Show HN: I used Claude Code to discover connections between 100 books
trails.pieterma.es494 points by pmaze 3 days ago
494 points by pmaze 3 days ago
I think LLMs are overused to summarise and underused to help us read deeper.
I built a system for Claude Code to browse 100 non-fiction books and find interesting connections between them.
I started out with a pipeline in stages, chaining together LLM calls to build up a context of the library. I was mainly getting back the insight that I was baking into the prompts, and the results weren't particularly surprising.
On a whim, I gave CC access to my debug CLI tools and found that it wiped the floor with that approach. It gave actually interesting results and required very little orchestration in comparison.
One of my favourite trail of excerpts goes from Jobs’ reality distortion field to Theranos’ fake demos, to Thiel on startup cults, to Hoffer on mass movement charlatans (https://trails.pieterma.es/trail/useful-lies/). A fun tendency is that Claude kept getting distracted by topics of secrecy, conspiracy, and hidden systems - as if the task itself summoned a Foucault’s Pendulum mindset.
Details:
* The books are picked from HN’s favourites (which I collected before: https://hnbooks.pieterma.es/).
* Chunks are indexed by topic using Gemini Flash Lite. The whole library cost about £10.
* Topics are organised into a tree structure using recursive Leiden partitioning and LLM labels. This gives a high-level sense of the themes.
* There are several ways to browse. The most useful are embedding similarity, topic tree siblings, and topics cooccurring within a chunk window.
* Everything is stored in SQLite and manipulated using a set of CLI tools.
I wrote more about the process here: https://pieterma.es/syntopic-reading-claude/
I’m curious if this way of reading resonates for anyone else - LLM-mediated or not.
I did something similar whereby I used pdfplumber to extract text from my pdf book collection. I dumped it into postgresql, then chunked the text into 100 char chunks w/ a 10 char overlap. These chunks were directly embedded into a 384D space using python sentence_transformers. Then I simply averaged all chunks for a doc and wrote that single vector back to postgresql. Then I used UMAP + HDBScan to perform dimensionality reduction and clustering. I ended up with a 2D data set that I can plot with plotly to see my clusters. It is very cool to play with this. It takes hours to import 100 pdf files but I can take one folder that contains a mix of programming titles, self-help, math, science fiction etc. After the fully automated analysis you can clearly see the different topic clusters. I just spent time getting it all running on docker compose and moved my web ui from express js to flask. I want to get the code cleaned up and open source it at some point. Thanks for the supportive comments. I'm definitely thinking I should release sooner rather than later. I have been using LLM for specific tasks and here is some sample stored procedure I had an LLM write for me. --
-- Name: refresh_topic_tables(); Type: PROCEDURE; Schema: public; Owner: postgres
-- CREATE PROCEDURE public.refresh_topic_tables()
LANGUAGE plpgsql
AS $$
BEGIN
-- Drop tables in reverse dependency order
DROP TABLE IF EXISTS topic_top_terms;
DROP TABLE IF EXISTS topic_term_tfidf;
DROP TABLE IF EXISTS term_df;
DROP TABLE IF EXISTS term_tf;
DROP TABLE IF EXISTS topic_terms; Can someone break this down for me? I'm seeing "Thanos committing fraud" in a section about "useful lies". Given that the founder is currently in prison, it seems odd to consider the lie useful instead of harmful. It kinda seems like the AI found a bunch of loosely related things and mislabeled the group. If you've read these books I'm not seeing what value this adds. I guess the lies were useful until she got caught? Why lie if it isn’t useful? Lying is generally bad, why do a generally bad thing if there isn’t at least a justification, a “use” if you will. Be careful with the 'utility' model of explaining behavior. It is fairly easy to slide into 'if behavior X is manifested, this must mean X must somehow be useful'. You can use this model to explain behavior, but be aware of the circularity trap in the model. "She lied thus the lie must have had use, even if it is not obvious we will discover the utility if we dig down enough". Another model can be post-rationalization. People just do stuff instinctively, then rationalize why they did them after the fact. "She lied without thinking about it, then constructed a reasoning why the lie was rational to begin with". At the extremes, some people will never lie, even to their detriment. Usually they seem to attribute this to virtue. Others will always lie. They seem to feel not lying is surrendering control. Most people are somewhere in between. Thanos is the comic book villain snapping his fingers. Theranos is the fraud mentioned in the piece. In a similar vein, I've been using Claude Code to "read" Github projects I have no business understanding. I found this one trending on Github with everything in Russian and went down the rabbit hole of deep packet inspection[0]. ValdikSS is the guy behind the SBC XQ patches for Android (that alas were never merged by G). I didn’t expect to see him here with another project! https://habr.com/en/articles/456476/ https://android-review.googlesource.com/c/platform/system/bt... That's a cool idea. There are so many interesting projects on GitHub that are incomprehensible without a ton of domain context. I got the idea from an old post on here called Story of Mel[0] where OP talks about the beauty of Mel's intricate machine code on a RPC-4000. This is the part that always stuck with me: I have often felt that programming is an art form,
whose real value can only be appreciated
by another versed in the same arcane art;
there are lovely gems and brilliant coups
hidden from human view and admiration, sometimes forever,
by the very nature of the process.
