Hello, everybody! Ah, dear friends. In one of the previous episodes, we have addressed the subject of licenses, open source, types of open source. And, uh, the commentaries have been making some reports, where you asked for a little more detailed on the subject. Tanya and I are gonna try to do this in our short, little special edition. But again, given that the audience is in our, uh, totally different levels of training, we'll try to expose this subject from the very beginning and show examples, uh, existing models. What are the licenses they can read? Well, Tanya, let's get started.
Well, if you're in general, what's the software license? , this is a legal agreement between the two parties between the developer and the user, which determines the conditions, the software can use the code.
So it protects rights and regulates responsibility both on the one hand and on the other., that's, she also protects intellectual property, that's, protects the developer. It makes it clear that it is possible to do what is not, to regulate distribution and copying. In some cases, er, it is permitted, in some cases, to be prohibited. We'll look at this further in detail. And as I said, uh, reincorporates the types of responsibility.
And if it's true, uh, that big picture in general terms, the types of licences are: pro-cable, free, open source, freemium, shareware, public domain and commercial.
Ah, they're different, depending on the restrictions. I mean, for example, license, she has the most severe restrictions. The code can't be copied, not changed, and so on.
This is all of us known, there, Windows, Mac, the system, the programmes of the Photoshop type, et cetera, Adobe., there are free GPL licenses that are free, changes remain free. It's like WordPress, again, all known, common sites.
Open source is personal to me as a man of a little far away from all this, it was hard, to draw a distinction in my mind between a free and open source. So open source is linked to a lot of interesting casees because they position it as an open code that can be used for commercial purposes, but with conditions. And these conditions are the most interesting, since even large companies were in violation of these conditions. And I'd like to see Illnar share, um, these briefcases are interesting, which are connected to the open source of licenses.
Yeah, here, uh, you know, it's really hard to just put it all together in one scale, so it doesn't cross. Ah, Linux is also open source, yes, but with some degree, yes. It's important here, though, that this is a GPL license. Mm-hmm, and, uh, here you have a very good written product, and all changes must be free, too. So you can't take Linux's core, change it a little, yeah, there's something to finish, redo and start selling it. I mean, if you've taken the foundation, the product that is under GPL license, the changes you make, they're also under this license.
This is an example, Wikipedia. I was actually a little surprised when I've been in touch with this fact a while, but he-- Vickipedia is under Creative Commons' license, and, uh, you can't do derivatives based on Wikipedia. under another license, yes. I mean, more restrictions on putting it. So we can't take Vickipedia, create a new product on its basis and make it pay-as-you. If it is under Creative Commons, then all further derivatives must also remain under this licence.
And I wonder, yes, Ilnar, I'm sorry to interrupt, but you were just telling me when I didn't really understand their motivation. Say it again, please. I wonder why they did that. Your version.
Yeah, it's just a hypothesis, yeah. But, uh, uh, Vickipedia creates a huge number of people. However, as anyone could become, it could gradually become an editor, start amending articles and so on. So it's a huge collective effort, you know, a real open source, yeah, when a lot of people are working on product creation. And there's just hypothesis that, accordingly, the work of this number of people who created, uh, this product, it should, in fact, remain accessible to all people. So you can't just take this job, you can't pay anyone, you know, roughly, and start getting money from it. It's like an extended effect from, uh-a, this license. But what is interesting is that, uh-a, models, for example, ChatGPT or other models, they're clearly on the Vickipedia texts, and they're all trained and paid. So here in any case, there are enough thin things. Legal questions you're talking about are not always clear. It is not always possible to prove that , material has been directly used and, accordingly, it should be under this licence.
At this point, it seems that this is something that is so unrealistic, yes, some license, something needs to be found somewhere, possibly a legally complex document. Let's show you, uh-a-a-a-a-a, like, let's start with DeepSeek. We've all heard and, uh, we've probably used DeepSeek-R1 and everyone's probably heard he's open-source. So what does that mean, open-sourceness? In this case, the product itself is licensed by MIT, yes, in the repository itself, where the product code is located, everything is in the model. Ah, there's a license, MIT-shone license. It's the most open, uh, product. And so very short, literally, there's a 20-line on one screen, and twenty-one lines are not very wide, and all the information on this license is posted. Uh, I can read it, I can translate it. What do you mean here? You can take this product, you can do whatever you want with it. You can reproduce him, you can then make a payload on it. There's no restriction. The only thing that starts 15 is, uh-a-a, dehydration of the company that created this product, yeah. That's what, the product is transmitted as it is, without any guarantee. And if you use this product, something broke down, you don't come to DeepSeek for damages. Yeah, that's, like, some kind of company's own protection. But in general, this is the whole license. I mean, if we're saying we need to read some document, that's the twenty-one line of the text, that's all that's in DeepSeek-R1.
