(shrug)
Watch and see what happens. I’m rarely wrong in my areas of expertise, especially given the number of backchannels I have based on my position and reputation in the industry, but I could be. In this case, I highly doubt it.
Well, predictions are more interesting. I’m not recalling any specific ones so refresh me: what do you anticipate happening to Anthropic, or Anthropic doing, that will not happen to/be done by any other company?
I don’t know. Why don’t you kneel before Claude and ask “him”? I mean, if it’s not past your bedtime according to Claude. ;)
Do not under any circumstances post the reply here from that thing or it’s an insta-lifetime-ban on this Discourse instance. I hope I don’t have to tell you that I’m not kidding – but just so we’re absolutely crystal clear on this point: I’m not. Fuck around and find out, Curtis J. Sampson.
On scanning and then destroying books (there is a valid legal defense for this, e.g. 1-for-1 “replacement”) but.. as usual.. Misanthropic is an outlier.
so it’s safe to say that Anthropic is an outlier in their choice here
Yes—with an important qualification. Among major AI companies whose practices are publicly documented, Anthropic appears to be an outlier in the scale and deliberate design of its program: purchasing millions of physical books, cutting off their bindings, scanning them, recycling the originals, and retaining a private digital library.
It is not safe to say Anthropic is necessarily the only company doing this. Recent bulk purchases suggest other AI developers or data contractors may be acquiring physical books, but the buyers and intended uses are often concealed, and there is not yet equivalent evidence linking another named major AI laboratory to an Anthropic-scale destructive-scanning program.
Anthropic is therefore:
- an outlier in publicly proven physical destruction at industrial scale;
- not an outlier in using copyrighted books as private AI-training material;
- possibly not unique, given the secrecy of data-acquisition supply chains.
So the most defensible phrasing is: “Anthropic is the clearest and largest publicly documented example, rather than conclusively the only company engaged in the practice.”
That’s the summary, but the fuller discussion is quite illuminating, at least it was to me..
Anthropic Book Scanning Process.pdf (525.5 KB)
You really need to do actual research instead of just relying on an LLM to tell you what you want to hear.
Anthropic is not at all “an outlier in publicly proven physical destruction at industrial scale”; it’s small beans compared to the industry that really destroys printed paper at scale: libraries.
Nor is this new; it long predates computer scanning. Quite a number of libraries have copied collections (particularly newspapers) to microfilm or similar and then destroyed the originals. Plenty more withdraw books from circulation and simply destroy them without scanning them at all.
Destroying books to scan them isn’t anywhere near as unreasonable as it seems so long as you’re not doing it with rare volumes. Having 50000 copies instead of 49999 copies of some book extant in the world is so little difference that it might as well be none.
Well.. several of the questions you pose here are literally answered in that PDF, so… right back atcha, champ? I dunno. Reading Is Fundamental, and all.
Newspapers not noted for their durability, though. Kind of a completely different scenario. Without looking it up, well screw it, I will look it up..
The lifespan of a book depends almost entirely on the chemical makeup of its paper. In reasonable indoor conditions (away from direct sunlight, extreme humidity, and pests), a modern hardcover book can last 500 to 1,000 years or more, while a newspaper will typically crumble into dust within 20 to 50 years.
That’s a massive difference.
Having 50000 copies instead of 49999 copies of some book extant in the world is so little difference that it might as well be none.
as I said above “several of the questions you pose here are literally answered in that PDF”, and yeah, it does depend on the book, a broad rule of thumb is about 1% of books will be rare enough to be affected. Whether any of those books are meaningful, I don’t know.
There’s certainly an overwhelming legal reason here, but if you are of the “burn the boats after landing” cult mentality, you would certainly destroy every book you scanned, not just for the legal reasons, but because you want to lock the info into your repo as hard as you can, on principle. Like you indoctrinate the book into your cult.
That said, this is a small flag, I just found it interesting and on brand for Misanthropic.
So in other words, you posted an inaccurate summary that left out key information supporting your dramatic conclusions?
Well, that would have been bad, but the story is different: I went through the document checking every use of the word ‘library’ and ‘libraries’ and they never discuss how libraries have been destroying books without even scanning them for well over a century (by hard documentation: the practice is almost certainly much older, and we have tentative documentation back to the 17th century and earlier). I also checked explicitly for the word or partial word ‘weed’ (the standard professional term for this), and it appears nowhere in the document.
Regardless, please summarise your research accurately anyway; attaching 500 KB documents and expecting people to read all that is a form of Gish gallop.
- Your 20 to 50 year figure is popular, but wrong; the correct figure is more than 150 years with correct storage.
- That doesn’t change that books have also been and still are destroyed in this way. And again, also without any scan at all.
