AI destroying economic system

[brucehoult]

my first bike with a fully electronic throttle and no... I don't like.  The 0...1 point is like a razor.  There is no play.  So that "getting back on the power" gently just at that "tuck point" is probably delayed by caution not to get more than I wanted in throttle lurch.  It's fine across the rest of it's "curve", but the "Crack the throttle open" is brutal and digital.

I haven't ridden one of those. My sister and her husband had one of some kind of NC750 each. The frunk is funny.

But I don't know why it would be worse than anything else with fuel injection with complete fuel cutoff at closed throttle and more than a certain RPM, which in my case goes back to my 1986 K100RT which I bought in early 1988 (ex police bike) and then owned for nine years.

In those low speed manoeuvres you should be keeping absolutely constant throttle and controlling speed with the rear brake.

[paulca]

In those low speed manoeuvres you should be keeping absolutely constant throttle and controlling speed with the rear brake.

Yes, it's the unexpected transition that gets me.  I suppose I could just slow a little more before I turn in and then go through the corner on nuetral throttle.

The frunk is epic.  I can go to the shop and get a full large sized shopping bag in there.  10 tins of beer.  Many bottles of wine.  A helmet if you must.  I also have a 50L top box for the back.

The Nc750 isn't a bad bike.  I bought it for commuting, just after the pandemic.  Never did go back to the office full time and now I wish I'd bought something else.  I miss the high-reving straight 4, litre bike.

[LenutaOvidia]

Apparently the human brain uses about 20W.

My SpacemiT K3 board uses 14W when running gemma-4-26B-A4B-it-QAT-Q4_0.gguf and can solve many kinds of problems and write or review or bugfind in code faster than I can.

How many gigawatts did it take to train Gemma 4?

Also, how many terawatt-hours would it take to learn to ride a bicycle, something a typical 5-year-old learns in under an hour?

Idk but a Chinese company called Agibot already pulled it off in March 2025: https://www.youtube.com/shorts/e_rzxz0sZ8o . They got their lingxi robot to ride a bike in an underground parking lot for some time.

[booscrawl]

Apparently the human brain uses about 20W.

My SpacemiT K3 board uses 14W when running gemma-4-26B-A4B-it-QAT-Q4_0.gguf and can solve many kinds of problems and write or review or bugfind in code faster than I can.

How many gigawatts did it take to train Gemma 4?

Also, how many terawatt-hours would it take to learn to ride a bicycle, something a typical 5-year-old learns in under an hour?

Idk but a Chinese company called Agibot already pulled it off in March 2025: https://www.youtube.com/shorts/e_rzxz0sZ8o . They got their lingxi robot to ride a bike in an underground parking lot for some time.

In a straight line for a few seconds. Okay.

[paulca]

I'm seeing an increasing number of viral videos showing the process of AI 'feeder' workshops using a break press to remove the spine from books, so the pages can be loaded into a scanner and OCRed for AI training.

Visually this looks like "Burning books!".  But they only need to scan one copy right?  They aren't burning all copies, they are sourcing one copy and scanning it then destroying it.

[Analog Kid]

Visually this looks like "Burning books!".  But they only need to scan one copy right?  They aren't burning all copies, they are sourcing one copy and scanning it then destroying it.

Unless that one copy happens to be the only copy ...

Sounds like a pretty ham-fisted way to treat your source material. A museum conservator would be pretty horrified at this behavior.

[PlainName]

But they only need to scan one copy right?

Are they keeping the copies verbatim, or disposing of them after whatever has been extracted? And if they are keeping them, does anyone else know or would be able to access those copies?

[Siwastaja]

A museum conservator would be pretty horrified at this behavior.

No, a museum conservator understands perfectly well that many books have been printed in tens of thousands of copies. This is a distribution medium, not archived material. Similar thing with traditional newspapers - print, read, dispose of.

[golden_labels]

Contents of those books are generally⁽¹⁾ not permanently lost, given many copies exist in archives dedicated exactly for the purpose of preservation. But this is not the point either.

It’s not Harry Potter series, still circulating in millions and likely to be republished 2–3 times in your lifetime, that is being destroyed. It’s not acquisition from common bookstores. The companies buy from antiquaries! They both order and then destroy indiscriminately. We may find all kinds of books in antiquaries, but in that business easily available titles bring no profit. Which means they take out of circulation books that are already very hard to get, even harder in a decent shape. One may wait years to find a single copy and then pray it’s not damaged or a reject.⁽²⁾ It does not matter some national library has a copy. That copy is for researchers, not for mere mortals, and you can never own it. For most of us that book is effectively gone. All this made worse by some items in antiquaries are internal publications that never went to any archives. It’s a sheer luck you find them in dumpsters, at antiquaries, among street vendors.

