bhaney
a day ago
Honestly I don't think it would be that costly, but it would take a pretty long time to put together. I have a (few years old) copy of Library Genesis converted to plaintext and it's around 1TB. I think libgen proper was 50-100TB at the time, so we can probably assume that AA (~1PB) would be around 10-20TB when converted to plaintext. You'd probably spend several weeks torrenting a chunk of the archive, converting everything in it to plaintext, deleting the originals, then repeating with a new chunk until you have plaintext versions of everything in the archive. Then indexing all that for full text search would take even more storage and even more time, but still perfectly doable on commodity hardware.
The main barriers are going to be reliably extracting plaintext from the myriad of formats in the archive, cleaning up the data, and selecting a decent full text search database (god help you if you pick wrong and decide you want to switch and re-index everything later).
serial_dev
20 hours ago
The main barriers for me would be:
1. Why? Who would use that? What’s the problem with the other search engines? How will it be paid for?
2. Potential legal issues.
The technical barriers are at least challenging and interesting.
Providing a service with significant upfront investment needs with no product or service vision that I’ll likely to be sued for a couple of times a year, probably losing with who knows what kind of punishment… I’ll have to pass unfortunately.
1vuio0pswjnm7
9 hours ago
But he did not mention anything about creating a "service"
It could be his own copy for personal use
What if computers continue to become faster and storage continues to become cheaper; what if "large" amounts data continue to become more manageable
The data might seem large today, but it might not seem large or unmanageable in the future
namlem
19 hours ago
It would be incredible for LLMs. Searching it, using it as training data, etc. Would probably have to be done in Russia or some other country that doesn't respect international copyright though.
jxjnskkzxxhx
19 hours ago
Do you have a reason to believe this ain't already being done? I would assume that the big guys like openai are already training on basically all text in existence.
IlikeKitties
18 hours ago
In fact, facebook torrented annas archive and got busted for it, because of course they did:
https://torrentfreak.com/meta-torrented-over-81-tb-of-data-t...
HDThoreaun
14 hours ago
Every LLM maker probably did the same. Facebook just has disgruntled employees who leaked it
gpm
12 hours ago
Google goes around legally scanning every book they can get their hands on with books.google.com. Legally scanning every paper they can get their hands on with scholar.google.com.
I doubt they'd resort to piracy for what is basically the same information as what they've already legally acquired...
lcnPylGDnU4H9OF
9 hours ago
That is a good reason to think they did not but it doesn't necessarily override reasons for them to do so. Perhaps it's dubious that the subset of data they could not legally get their hands on is an advantage for training but I really don't know, and maybe nobody does. Given that, Google's execs may have been in favor of similar operations as Facebook's and their lawyers may have been willing to approve them with similar justifications.
sneak
2 hours ago
Downloading a torrent isn't piracy if you are a license holder for the information that you are downloading.
gpm
2 hours ago
*If the license you have authorizes you to make a copy in that fashion.
But here, Google isn't a license holder. Google doesn't license the text in Google Books (unless something has changed since the lawsuits). Google simply legally acquires (buys, borrows, etc) a copy of the book and does things with it that the US courts have found are fair use and require no license.
Incidentally I believe the French courts disagreed and fined them half a million dollars or so and ordered them to stop in France.
ar_lan
10 hours ago
Wasn't this confirmed what Meta does?
https://www.forbes.com/sites/danpontefract/2025/03/25/author...
executesorder66
18 hours ago
> or some other country that doesn't respect international copyright though.
Like the US? OpenAI et al. don't give a shit.
TeMPOraL
14 hours ago
There's a difference between feeding massive amounts of copyrighted material to a training process that blends them thoroughly and irreversibly, and doing all that in-house, vs. offering people a service that indexes (and possibly partially rehosts) that material, enabling and encouraging users to engage directly in pirating concrete copyrighted works.
sellmesoap
6 hours ago
Ironically the low tech infringing proposal would lead to more reliable results grounded in the raw contents of the data, using less computing/power and without the confidently incorrect sycophanty we see from the LLMs.
TeMPOraL
5 hours ago
Nah. It would just lead to more of classical search. Which is okay, as it always has been.
LLMs are not retrieval engines, and thinking them as such is missing most of their value. LLMs are understanding engines. Much like for humans, evaluating and incorporating knowledge is necessary to build understanding - however, perfect recall is not.
Another, arguably equivalent way of framing it: the job of an LLM isn't to provide you with the facts; it's main job is to understand what you mean. The "WIM" in "DWIM". Making it do that does require stupid amounts of data and tons of compute in training. Currently, there's no better way, and the only alternative system with similar capabilities are... humans.
IOW, it's not even an apples to oranges comparison, it's apples to gourmet chef.
r14c
10 hours ago
That's Uber's Gambit. Nothing is illegal for large enough corporations with strong network effects and deep pockets.