You can learn a lot about an individual
just by reading through his code,
even in hexadecimal.
Mel was, I think, an unsung genius. Thank you for sharing that story. Mel seems virtuousic, but is that really art? Optimizing pattern positioning on a drum for maximum efficiency. Is that expression? > Is that expression? If it wasn't expression everyone would get the same result. But no one else at Royal McBee did things the way Mel Kaye did things. Kaye had a strong artistic vision for how things should be done; he didn't want to use the ergonomic features of the RPC-4000 because they didn't align with his vision. I think he found the idea of rigging the blackjack program offensive in part for the same reason. Speaking for myself, I have always found the story and "pessimal" instructions beautiful. It's my favorite piece of folklore of all time. Kaye and Nather are both artists to me. Tangentially, Kaye is standing on the far right in this photo. https://zappa.brainiac.com/MelKaye.png And here is Nather. https://en.wikipedia.org/wiki/Ed_Nather#/media/File:Ednather... If you consider engineering the art of the possible. (Yes, I know it's a politician's phrase, that's because politics is the art of the plausible ... ) I read a book maybe a decade ago on the "digital humanities". I wish now I could remember the title and author. :( Anyway, it introduced me to the idea of using computational methods in the humanities, including literature. I found it really interesting at the time! One of the the terms it introduced me to is "distant reading", whose name mirrors that of a technique you may have studied in your gen eds if you went to university ('close reading"). The idea is that rather than zooming in on some tiny piece of text to examine very subtle or nuanced meanings, you zoom out to hundreds or thousands of texts, using computers to search them for insights that only emerge from large bodies of work as wholes. The book argued that there are likely some questions that it is only feasible to ask this way. An old friend of mine used techniques like this for dissertation in rhetoric, learning enough Python along the way to write the code needed for the analyses she wanted to do. I thought it was pretty cool! I imagine LLMs are probably positioned now to push distant reading forward in an number of ways: enabling new techniques, allowing old techniques to be used without writing code, and helping novices get started with writing some code. (A lot of the maintainability issues that come with LLM code generation happily don't apply to research projects like this.) Anyway, if you're interested in other computational techniques you can use to enrich this kind of reading, you might enjoy looking into "distant reading": https://en.wikipedia.org/wiki/Distant_reading > I wish now I could remember the title and author. LLMs are great at finding media by vague descriptions. ;) According to Claude (easy guess from the wikipedia link?): The book is almost certainly by *Franco Moretti*, who coined the term "distant reading." Given the timeframe ("maybe a decade ago") and the description, it's most likely one of these two: 1. *"Distant Reading"* (2013) — A collection of Moretti's essays that directly takes the concept as its title. This would fit well with "about a decade ago." 2. *"Graphs, Maps, Trees: Abstract Models for Literary History"* (2005) — His earlier and very influential work that laid out the quantitative, computational approach to literary analysis, even if it didn't use "distant reading" as prominently in the title. Moretti, who founded the Stanford Literary Lab, was the major proponent of the idea that we should analyze literature not just through careful reading of individual canonical texts, but through large-scale computational analysis of hundreds or thousands of works—looking at trends in genre evolution, plot structures, title lengths, and other patterns that only emerge at scale. Given that the commenter specifically remembers learning the term "distant reading" from the book, my best guess is *"Distant Reading" (2013)*, though "Graphs, Maps, Trees" is also a strong possibility if their memory of "a decade" is approximate. After some digging, I think it was likely this one: https://direct.mit.edu/books/book/5346/Digital-Humanities I dont understand the lines connecting two pieces of text. In most cases, the connected words have absolutely zero connection with each other. In "Father wound" the words "abandoned at birth" are connected to "did not". Which makes it look like those visual connections are just a stylistic choice and don't carry any meaning at all. Really great work but have to agree with others that I don’t see the threads. The one I found most connected that the LLm didn’t was a connection between Jobs and the The Elephant in the Brain The Elephant in the Brain: The less we know of our own ugly