They were able to place such limits in the twenty-one line .
And here, it's important to understand that, uh, not always easy., for example, if we take another product of DeepSeek, DeepSeek V3, which you may have heard, too, but it's a little less known than R1, because R1 was free, and he hit almost all the baccalaureate in his time., V3 is also a very good version, but it is therefore important to understand: there is a separate licence for the code that is being distributed and a separate licence for the model. So, the code license is nothing different. Same twenty-one line, same MIT license. Take it, use it. There's no restriction. You can even make commercial products on the basis of it.
But there's a license for the model itself, respectively. And if we're talking about a model license, there are some limitations that are inside. I'm not reading everything. I'll stop right there on the last block. Yes, app. Annex A is called here. There is, uh-a-a-a, that text that you agree with some limitations when you use this model. Ah, they're a lot, they're curious. Translate into Russian if English is not very comfortable with reading. Well, for example, I'd like to point out that for, uh, unwanted-- well, not desirable. You agree not to use this model for a fully automatic, uh, decision making, uh, you know, decision-making systems, and that have an impact on human rights. Uh, let's try to give us an example. So you can't use this model to make a full automatic decision, for example, about hiring. I mean, er, human rights must not be violated by decision-making, full-on-based, automatic decision-making., interesting here is, of course, the question: what are these restrictions imposed for? This may be linked to the entry into the international market, European legislation that we often talk about--
Yeah.
...yes, or some other restrictions. But somehow, the V3 model has a different license. In fact, a very short document, too, can be read in its entirety. This is, uh, version 1.0 of 23 October of the twenty-third year, but even the most modern models under this licence are coming out.
License's coming out. That's why if you ever make your products based on, uh, DeepSeek, you'll have to read your license. That's the first. And the second, in fact, point out that different licenses can be different under different parts because under R1 one license, and under V3 the licenses are already on a separate code, separate to the model.
Aaah, I also want to highlight this characteristic example, Llama, Llama 2 Community License, yes. And here, aaaa, but, among other things, what is there, I want to highlight the following. Here's paragraph two: additional commercial conditions, additional commercial restrictions that are in place.
Mm-hmm.
Aaah, as soon as you have, uh, seven-de-seven hundred million monthly active users, yes, so you have more than seven hundred million monthly active users, Accordingly, you cannot use this product without the approval of the model developer. And it's obvious, uh, dehydration of itself from competitors. Because Sasha Masrabov said in one of the episodes that Snapchat would be happy to use Llama in the time, but there is no possibility because active users are already very much, and therefore, We can't use this model. I mean...
Mm-hmm.
Yeah.
Besides, I'd like to...
Thank you for that.
Yeah. Yes, yes, yes. I mean, when you just hear about the license, I have at least, yeah, I think it's a big, complicated document that doesn't--
Page 50, as usual.
Yes, in which, without a magisterial degree, law-enforcement
Mm-hmm.
In some other way, aaaaa, there's no English law that you'll never get through.
Yeah.
And it's just curious to see what the company's conditions are. I'd like to be here the last thing to turn, uh, your attention, Tanya, and our viewers, which, for example, is, uh, in the first paragraph, the fifth paragraph, subparagraph that you won't use Llama's material, or Any exits that are made from the materials from Llama are produced. So any material she's generity will not be used to improve other models. That's, uh, again, and Sasha Mashrabov talks a lot about it, and at the launch, we said that once a good model emerged, we can get a lot of data from her, and then we can get a lot of data on it. I'm gonna teach you my model. And it's much cheaper than starting from scratch. So there's a rumor, aaaa, that DeepSeek is trained on the OpenAI and you think that-
Yeah, so it's not gonna be traceable, right?
Yeah.
They just banned it. How can you track it?
We've left a lever like this, if we push, we come, we'll get it all figured out.
Mm-hmm.
The lawyers in this company are clearly very strong, but, you see, they leave these levers in their own license.
Yeah.
We'll leave all these resources under this video. If you're not using, there's DeepSeek, not Llama, and Qwen or anyone, it's also very easy to find through the search for a license for the model. If you want to create some food, you'll have to spend at least ten--
Yeah.
Fifteen minutes to study these
materials.
Well, I understand correctly, Elnar, that you were talking about this MIT license, too. Where's that name from? What you said, it's the Massachusetts Institute of Technology. And the idea of all these licenses is in the first place as a good thing for mankind, right? I mean, just so that ordinary people use things, move on like that. But they still have to be dehydrated in documents so that if anything, some trial or something, and, seriously, commercial use, is no longer in the way of their own. If it's all that way.