TeMPOraL
4 hours ago
That's not Uber's Gambit.
Uber was blatantly ignoring the local laws in order to break into the market and quickly defeat local competition. They used their infinite VC money supply to interfere with and delay investigations and enforcement, betting that if they do it fast enough, they'll have the general population on their side.
LLM vendors found and exploited[0] a legal uncertainty - correct me if I'm wrong, but AFAIK it still isn't settled whether or not their actions were actually illegal. Unlike Uber, LLM vendors aren't breaking into markets by ignoring the laws to outcompete incumbents, and burning stupid amounts of money just to get away with it. On the contrary, LLM vendors are simply providing an actually useful product, and charging a reasonable price for it, while reinvesting it into improving the product. Effects it has on other markets aside[1], their business model is just providing actual value in exchange for money. That's much more direct and honest than most of the tech industry.
The product itself is also different. Uber is selling a mirage, a "miracle" improvement that quickly turns not so, and is destined to eventually destroy the markets it disrupted. LLM vendors are developing and serving systems that provide actual value to users, directly and obviously so.
--
[0] - Probably walked into this without initially realizing it. No one complained 5-10 years ago, where the datasets were smaller and the resulting models had no real-world utility. It's only when the models became useful, that some people started looking for ways to make them go away.
[1] - That's an unfortunate effect of it being a general AI tool, and would be the same regardless of how it was created.
corgi912
12 hours ago
There's this famous phrase in Russian that was born out of a short interview with a woman, a strong Putin supporter, that's often been used as a sarcastic remark for pointing out someone's double standards and/or hypocrisy.
It can be roughly translated to "you don't understand, it's a completely different situation". That's what's constantly on my mind when I'm reading discussions like this one.
Everybody and their dog torrenting petabytes of data and getting away with it (Meta is the only one that got caught and they've still gotten away with doing it)?
The very same data poor American students were forced to commit suicide over? The same data that average American housewives were sued over for millions of dollars of "damages"? The same data that often gets random German plumbers or steelworkers to pay thousands of euros of "fines" to the copyright mafia so they won't get sued and have their lives ruined?
Yet when giant corporations are doing the exact same thing on a massive scale, it's fine? It's not even the same thing, an American student torrenting books isn't making any money off it, while Meta very much is.
Of course it's not the same, a simple-minded and poorly educated person like me isn't capable of understanding the difference. You keep believing in your moral superiority, the rest of the world has finally woken up.
TeMPOraL
4 hours ago
Is there also a famous Russian phrase that translates to "details are irrelevant, it kinda looks similar to me therefore it's the same"? If not, there definitely should be.
The details are the entire point. Arguing that a corporation can get away doing something, while an individual can't, isn't useful, because there are great many of such somethings, and in most cases it turns out perfectly reasonable, once you dig into details.
Exoristos
6 hours ago
There are those who are in charge and those who aren't.
sneak
2 hours ago
> The very same data poor American students were forced to commit suicide over
Leaving the rest of your argument aside, precisely nobody forced aaronsw to commit suicide.
gosub100
10 hours ago
> that blends them thoroughly and irreversibly
It's okay, you can say 'laundering'
TeMPOraL
4 hours ago
I can, but I don't, because that's at best an unintended side effect.
freedomben
12 hours ago
> > or some other country that doesn't respect international copyright though.
> Like the US? OpenAI et al. don't give a shit.
OpenAI is not a country and therefore cannot make laws that don't respect international (or domestic) copyright. Also the US is a lot bigger than OpenAI and the big tech corps, and the law is very much on the side of copyright holders in the US.
diggan
11 hours ago
> the law is very much on the side of copyright holders in the US.
Remind me again what the status of the case is with Meta/Facebook using pirated material to train their proprietary LLMs, and even seeding the data back to the community while downloading it?
SR2Z
5 hours ago
In progress. Nobody is expecting the original protections afforded by copyright to apply here, but the fact that the material is pirated is less relevant than whether or not an LLM is a transformative use of the material.
We will almost certainly see copyright law weakened by the case, but I do not believe that FB will get off with no penalties.
gosub100
10 hours ago
The money is definitely in the side of big tech vs book publishers. There may be a nominal settlement to end the matter, perhaps after a decade of litigation
andrepd
13 hours ago
> Would probably have to be done in Russia or some other country that doesn't respect international copyright though.
Incredible, several years of major American AI companies showing that flaunting copyright only matters if it's college kids torrenting shows or enthusiasts archiving bootlegs on whatcd, but if it's big corpos doing it it's necessary for innovation.
Yet some people still believe "it would have to be done in evil Russia".