motives, the easier it is to hide them from others. Self-deception is therefore strategic, a ploy our brains use to look good while behaving badly. Jobs: “He can deceive himself,” said Bill Atkinson. “It allowed him to con people into believing his vision, because he has personally embraced and internalized it.” Wow! Amazing! Have you read the Syntopicon by Mortimer J Adler? It's right up your alley on this one. It's essentially this, but in 1965, by hand, with Isaac Asimov and William F Buckley Jr, among others. Where did you get the books from? I've been trying to do something like this myself, but haven't been able to get good access to books under copyright. Yeah, thinking a bit more here, you've created a Syntopicon. I've always wanted to make a modern one too! You can do the old school late night Wikipedia reading session with the trails idea of yours. Brilliant! Really though, how can I help you make this bigger? This feels like a nice idea but the connection between the theme and the overarching arc of each book seems tenuous at best. In some cases it just seems to have found one paragraph from thousands and extrapolated a theme that doesn’t really thread through the greater piece. I do like the idea though — perhaps there is a way to refine the prompting to do a second pass or even multiple passes to iteratively extract themes before the linking step. I don’t like this product as a service to readers (i.e., people who read as a cognitive/philosophical exploit) but I do think that somewhere embedded in its backend there are things of benefit. I think that this sucks the discreet joy out of reading and learning. Having the ways that the topics within a certain book can cross over in lead into another book of a different topic externalized is hollowing and I don’t find it useful. On the other hand I feel like seeing this process externalized gives us a glimpse at how “the algorithms” (read: recommender systems) suggest seemingly disjunctive content to users. So as a technical achievement I can’t knock what you’ve done and I’m satisfied to see that you’re the guy behind the HN Book map that I thought was nice too. At its core this looks like a representation of the advantages that LLMs can afford to the humanities. Most of us know how Rob Pike feels about them. I wonder if his senior former colleague feels the same: https://www.cs.princeton.edu/~bwk/hum307/index.html. That’s a digression, but I’d like to see some people think in public about how to reasonably use these tools in that domain. > Having the ways that the topics within a certain book can cross over in lead into another book of a different topic externalized is hollowing and I don’t find it useful. Intuitively, I agree. This feels like the different between being a creator (of your own thoughts as inspired by another person's) and a consumer (although in a somewhat educational sense). There would need to be a big advantage to being taught those initial thoughts, analogous to why we teach folks algebra/calculus via formulas rather than having every student figure out proofs for themselves. "There are, you see, two ways of reading a book: you either see it as a box with something inside and start looking for what it signifies, and then if you're even more perverse or depraved you set off after signifiers. And you treat the next book like a box contained in the first or containing it. And you annotate and interpret and question, and write a book about the book, and so on and on. Or there's the other way: you see the book as a little non-signifying machine, and the only question is "Does it work, and how does it work?" How does it work for you? If it doesn't work, if nothing comes through, you try another book. This second way of reading's intensive: something comes through or it doesn't. There's nothing to explain, nothing to understand, nothing to interpret."
— Gilles Deleuze I am not familiar with the source of this quote, but I don't disagree, it is just incredibly reductive. Gilles Deleuze him-/her-self was not born and did not live in a vacuum. They were influenced and mimetically reproduced ideas they were exposed to, like we all do. I don't find the point of this project meaningless myself. The opposite in fact. But the results are not accurate for anyone who has actually read any of these texts. The feedback loop you describe—watching Claude's logs, then just asking it what functionality it wished it had—feels like an underexplored pattern. Did you find its suggestions converged toward a stable toolset, or did it keep wanting new capabilities as the trails got more sophisticated? I do this all the time in my Claude code workflow:
- Claude will stumble a few times before figuring out how to do part of a complex task
- I will ask it to explain what it was trying to do, how it eventually solved it, and what was missing from its environment.
