah, right. So you can have a prompt with all of the philosophy, using the prompt, and [on] the other [end] it's all trivial. And then you can slowly move from the other end — right? — [all the] way to — yeah, "what is the meaning of life?" And it gives you the [appropriate response]. Yeah. ENRICO: Good. Maybe — also, for [a] more editorial thing — I — I wonder whether we could use a fourth challenge, which is the challenge from prompting — something like: oh, but, you know, the philosophy is in the prompting — and then say: OK, but even if [it] is collaborative, so — [structuring the paper] So we have — it's also, I think, nice, because the challenge, in a sense — prompting is symmetrical with the challenge from authorship. Because [with authorship, the objection is]: oh, [there's no author]. No, no, we say: no, it's in the text. But [with prompting], they say: oh, but the text still has some kind of authorship, because there is the prompter. And then the [middle] two are much more connected, because abduction and phenomenology — these [have] more to do with whether certain features of the text can be — so, [they] are more intrinsic: whether the text can have certain philosophical merits without an abductive process and [without] experience of the [world]. Whereas the first and the last would be more relational. So: how can there be philosophy without a philosopher? Or: how can there be LLM philosophy without a prompter? Yeah. NICK: OK. That's — [that's] a nice structure. This would also make the paper a bit longer, which is good. I think — because it's a big issue. ENRICO: So yeah, [it would be] a substantial [contribution]. Yeah. NICK: Yeah. Because there's one other idea I've started to think about. I don't know if it's this paper or another one. Have you heard of something — it's in LLMs — called "The Bitter Lesson"? The paper from 2019? The lesson was basically: it's all to do with just general computation. So, whenever you want to sort of say "how do we advance AI?" — [the answer is]: never specialise, always [go] general. OK? So, original[ly] — a few years ago, apparently people thought: well, say Bloomberg, and you have all of your news archives — what you should be able to do is just have a perfect specialist news bot. So, go to Bloomberg. Yeah. And that doesn't work, apparently, right? What works much better is having a general model and then making it specialise. And that's interesting for [philosophy], because if you think about "hanging together in the broadest possible sense" — yeah. And so, this is something else to think about. So [the idea] would show that you would have a better [outcome]. ENRICO: If you [trained on just] the masterpiece [works], all the papers of the [particular] field — but it's better having also the internet. It's better having everything. It's better having blog posts. NICK: Yeah, that maybe can connect also to the last point we were discussing — this idea that you need secondhand experience, and usually it's easier to find that maybe in novels or diaries rather than in philosophy. So maybe — with the — I don't know if it's an explanation, but the idea that you just don't need only the corpus and all the methodologies and the results of the discipline. You also need the empirical basis of the discipline, which is — yeah. ENRICO: Exactly. The [breadth is] good. And that's actually also interesting because a few years ago — especially — honestly, Nicholas Silins [?] — he trained up an LLM on Daniel Dennett's philosophy, and I think — I don't know, I think it was quite a small project — but [the idea was]: can the students tell the difference between the Dennett bot and Dennett? NICK: Interesting contrast case of what we've just said there, because that is a specialised thing. We can say: we shouldn't be doing that — we should be doing this much broader [approach]. And [I] wonder whether this [broader approach] can also be creative, or [whether it's] just recombining [existing] ideas — or can [it] really deepen [things]? Because in art there are also these cases, like the Next Rembrandt, in which they just created [a painting from] all the Rembrandt paintings. And that's really probably a better Rembrandt than those made by Midjourney. But it's much less creative — you just have a variation on [a] Rembrandt. Yeah, yeah — it's style [transfer]. ENRICO: And then also — the philosophy and engineering [event?] went very well, and Adisa [?] was there. Also, I discussed it with her. She thinks that this aesthetic functioning [idea] is a very good [one]. NICK: Oh, good. We just need to find the time. Yeah. Probably after that. Yeah, yeah. Good, good. What am I saying? So Wednesday, I see you. Yeah. Wednesday and Thursday we have things as well, right? ENRICO: Yes. NICK: And I'll try and have something. Yeah — if you have something already — something? Yeah, something small we can discuss. Yeah. Maybe what I'll do is I'll focus again on Section 3 for the next few days. Yeah. ENRICO: Yeah. And then, yeah, probably adding this Section 4 on prompting. And [it] also maybe helps us to solve some problems, because at least [at one end] it's OK — but even because — yeah, there is the ideal case in which you just write: "Please, [take these] 50 [sources], make this the paper [on] this topic, the deadline is tomorrow." Yeah, yeah. But then also, it seems — at least the weakest case — [is one] in which the prompter also plays a role. And then we can also rehearse the generating model. It's already something, because in a sense it's [been] proven that we can — you can do good philosophy by deferring a significant part to [the] LLM. NICK: So [there's a] weaker claim and then the bigger claim. ENRICO: Yeah. OK. Good. Now, I'm sorry we haven't got — I don't know about you, but when I'm writing a paper: fun, fun, fun until the last [twenty percent]. Yeah, I know. NICK: I'm going to run to my train. So I'll say — yeah, I'll see you Wednesday. My intention after I come back from my [break] is to spend more time in the office, which is maybe good news or bad news. ENRICO: Oh, yes. So yeah, you'll see me in the office just sweating in the corner. Sweating and whining. NICK: Ah, OK. I think I forgot this one. Have a good evening.