DataDaoDe
12 hours ago
OP does have an exaggerated statement - its not like there aren't laws in Russia or something and I largely agree with your sentiment. I think there are levels to this though and its pretty clear that Russia is much riskier than the USA when it comes to IP - just look up anything to do with insuring IP risk in Russia (here's one such example: https://baa.no/en/articles/i-have-ip-in-russia-is-my-ip-at-r...)
Also according to the office of US trade representative, Russia is on the priority watch list of countries that do not respect IP [1] and post 2022, largely due to the war, Russia implemented measures negatively effecting IP rights. [2,3]
If you think it isn't the case and Russia is just as risky as the US when it comes to copyright and IP, I would be interested to know why.
1. https://ustr.gov/about/policy-offices/press-office/press-rel... 2. https://www.papula-nevinpat.com/executive-summary-the-ip-sit... 3. https://www.taftlaw.com/news-events/law-bulletins/russia-iss...
mdp2021
11 hours ago
> evil
In this case and context, a label like "evil" is a twisted interpretation.
sam_lowry_
19 hours ago
LLMs already use it, dude )
exe34
16 hours ago
I think one use would be to search for information directly from a book, rather than get a garbled/half-hallucinated version of it.
echollama
10 hours ago
garbled/half-hallucinated is probably what you would've gotten 8-12mo ago but now adays im sure with good prompting you can pull value from any book.
jdironman
14 hours ago
You don't need AI for that. I get the optimistic spirit of what you mean though.
mdp2021
11 hours ago
Optimized information retrieval of complex text is AI.
bbor
13 hours ago
1. It'd be for the scientific community (broadly-construed). Converting media that is currently completely un-indexed into plaintext and offering a suite of search features for finding content within it would be a game-changer, IMO! If you've ever done a lit review for any field other than ML, I'm guessing you know how reliant many fields are on relatively-old books and articles (read: PDFs at best, paper-only at worst) that you can basically only encounter via a) citation chains, b) following an author, or c) encyclopedias/textbooks.
2. I really don't see how this could ever lead to any kind of legal issue. You're not hosting any of the content itself, just offering a search feature for it. GoodReads doesn't need legal permission to index popular books, for example.
In general I get the sense that your comment is written from the perspective of an entrepreneur/startup mindset. I'm sure that's brought you meaning and maybe even some wealth, but it's not a universal one! Some of us are more interested in making something to advance humanity than something likely to make a profit, even if we might look silly in the process.
Aachen
12 hours ago
> I really don't see how this could ever lead to any kind of legal issue. You're not hosting any of the content itself, just offering a search feature for it.
You don't need to host copyrighted material. It's all about intent. The Pirate Bay is (imo correctly, even if I disagree with other aspects about copyright law and its enforcement) seen as a place where people go to find ways to not pay authors for their content. They never hosted a copyrighted byte but they're banned in some form (DNS, IP, domain seizures) in many countries. Proxies of TPB also, so being like an ISP for such a site is already enough, whereas nobody is ordering blocks of Comcast's IP addresses for providing access to websites with copyrighted material because they didn't have a somewhat-provable intent to provide copyright infringement
When I read the OP, I imagine this would link from the search results directly to Anna's archive and sci-hub, but I think you'd have to spin it as a general purpose search page and ideally not even mention AA was one of the sources, much less have links
(Don't get me wrong: everyone wants this except the lobby of journals that presently own the rights)
It would be a real shame if an anonymous third party that's definitely not the website operator made a Firefox add-on that illegitimately inserts these links to search results page though
DaSHacka
10 hours ago
> When I read the OP, I imagine this would link from the search results directly to Anna's archive and sci-hub
You could just give users ISBNs or link to the book's metadata on openlibrary[0], both of which AA's native search already does.
coolThingsFirst
8 hours ago
Yeah but how does the search work, does it show a portion of the text? If it's a portion of the text isn't that also a part of the book?
carlosjobim
13 hours ago
> 1. Why? Who would use that?
Rather who would use a traditional search engine instead of a book search engine, when the quality of the results from the latter will be much superior?
People who need or want the highest quality information available will pay for it. I'd easily pay for it.
notpushkin
21 hours ago
I think there’s a couple ways to improve it:
1. There’s a lot of variants of the same book. We only need one for the index. Perhaps for each ISBN, select the format easiest to parse.
2. We can download, convert and index top 100K books first, launch with these, and then continue indexing and adding other books.
throwup238
17 hours ago
How are you going to download the top 100k? The only reasonable way to download that many books from AA or Libgen is to use the torrents, which are sorted sequentially by upload date.
I tried to automate downloading just a thousand books and it was unbearably slow, from IPFS or the mirrors both. I ended up picking the individual files out of the torrents. Even just identifying or deduping the top 100k would be a significant task.