- Trivial pointers go into the CLAUDE.md. Complex tasks go into a new project skill or a helper script This is the best way to re-enforce a copilot because models are pretty smart most of the time and I can correct the cases where it stumbles with minimal cognitive effort. Over time I find more and more tasks are solved by agent intelligence or these happy path hints. As primitive as it is, CLAUDE.md is the best we have for long-term adaptive memory. I ended up judging where to draw the line. Its initial suggestions were genuinely useful and focused on making the basic tool use more efficient. e.g. complaining about a missing CLI parameter that I'd neglected to add for a specific command, requesting to let it navigate the topic tree in ways I hadn't considered, or new definitions for related topics. After a couple iterations the low hanging fruit was exhausted, and its suggestions started spiralling out beyond what I thought would pay off (like training custom embeddings). As long as I kept asking it for new ideas, it would come up with something, but with rapidly diminishing returns. This is really, really, good. Ignore the commenters in this thread who don’t see the connections. It takes a very high degree of artistic creativity and linguistic imagination to see these types of connections, and many of the “engineer types” on this forum are unfamiliar with that mode of thinking. Ignore them. Every one of these connected threads are really good. It’s the opposite. The connections are so trivial/obvious as to be uninteresting. I’m not surprised that it found connections when you told it to find connections. Most of those connections seem rather dubious to me. I think you’d have been better off coming up with these yourself. Earlier today, I was thinking about doing something somewhat similar to this. I was recently trying to remember a portal fantasy I read as a kid. Goodreads has some impressive lists, not just "Portal Fantasies"[0], but "Portal Fantasies where the portal is on water[1], and a seven more "where/what's the portal" categories like that. But the portal fantasy I was seeking is on the water and not on the list. LLMs have failed me so far, as has browsing the larger portal fantasy list. So, I thought, what if I had an LLM look through a list of kids books published in the 1990s and categorize "is this a portal fantasy?" and "which category is the portal?" I would 1. possibly find my book and 2. possibly find dozens of books I could add to the lists. (And potentially help augment other Goodread-like sites.) Haven't done it, but I still might. Anyway, thanks for making this. It's a really cool project! [0] https://www.goodreads.com/list/show/103552.Portal_Fantasy_Bo... [1] https://www.goodreads.com/list/show/172393.Fiction_Portal_is... This is a beautiful piece of work. The actual data or outputs seem to be more or less...trash? Maybe too strong a word. But perhaps you are outsourcing too much critical thought to a statistical model. We are all guilty of it. But some of these are egregious, obviously referential LLM dog. The world has more going on than whatever these models seem to believe. Edit/update: if you are looking for the phantom thread between texts, believe me that an LLM cannot achieve it. I have interrogated the most advanced models for hours, and they cannot do the task to any sort of satisfactory end that a smoked-out half-asleep college freshman could. The models don't have sufficient capacity...yet. When I saw that the trail goes through just one word like "Us/Them", "fictions" I thought it might be more useful if the trail went through concepts. The links drawn between the books are “weaker than weak” (to quote Little Richard). This is akin to just thumbing the a book and saying, “oh, look, they used the word fracture and this other book used the word crumble, let’s assign a theme.” It’s a cool idea, but fails in the execution. Yes. It's flavor-of-the-month Anthropic marketing drivel: tenuous word associations edition¹. ¹ Oh, that's just LLMs in general? Cool! I spent 30 seconds and the first word that came to mind was drivel. As an English teacher this shit makes me hate LLMs even more. Like so much techbro nonsense, it completely ignores what makes us human. give it a more thorough look maybe? It’s any interesting thread for sure, but while reading through this I couldn’t help but think that the point of these ideas are for a person to read and consider deeply. What is the point of having a machine do this “thinking” for us? The thinking is the point. And that’s the problem with a lot of chatbot usage in the wild: it’s saving you from having to think about things where thinking about them is the point. E.g. hobby writing, homework, and personal correspondence. That’s obviously not the only usage, but it’s certainly the basis for some of the more common use cases, and I find that depressing as hell. so consider them deeply. Why does the value diminish if discovered by a machine as long as the value is in the thinking? I had a look at that. The notion of a "collective brain" is similar to that of "civilization". It is not a novel notion, and the connections shown there are trivial and uninspiring. This is a software engineering forum. Most of the engineer types here lack the critical education needed to appreciate this sort of thing. I have a literary education and