Yeah, I have a clear parallel with an
academic environment, yeah.
Mm-hmm.
That is when we say, for example, MIT license, one of the main restrictions that it has, it's a copyright, to say. I mean, if you use these materials and do something on its basis, you have to keep, uh, information that you use this stuff there. And it's very similar to, uh, writing scientific articles, yeah. I got a...
Mm-hmm.
There's a number of publications in math. If I take some theorem, I'll take some kind of approval--
Yeah.
That's proved by another man.
I'll mention the source. Mm-hmm.
Yeah. I can use them completely. The only thing I need to do is just refer to the material I use on the author--
Mm-hmm.
Mm-hmm.
There are different, uh, licenses, yes. So, you and I were talking about what Apache is, Apache, two zeros are calling it now...
Yeah, yeah, yeah.
Second version. It's also just from a company that makes products, it's a license
that's pretty popular for it.
Mm-hmm.
We must have literally, Tanya, yes, there's a few minutes. We could, uh, say a few words about open source. Yeah, here's--
Yeah.
Many might have a question that products were being created free of charge. Like, like, some kind of good, big products.
Yeah, yeah, yeah, that's an interesting moment. Mm-hmm.
And here I want to give you an example of a company that created Linux, and now Linus Torvalds is in time--
What are they making?
Linux core, yes. And it's like a huge number of people, uh, wasting time, and trying to make the product work. There must be three main elements in here. First, if the product is a product that is worth it, then people are all investing in this story. That's professional points for what you are, uh, speaking of a English-language meinthene, I don't know how Russian is.
Mm-hmm.
You're involved in the creation of this product, aren't you?
You're supporting him, yes.
You're supporting him, yes. Besides, you work with a very, uh, good, serious, most often the technology. It's actually helpful, but one point is useful for professional growth. Another story where the money comes from? If the company uses this product, for example, there's a library in the software or something, and it needs to, uh, make some changes, to the point of fact, it's in the first place, may simply charge money to an organization that develops this product and prioritises changes that would be important to it. On the one hand, the company gets the money, and she'd still like to do it. She'll do it, just do it faster, and she'll, uh, hire the staff who could do it. For a company that uses these, uh, products, it produces a product that better meets its requirements.
But it's in the form of donations, right? Right? Mm-hmm.
Most often in this format, it happens.
Yeah, yeah.
And the third paragraph that would like to be distinguished is, respectively, aa, the infrastructure that arises around. If we take Linux, it's an operating system that uses a very, very large number of places. And for large companies, the lack of sustainability is important. It's important that all, uh, mistakes, the bahts that arise, they're quickly exhausted. And, accordingly, it's important that if something happened there, I broke up, I took my troubles to some company, and she decided. Accordingly, the, uh, bonding of additional services for some open source of product, it also makes a lot of money. For example, Red Hat, which provides Linux distributives to many companies, is exactly what it does. I mean, the distributor himself, he's free, yes, so Linux you can supply free, but the whole set of services they're providing on top of it, he's bringing the company big money, respectively. Well, I guess if you're talking about open source and earnings, you could have had three of those things.
Mm-hmm.
Oh, if you're interested in this subject, you can just get deeper into it.
Or write in the comments. We're gonna try and answer them, too.
Yeah. Well, it might seem so much, you know, obviously, you showed such short documents where they all sorted out. And the sl-- our listeners might seem to think that everything just read 20 lines and how good it is. But actually, even big companies have been caught in a violation of such, uh, licenses. And there are a few examples. They're not really fresh, but they're so very indicative. So, there was a suit in America Adobe v. Forever 21, this is some brand of youth clothing, two thousand and seventeen years, which used several Adobe products for their promotional campaigns, you know, there, Photoshop, Illustrator, but, uh, most likely pirate versions, that is, no commercial license. And, of course, they won the court. And now all companies are afraid of using fire like that, you know, pirate versions.
And the second case, which is more interesting, is exactly Linux, and it's about
Linux, and he lasted 13 years, there were litigations. And he's just showing that we've been talking about open source so much, that the connotations of these paragraphs of the treaty depend on the skill of a better lawyer. Here's the company, the developer's company's lawyer will be able to convince the judge to correct his arguments or the user. And for 13 years, SCO Group versus IBM. So SCO claimed that IBM illegally included its Unix code in Linux. And this case has become one of the most important things in the open source software licenses to date. It is a multi-year litigation that has led the court toward the IBM side. But the root thing has changed the approach to licensing open-speed software. So take care of yourself, keep your company safe, keep your money and hire good lawyers.
Put the liki down, sign up on the canal, stay with us, press the bell, write the comments. Have a good day!
Thank you.
I hope you enjoyed it.
We're waiting for comments and questions.