Nick

You may have heard about the incredible way in which artificial intelligences like ChatGPT manage to learn the contents of the most popular books. Here's how the Internet giants managed to do it – and it's not pretty!

How AI knows about the content of a book?

First, here's a little essay in which I quizzed ChatGPT and its competitor Gemini by asking a question about the 3rd book in the Harry Potter saga (warning spoilers):

Image: a question about Harry Potter on Gemini (from Google)

Image: a question about Harry Potter on chatGPT (from openAI)

Then I asked the AIs how they knew that, and here's their answer:

> Gemini: I haven't personally read the Harry Potter books. […] In this case, I've likely accessed and processed information from sources like:[…] Digital copies of the books themselves

> ChatGPT: I've been trained on a diverse range of text, including summaries and analyses of popular books, which allows me to provide information about their content.

While Gemini clearly states that it had access to the text of the book (or not – it is not clear), ChatGPT is more evasive, stating that it only had access to publicly available data…

However, given the details known about many books, many people have speculated that artificial intelligences have used books to learn all this, which raises many questions.

How did they get the books?

Obtaining classic literature is a relatively straightforward task. Many of these books are in the public domain, and there are numerous online resources (some of which are listed on this page).

But to get hold of a large number of well-known books and bestsellers, you need to get permission from the authors and publishers.

The problem is that this would take too much time, and would probably be rejected by many authors and organizations.

Another solution would be to buy the books one by one in paper or digital format and use them. But it would take a lot of time to buy hundreds of thousands of books.

So we had to find another way to enable artificial intelligences to read all these texts.

A quick way would be to "find" a ready-made collection of well-known books!

This is where the famous collection of ebooks known as "books3" comes in.

What is books3 and why does it change everything?

It's now accepted that books freely available on the Internet have been used to train and design artificial intelligences such as chatGPT.

These books include 2 collections called books1 (also known as bookcorpus) and books2.

These two archives contain thousands of books that raise no particular copyright issues. These are ebooks that are officially available free of charge on the Internet, either because they are in the public domain, or because their authors have decided to publish them free of charge on the web.

But in October 2020, Shawn Presser published the books3 collection, which contains almost 200,000 books, in the form of ebooks in TXT format, with a total weight of 37 GB.

This collection comes from an ebook piracy website known as bibliotik.

The complete collection is now available as a book download from The-Eye website. This site is designed by the EleutherAI organization, which provides content for the creation of new artificial intelligences to compete with Internet giants such as OpenAI, Google and Microsoft.

The Eye also offers a very large archive of AI data called "The Pile", weighing in at hundreds of gigabytes!

Today, the books3 collection is no longer freely available on the Internet, but there is every reason to believe that it is still widely used and distributed under the radar, far from the eyes of the general public.

Who used books3 and what were the consequences?

The first company to acknowledge the use of books3 was Meta (parent company of Facebook), which admitted in a Californian court that it had used this collection of books for its Llama artificial intelligence model.

As my tests show, ChatGPT (for GPT3 and especially GPT4) and Gemini (Google) are also suspected of having made extensive use of books3 to create their conversational chat. But there's no proof of this, and it's very difficult to know for sure.

Today, the publishing houses have obtained the removal of books3 from the Internet, but it's fair to say that the damage has been done (source: https://www.reuters.com/technology/meta-used-copyrighted-books-ai-training-despite-its-own-lawyers-warnings-authors-2023-12-12/).

Because of this, it seems difficult to create a relevant competitor to ChatGPT, LLama or Gemini without resorting to a large collection of published books.

It would therefore make sense for authors and publishers to license their works, or else legislate to prevent AIs from using copyrighted works.

Maybe you downloaded a giant fantasy saga or a comprehensive textbook. While these tomes offer hours of reading material, navigating them on your e-reader can be cumbersome. In this tutorial, we will see how to split an EPUB file into multiple chunks.

(more…)

The figures for the US book market have just come in, and one piece of news is sure to delight digital book lovers: ebook sales topped the billion-dollar mark in 2023.

This is an impressive figure: in the USA, in 2023, digital books of the ebook (text) type brought in over a billion dollars (+0.6%)!

This figure covers sales of digital books for e-readers, of course, but also sales of ebooks to be read on computers (PDF documents), smartphones and tablets (ebooks from the apps from Google and Apple).

While printed books are holding up remarkably well, with estimated sales of almost $9 billion by 2023, the same cannot be said for physical audiobooks. Unsurprisingly, sales of physical audiobooks (in CD form) are down to a paltry $12.9 million.

Now, if we add up all these figures, we see that the book market is slightly down by 0.3% over the year 2023.

The market is therefore holding steady, with digital sales continuing to increase year on year.

These figures come from the AAP (Association of American Publishers). Source : https://publishingperspectives.com/2024/03/aap-statshot-in-2023-us-revenues-were-12-6-billion/

Ever downloaded multiple ebooks of the same story, but wished you had them all in one file? Or maybe you have short stories you’d love to combine into a neat anthology.