I’m actually shocked at how good most of these threads are. I think most engineer types avoid that kind of analysis on purpose. Programmers tend to lean two ways: math-oriented or literature-oriented. The math types tend to become FAANG engineers. The literature oriented ones tend to start startups and become product managers and indie game devs and Laravel artisans. That doesn’t speak well towards your literary education, candidly. We should try posting this on a literary discussion forum and see the responses there. I expect a lot of AI FUD and envy but that’ll be evidence in this tools favor. lol yes that’s the only reason anyone could find this uh literary analysis less than compelling Build a rag with significant amount of text, extract it by key word topic, place, date, name, etc. … realize that it’s nonsense and the LLM is not smart enough to figure out much without a reranker and a ton of technology that tells it what to do with the data. You can run any vector query against a rag and you are guaranteed a response. With chunks that are unrelated any way. unrelated in any way? that's not normal. have you tested the model to make sure you have sane output? unless you're using sentence-transformers (which is pretty foolproof) you have to be careful about how you pool the raw output vectors I checked 2-3 trails and have to agree. Take for example the OODA loop. How are the connections made here of any use? Seems like the words are semantically related but the concept are not. And even if they are, so what? I am missing the so what. Now imagine a human had read all these books. It would have come up with something new, I’m pretty sure about that. Indeed, I'm not seeing a "so what" here. LLMs make mental models cheap, but all models are wrong, and this one is too. The inclusion of Donalla Meadows' book and the quote from The Guns of August are particularly tenuous. What meaningful connections did it uncover? You have an interesting idea here, but looking over the LLM output, it's not clear what these "connections" actually mean, or if they mean anything at all. Feeding a dataset into an LLM and getting it to output something is rather trivial. How is this particular output insightful or helpful? What specific connections gave you, the author, new insight into these works? You correctly, and importantly point out that "LLMs are overused to summarise and underused to help us read deeper", but you published the LLM summary without explaining how the LLM helped you read deeper. The connections are meaningful to me in so far as they get me thinking about the topics, another lens to look at these books through. It's a fine balance between being trivial and being so out there that it seems arbitrary. A trail that hits that balance well IMO is https://trails.pieterma.es/trail/pacemaker-principle/.
I find the system theory topics the most interesting. In this one, I like how it pulled in a section from Kitchen Confidential in between oil trade bottlenecks and software team constraints to illustrate the general principle. Can you walk me though some of the insights you gained? I've read several of those books, including Kitchen Confidential and Confessions of an Economic Hit Man, and I don't see the connection that the LLM (or you) is trying to draw. What is the deeper insight into these works that I am missing? I'm not familiar with he term "Pacemaker Principle" and Google search was unhelpful. What does it mean in this context? What else does this general principle apply to? I'm perfectly willing to believe that I am missing something here. But reading thought many of the supportive comments, it seems more likely that this is an LLM Rorschach test where we are given random connections and asked to do the mental work of inventing meaning in them. I love reading. These are great books. I would be excited if this tool actually helps point out connections that have been overlooked. However, it does not seem to do so. > Can you walk me though some of the insights you gained? This made me realize that so many influential figures have either absent fathers, or fathers that berated them or didn't give them their full trust/love. I think there's something to the idea that this commonality is more than coincidence. (that's the only topic of the site I've read through yet, and I ignored the highlighted word connections) > we are given random connections and asked to do the mental work of inventing meaning in them How is that different from having an insight yourself and later doing the work to see if it holds on closer inspection? Don't ask me to elaborate on this, because it's kinda nebulous in my mind. I think there's a difference between being given an insight and interrogating that on your own initiative, and being given the same insight. I don't doubt there is a difference in the mechanism of arriving at a given connection. What I think it's not possible to distinguish is the connection that someone made intuitively after reading many sources and the one that the AI makes, because both will have to undergo scrutiny before being accepted as relevant. We can argue there could be a difference in quality, depth and search space, maybe, but I don't think there is an ontological difference. The one that you thought of in the shower has a much greater chance of being right, and also of being relevant to you.