The initial wave of reports mention neither the scale nor the style of such acquisitions. It may feel like: oh, some company bought a bunch of books from a few stores — so what? To give some picture, just a single store in Poland received an order for 800,000 items. Eight. Hundred. Thousand. Let it sink in. LLM companies do not share the books between themselves, so each one repeats the same procedure, many, many times. It is not normal, healthy economic activity. This is a frantic pillage, a race over who depletes the resources first. No different to what they do to this very forum and a lot of other sites right now. This is not the end of it. Acquisitions are not made by the LLM companies themselves. Instead dummy companies are opened, pretending to e.g. be buying books for charitable purposes like donating to local libraries.


On a separate note, if you believe singular archived copies are protecting titles from destruction, let me remind you something. We already irreversibly lost majority of early cinema, mostly in archive fires. Europe also lost a huge part of its archival documents in numerous wars. The same happened to archeological artifacts, in particular during WW2.


What may be the reason behind the destruction? The most likely explanation is that the destruction is cheaper and Ockhamially I’ll stay with this one. But there is another advantage. Lower exposure to intellectual property infringement claims. If they used the book for training and then put it back in circulation, it’s trivial to sue them for making an unlicensed copy. It requires more legal acrobatics if the paper version is destroyed.

⁽¹⁾ But see the end of the following paragraph.
⁽²⁾ This already happened to me once. A “perfect condition” book I imported from UK. No wonder it was never read: it was a misprinted copy.

[Marco]

They probably think destroying the original book plays better in court too.

[Analog Kid]

A museum conservator would be pretty horrified at this behavior.

No, a museum conservator understands perfectly well that many books have been printed in tens of thousands of copies. This is a distribution medium, not archived material. Similar thing with traditional newspapers - print, read, dispose of.

Read @golden_labels' post, just below yours in case you missed it, for a perfect refutation of what you wrote.

[brucehoult]

They probably think destroying the original book plays better in court too.

They are doing it PURELY for legal copyright reasons. They can perfectly well scan the books while preserving them if legally allowed to.

[Marco]

Copyright doesn't have a concept of format shifting. The millions of temporary copies before and during training all need a license regardless. They are doing it to hopefully edge closer to a fair use ruling once it gets to the US supreme court. It by no means is an open and shut case because they cut them up.

PS. the lower court judge was legislating from the bench and being silly, but that's par for the course.

[Siwastaja]

A museum conservator would be pretty horrified at this behavior.

No, a museum conservator understands perfectly well that many books have been printed in tens of thousands of copies. This is a distribution medium, not archived material. Similar thing with traditional newspapers - print, read, dispose of.

Read @golden_labels' post, just below yours in case you missed it, for a perfect refutation of what you wrote.

It's a good reply, but not a "perfect refutation" - that would have sources. This had none - just pure speculation.

Though, it could be still true, I don't deny it! Just like the attacks on this forum as mentioned are true. However, the attacks did not come from AI firms most of us know, they come from whatever Chinese, Indian, Russian origins, and from companies/individuals no one have heard about, and no one can trace.

No one can say if it's some kind of legit AI research, or just modern-day hybrid warfare against "our" western values, like "do not burn books". It could be both, and clearly it is working: internet warriors are first to blame OpenAI or Google or X AI or Anthropic about things others do.

What strikes me as odd: while the mention of LLM companies "not sharing" is true, there are only like <10-20 serious LLM companies teaching their models "everything".

There might be an explanation for this - explosive phase of AI craze before the bubble bursts - every startup with no serious capability of developing serious AI (which is extremely expensive) gets some millions of funding from idiot investors, and do poorly planned, poorly executed actions, like bombing EEVBlog forum with some runaway script, or ordering 800 000 books from Poland to be destroyed. It would be a gold rush, where almost no one survives, and devastation is left behind.

I don't deny it, but I would like to see a real discussion about it - does it (destruction of relatively rare pieces) happen, and if so, who does it, how and why?

[paulca]

Interesting Forbes article, contains a quota of evidence.
https://www.forbes.com/sites/maryroeloffs/2026/08/17/ai-companies-are-buying-and-destroying-antique-books-heres-why/

[PlainName]

Though, it could be still true, I don't deny it! Just like the attacks on this forum as mentioned are true. However, the attacks did not come from AI firms most of us know, they come from whatever Chinese, Indian, Russian origins, and from companies/individuals no one have heard about, and no one can trace.

I thought the Apple bot was in the lead on the graphs, and disabling that one had a great effect. Of course, hardly anyone knows of Apple

https://www.eevblog.com/forum/news/getting-a-lot-of-502-bad-gateway-errors/msg6361500/#msg6361500

[Siwastaja]

I thought the Apple bot was in the lead on the graphs, and disabling that one had a great effect. Of course, hardly anyone knows of Apple

Maybe you should have read what was said in the thread... 

[PlainName]

Maybe you should read the message that quote links to? 

Edit:
For clarity, we don't know the forum is besieged by AI bots (because they don't typically advertise themselves as such). Could be an ex-forum member with a grudge doing a DDOS or something. So your statement that this forum is brought to its knees by the book scanning crowd is just supposition. Further, you said that it is the unknown bots bringing the forum down.

So, I posted a link that shows we know the known Apple bot had an effect, which shows that your supposition was not fact in that case.