WillAdams
16 hours ago
The thing is, for an ISBN, that is one edition, by one publisher and one can easily have the same text under 3 different ISBNs from one publisher (hardcover, trade paperback, mass-market paperback).
I count 80+ editions of J.R.R. Tolkien's _The Hobbit_ at:
https://tolkienlibrary.com/booksbytolkien/hobbit/editions.ph...
granted some predate ISBNs, one is the 3D pop-up version, so not a traditional text, and so forth, but filtering by ISBN will _not_ filter out duplicates.
There is also the problem of the same work being published under multiple titles (and also ISBNs) --- Hal Clement's _Small Changes_ was re-published as _Space Lash_ and that short story collection is now collected in:
https://www.goodreads.com/book/show/939760.Music_of_Many_Sph...
along with others.
palmfacehn
20 hours ago
There should be a way to leverage compression when storing multiple editions of the same book.
bawolff
19 hours ago
From a good search perspective though you probably dont want 500 different versions of the same book popping up for a query
qingcharles
4 hours ago
And without some sort of weighting system, it wouldn't even know which one is the best one to show the user.
palmfacehn
15 hours ago
Agreed. I would prefer to see a single result for a single title. The option of pursuing different editions should follow from there.
tomthe
19 hours ago
I wonder if you could implement it with only static hosting?
We would need to split the index into a lot of smaller files that can be practically downloaded by browsers, maybe 20 MB each. The user types in a search query, the browser hashes the query and downloads the corresponding index file which contains only results for that hashed query. Then the browser sifts quickly through that file and gives you the result.
Hosting this would be cheap, but the main barriers remain..
ThatPlayer
17 hours ago
I've done something similar with a static hosted site I'm working on. I opted to not reinvent the wheel, and just use WASM Sqlite in the browser. Sqlite already splits the database into fixed-size pages, so the driver using HTTP Range Requests can download only the required pages. Just have to make good indexes.
I can even use Sqlite's full-text search capabilities!
Aachen
12 hours ago
I wonder if you could take this one step further and have opaque queries using homomorphic encryption on the index and then somehow extracting ranges around the document(s) you're interested in
Inspired by: "Show HN: Read Wikipedia privately using homomorphic encryption" https://news.ycombinator.com/item?id=31668814
showerst
14 hours ago
How would that scale to 10TB+ of plain text though? Presumably the indexes would be many gigabytes, especially with full text search.
wolfgang42
7 hours ago
The client only needs to get indexes for the specific search; if the index is just a list of TF-IDF term scores per document (which gets you a very reasonable start on search relevance) some extremely back-of-the-envelope math leads me to guess at an upper bound in the low tens of megabytes per (non-stopword) term, which seems doable for a client to download on demand.
qcic
12 hours ago
Super interesting.
greggsy
19 hours ago
It's trivial to normalise the various formats, and there were a few libraries and ML models to help parse PDFs. I was tinkering around with something like this for academic papers in Zotero, and the main issue I ran into was words spilling over to the next page, and footnotes. I totally gave up on that endeavour several years ago, but the tooling has probably matured exponentially since then.
As an example, all the academic paper hubs have been using this technology for decades.
I'd wager that all of the big Gen AI companies have planned to use this exact dataset, and many or them probably have already.
fake-name
19 hours ago
> It's trivial to normalise the various formats,
Ha. Ha. ha ha ha.
As someone who as pretty broadly tried to normalize a pile of books and documents I have legitimate access to, no it is not.
You can get good results 80% of the time, usable but messy results 18% of the time, and complete garbage the remaining 2%. More effort seems to only result in marginal improvements.
bawolff
19 hours ago
98% sounds good enough for the usecase suggested here.
pastage
16 hours ago
Writing good validators for data is hard. You can be 100% sure that there will be bad data in those 98%. From my own experience I thought I had 50% of the books converted correctly and then I found I still had junk data and gave up, it is not an impossible problem I just was not motivated to fix it on my own. Working with your own copies is fine, but when you try to share that you get into legal issues that I just do not feel are that interesting to solve.
Edit: my point is that I would like to share my work but that is hard to do in a legal way. That is the main reason I gave up.
landl0rd
13 hours ago
2% garbage, if some of that garbage falls out the right way, is more than enough to seriously degrade search result quality.
carlosjobim
13 hours ago
It's better than nothing, and nothing is what we currently have.
trollbridge
13 hours ago
Decent storage is $10/TB, so for $10,000 you could just keep the entire 1PB of data.
A rather obvious question is if someone has trained an LLM on this archive yet.
moffkalast
13 hours ago
A rather obvious answer is Meta is currently being sued for training Llama on Anna's archive.
You can be practically certain that every notable LLM has been trained on it.
rthnbgrredf
9 hours ago
> You can be practically certain that every notable LLM has been trained on it.
But only Meta was kind of not so smart to publicly admit it.