johnwatson11218 - 3 days ago
johnwatson11218 - 2 days ago
EXCEPTION
WHEN OTHERS THEN
RAISE EXCEPTION 'Error refreshing topic tables: %', SQLERRM;
END;
$$; -- Recreate tables in correct dependency order
CREATE TABLE topic_terms AS
SELECT
dt.term_id,
dot.topic_id,
COUNT(DISTINCT dt.document_id) as document_count,
SUM(frequency) as total_frequency
FROM document_terms dt
JOIN document_topics dot ON dt.document_id = dot.document_id
GROUP BY dt.term_id, dot.topic_id;
CREATE TABLE term_tf AS
SELECT
topic_id,
term_id,
SUM(total_frequency) as term_frequency
FROM topic_terms
GROUP BY topic_id, term_id;
CREATE TABLE term_df AS
SELECT
term_id,
COUNT(DISTINCT topic_id) as document_frequency
FROM topic_terms
GROUP BY term_id;
CREATE TABLE topic_term_tfidf AS
SELECT
tt.topic_id,
tt.term_id,
tt.term_frequency as tf,
tdf.document_frequency as df,
tt.term_frequency * LN( (SELECT COUNT(id) FROM topics) / GREATEST(tdf.document_frequency, 1)) as tf_idf
FROM term_tf tt
JOIN term_df tdf ON tt.term_id = tdf.term_id;
CREATE TABLE topic_top_terms AS
WITH ranked_terms AS (
SELECT
ttf.topic_id,
t.term_text,
ttf.tf_idf,
ROW_NUMBER() OVER (PARTITION BY ttf.topic_id ORDER BY ttf.tf_idf DESC) as rank
FROM topic_term_tfidf ttf
JOIN terms t ON ttf.term_id = t.id
)
SELECT
topic_id,
term_text,
tf_idf,
rank
FROM ranked_terms
WHERE rank <= 5
ORDER BY topic_id, rank;
RAISE NOTICE 'All topic tables refreshed successfully';
8organicbits - 3 days ago
Closi - 3 days ago
irishcoffee - 3 days ago
PeterStuer - 2 days ago
Terretta - 2 days ago
theturtletalks - 3 days ago
noname120 - 3 days ago
dinkleberg - 3 days ago
theturtletalks - 3 days ago
coolewurst - 2 days ago
maxbond - 2 days ago
Abstract_Typist - 2 days ago
pxc - 3 days ago
plutokras - 3 days ago
ako - 3 days ago
pxc - 2 days ago
smusamashah - 3 days ago
chrisgd - 2 days ago
Balgair - 2 days ago
urbandw311er - 3 days ago
tolerance - 3 days ago
mathgeek - 3 days ago
bonkusbingus - 3 days ago
drakeballew - 3 days ago
amadeuswoo - 3 days ago
samuelknight - 3 days ago
pmaze - 3 days ago
znnajdla - 2 days ago
catlifeonmars - 2 days ago
nkrisc - 2 days ago
lkbm - 3 days ago
drakeballew - 3 days ago
liqilin1567 - 2 days ago
tmountain - 2 days ago
usefulposter - 2 days ago
georgebcrawford - 2 days ago
rtgfhyuj - 2 days ago
eloisius - 2 days ago
DrewADesign - 2 days ago
rtgfhyuj - 2 days ago
pcrh - 2 days ago
znnajdla - 2 days ago
PinkMilkshake - 2 days ago
znnajdla - 2 days ago
only-one1701 - 2 days ago
znnajdla - 2 days ago
only-one1701 - 2 days ago
what-the-grump - 2 days ago
electroglyph - 2 days ago
baxtr - 2 days ago
wry_durian - 2 days ago
timoth3y - 3 days ago
pmaze - 3 days ago
timoth3y - 3 days ago
varenc - 2 days ago
gchamonlive - 3 days ago
delusional - 3 days ago
gchamonlive - 3 days ago
fwip - 2 days ago