[Siwastaja]

Maybe you should read the message that quote links to? 

Read https://www.eevblog.com/forum/news/getting-a-lot-of-502-bad-gateway-errors/msg6359000/#msg6359000

(and the whole thread, as I suggested, really, to get some context understanding)

[PlainName]

Read it all, no doubt as soon as you did. My comment stands in the context, and given the link I included.

 

[Siwastaja]

OK, I hoped you would find answers, but apparently not, so let me help:

> I thought the Apple bot was in the lead on the graphs
Yes, because the attackers didn't show in the graphs, so graphs were irrelevant. Read and understand the thread.

> and disabling that one had a great effect.
Earlier it didn't have any effect. Later it "did". Again later it didn't again. Most likely explanation is random variation in the actual attack, during that adjustment. Scroll a few messages forward, the problems came back with Apple still disabled. Read and understand the thread.

HTH.

[PlainName]

Let me help your comprehension:

I thought the Apple bot was in the lead on the graphs, and disabling that one had a great effect. Of course, hardly anyone knows of Apple

See that smiley thing on the end? Hint: it's a clue.
Never mind, as usual the thing just escalates.

However, the fact is you don't know what is attacking the forum, only that the forum gets a ton of traffic. Even if you could show they are AI bots (which they probably are) you don't know if they are bots from the same source that is destroying books. All of those are your supposition only.

The only thing about this that we do know is that Dave turned off Apple and it had an effect, or seemed to, that particular time. Which I highlighted as a tongue-in-cheek thing. I apologise for not realising you are devoid of any humour, even given the obvious hint (see above).

[golden_labels]

It's a good reply, but not a "perfect refutation" - that would have sources. This had none - just pure speculation.

This is because one doesn’t include source references in normal speech or writing? Sparsely they appear as the means for the reader to get wider insight into a side topic (example: this post of mine linking Wikipedia.), without cluttering the main statement, or to explain to the reader what may otherwise be obviously obscure to them (example: this post, linking to a missile incident almost nobody outside Poland would know about, but is relevant to explaining my line of thought.), or as an example (examples: parentheses in this sentence ), or as attribution (example: Google N-gram data attribution here). Abundant use of unsolicited references is rather a sign of manipulative tactic in persuasive writing. There are very few styles of communication where source are expected and then their use is following particular rules too (written or not). I therefore seen no reason to make that effort. Nobody else in this forum does it either. The pointless and absurd length of this very paragraph shows why.

If sources are needed, one may always ask for them. So can you. Unreasonable requests may of course be declined, but otherwise I see no problem. Just point to the fragment and ask. Done.

I myself see only one fragment in my post that, in my eyes, may need sourcing. It’s the middle, big paragraph mentioning the numbers and methods, from “one Polish company.” Before you make a request for that source, I implore you: please be sure you actually need it. The reason is simple: the effort you request from me. I’ll have to read through hundreds pages of paper magazines of the past months. If this is of value to you, I am happy to do it. But I’d prefer to not do it for no reason.

[Siwastaja]

It's a good reply, but not a "perfect refutation" - that would have sources. This had none - just pure speculation.

This is because one doesn’t include source references in normal speech or writing?

I know - I didn't mean that as critique against you, I rarely give any sources myself either. I just disputed the idea that it's "perfect refutation". It isn't that. It's usual chat - speculation - thought-provoking stuff, non-guaranteed correctness - that's all fine. To Analog Kid, it was more than that, because it happened to align with his feelings.

To me, the idea that 5-10 AI giants would be able to do significant damage by finding books that exist in maybe just ~10-20 of copies seems far-fetched by simple logic. The baseline is, after all, that anyone can buy those same copies and put them in their own shelf, where no one can ever access it again, anyway. Anthropic does not need 100 copies of the same book. It needs 1. So the damage, per book, is the same as any random individual buying the same single book. The fact that providers don't share is not many-orders-of-magnitude thing, it's a single order of magnitude thing because there can be roughly ~10 AI giants capable of creating large enough training they need "every book". IOW - if we have 1000 AI companies building large models, we have much bigger questions about energy and resource use.

So that part of the (implied, and not by you) story does not make sense. What would make the pieces click? Gold rush would explain what you are seeing - not tens, not even hundreds, but thousands of relatively small companies, backed by some millions of investor money, working in parallel, everyone trying to do poorly planned actions, when in reality they never can produce a working, large AI model. They could total to emptying supply of semi-rare books, from which some hundreds hard-to-find copies exist.

If that theory is correct, I have no solution nor suggestion to it. Forbidding others than Anthropic and OpenAI and Google from developing AI obviously isn't an option, even though those vendors would very much like it.

[PlainName]

The baseline is, after all, that anyone can buy those same copies and put them in their own shelf, where no one can ever access it again, anyway.

Kind of. The physical copy will exist wherever it is, and may well later see the light - the owner (or heirs) might donate, sell, auction, gift it on. The book may be hidden for decades or even eons, and then be discovered.