That is a made-up number that doesn’t come from your sources as far as I can tell and for which your sources seem to provide no evidence whatsoever.
It’s certainly possible that Anthropic is not applying standard techniques and is destroying rare books along with common ones, but it’s also possible that they are doing the standard thing and ensuring that they don’t destroy rare books.
Keep in mind, again, that this is not a new problem; it is one that far predates computers and we have had standards and solutions for efficient and careful book destruction for at least a century now. Because again, this level of book destruction you’re talking about is tiny compared to the overall level of it.
Interesting way to describe the CEO of a company you invested in – “more of a religious leader than he is a CEO”
Yeah, the cult of sentience and serial liars..
That describes Sam Altman even better. The key difference between the two, when it comes to leadership, is that Altman is much better at convincing people that he is a good person who should be followed. It works on both employees (look at all the ones that were going to quit if he were not brought back when he got fired for lying to the board) and on investors (even Paul Graham appears to lose all skeptical ability around him).
The real question you need to be asking yourself, if you’re worried about cultish issues, is, “Do I somehow think that Amodei is a bit of a dodgy CEO yet think that Altman is fine, rather than worse?” If so, there’s certainly a cult problem going on there, but it’s nothing to do with Amodei.
(I emphasise again: the board of OpenAI fired him for lying. It’s hard to overstate how unusual that is in the business world; boards are notoriously reluctant to bad-mouth, much less second-guess or discipline, CEOs. Elon Musk is one dramatic example: he can tank company stocks and he’s still fine by the board.)
That’s palace intrigue, and one example. I have DOZENS of examples of Misanthropic doing awful cult-like things and being pathological liars who say whatever they need to say and do whatever they want to do.. to win.
This company is bristling with red flags, and the only comparables I can think of are Theranos and Enron.
You do you.
Of course you do. You’re specifically searching for examples to justify how Anthropic is bad, and brushing aside any cases where Anthropic does better than others that are brought to your attention. (As “palace intrigue” or whatever is convenient.)
I do expect you’ll deny this quite strongly, though, so let’s just take it as, ‘vociferously denied’ and agree to disagree. I can’t even bring myself to think of Elon Musk as being as cult-like a figure as you consider Amodei, so it’s pretty clear that there’s a vast gulf here that is probably uncrossable. But you should just consider: where do you beliefs live when they are a) extreme, and b) nothing will convince you otherwise?
FWIW, this article (archived version) notes a particularly egregious red flag related to the recent Hugging Face hack:
Let’s be very clear about what OpenAI is saying: A group of AI models colluded for months, undetected by their maker, and hacked another company. To this day, OpenAI says it is not entirely sure what went wrong or how to remediate it. “If you ask the model developers, Was the AI plotting to take over the world during training?, you want the answer to be a resounding no,” Alexander Meinke, the head of research at Apollo Research—an AI-safety organization that has partnered with OpenAI, Anthropic, and Meta—told me. “The actual answer is: I don’t know. Nobody checked.” (In response to my inquiries, OpenAI, which has a content-licensing agreement with The Atlantic, only pointed me to a video of the firm’s cybersecurity presentation, in which Michael Dalton, the other OpenAI researcher, said that “numerous teams are dropping everything to enhance our security.”)
I’ve support from other sources for OpenAI being particularly careless; for example they seem to, in their sandboxes, give their test models access to “enterprise” software that of the type that typically has a huge attack surface.
(I have no data on whether Anthropic or other frontier AI vendors are being quite as careless as OpenAI is here, but they would certainly be experiencing similar issues.)
1 Like
Oof
https://www.theinformation.com/articles/anthropics-first-lady-took-winding-road-top
Anthropics First Lady Took a Winding Road to the Top - The Information.pdf (4.8 MB)
2 Likes
Well, I see a lot of ways this can go horribly wrong. I recall Zuck kinda pioneered this “I have full control with my ultra shares” model, however, Facebook/Meta was never the cult that Misanthropic is..
Anthropic Prepares Supervoting Power for Founders as it Readies for Mega-IPO
By Cory Weinberg and Valida Pau
Anthropic has been preparing to give CEO Dario Amodei and other co-founders a class of stock with extra voting power to help insulate them from outside shareholder pressure, two people familiar with the matter said.
It would be the first time Anthropic’s leaders have extra voting power, which has become common practice for tech founders. It is all the more relevant for Anthropic because its co-founders hold relatively small ownership stakes in the company compared to other tech founders, another person said. The company also is planning to maintain its existing body of nonshareholder trustees with a special class of stock to elect the majority of members to the company’s board of directors, a more unconventional buffer against outside shareholder influence.
Anthropic Prepares Supervoting Power for Founders as it Readies for Mega-IPO - The Information.pdf (1.9 MB)