(more…)

E-readers have revolutionized the way we read books, and millions of us around the world use one every day. But are you getting the most out of your e-reading experience?

This guide will show you how to change the settings on your e-reader to make reading more comfortable and enjoyable.

> TLDR: Don’t be afraid to experiment with different text settings to find what works best for you. Use the built-in reading statistics to track your progress and stay motivated. Consider using an external light source.

Find the right settings for your e-reader

image: configure text settings on a Pocketbook Verse Pro e-reader

When we buy an e-reader and use it, we often make the mistake of configuring it to reproduce the reading experience of a paper book.

But an e-reader allows us to be much more flexible and creative when it comes to displaying text.

So we need to experiment to find the text settings that suit us best. Here are a few examples:

1. Change screen orientation
2. Increase font size significantly
3. Changing the font and installing new fonts on your e-reader
4. Switch the e-reader to night mode with white characters on a black background

Personally, I’ve set up my e-reader with very large fonts and plenty of space between lines, which means I can read faster – even if I have to change pages more often.

Use the built-in reading progress statistics

Image: configure reading progress on Kobo e-readers

If you want to stay motivated or read more, you can use the built-in reading progress statistics on your e-reader:

Visualization of Progress: statistics like "time left in chapter" or "percentage of book read" give you a tangible sense of accomplishment. Seeing a number tick down or a percentage climb can be very motivating, especially when tackling a long book. Setting Goals: You can use statistics to set realistic reading goals. Knowing how long a typical chapter takes you to read allows you to aim for a certain number of chapters per day or week. Maintaining Momentum: Statistics can help you stay on track. Seeing how much you’ve read in a single sitting or over a period can encourage you to keep going, especially if you’re close to a milestone like finishing a chapter or reaching a certain percentage.

I therefore strongly recommend that you configure the display of statistics on your e-reader.

Use an external light source

Image: you may feel tired when reading with the built-in light of your e-reader

Just because your e-reader has a screen with built-in lighting doesn’t mean you have to use it.

If you feel tired when reading an ebook on your e-reader, you can change the lighting settings, such as its intensity or the blue light filter.

I’d also recommend using a bedside lamp if your eyes get tired, or at least more tired than when reading a paper book.

There’s a lot of information all over the Internet about the so-called polluting nature of ebooks. Quite simply, the figures are announced at the drop of a hat. It’s time to find out how eco friendly are eBooks!

TLDR: it’s much better (eco friendly) to read an ebook than to buy and read the paper version if you already own hardware capable of reading the ebook.

What is the carbon footprint of an ebook?

There are studies on the carbon footprint of e-readers and paper books, but not ebook reading alone.

Moreover, not all sources agree, since the studies are commissioned by e-reader manufacturers (Kindle) or paper book publishers (Hachette french publisher). And I think it’s all more or less false – but you can read it here if you’re interested: https://reporterre.net/Livre-papier-ou-liseuse-electronique-qui-est-le-plus-ecolo – French link.

Of course, the media pick up on all this without paying any attention to the fine print.

So let’s start again with what an ebook is.

> An ebook is a computer file of a few hundred Kb to a few Mb in size.

> This file is transmitted by an Internet server to your e-reader when you download it.

To take into account everything that goes into reading an ebook, we also need to take into account the download when calculating the carbon footprint of an ebook.

Note that this mode of operation is simple and corresponds almost identically to the way a web page is retrieved from a web server by your Internet browser.

We can therefore rely on existing studies on the subject to calculate this.

A study that seems to me to be fairly serious and reliable arrives at an average carbon footprint of 5 g of CO² for a web page. And therefore, for an ebook too. See here: https://en.reset.org/whats-carbon-footprint-your-website/

An ebook can be read, with a carbon footprint of around 5g of CO² on a smartphone, tablet, e-reader or computer! After all, it’s just a file downloaded onto a device.

On the other hand, the carbon footprint of a e-reader is undoubtedly greater, at around 100 to 200kg of CO² (compared with less than 100kg of CO² for a smartphone).

If you already own a device that allows you to read books, the most eco-friendly thing to do is to use the equipment you have at your disposal to read your digital books (ebooks), giving priority to the most energy-efficient devices:

• Smartphone
• Tablet
• E-reader

The use of a computer is not recommended, however, as the time it takes to read your book (5 to 10 hours) can consume a lot of electricity.

As far as the carbon footprint of electricity is concerned, if the vast majority of which comes from nuclear power, it is negligible. But this is a different and more complex subject.

Yes, ebooks are green!

In conclusion, it’s much better (eco friendly) to read an ebook than to buy and read the paper version if you already own hardware capable of reading the ebook.

So if you have a smartphone, tablet or e-reader, the ebook is much greener than the paper book when you buy it.

Reading an ebook could help combat global warming.