Home > News content

Is Kaifu Lee's Zero big model shell LLaMA Plagiarism?

via:新浪科技     time:2023/11/21 13:01:02     readed:85

Author: You Yunting

Recently, Kaifu Lee invested in the company was exploded its Large Model "Yi" using the Facebook parent company META's LLaMA large model (full name "Large Language Model META AI") open source code, but did not mention. The practice of hiding without saying so has raised a lot of questions. In response, the company apologized that it is in the process of resubmitting models and code to various open source platforms and supplementing copies of the LLaMA protocol. The head of the company said there was no problem with plagiarism. Today to talk with you, legally, this behavior is not plagiarism? What are the other legal consequences?

First, the conclusion. China's "Copyright Law" stipulates: the right of authorship, that is, the right to show the identity of the author, the right to sign on the work. The so-called plagiarism is the violation of the right of authorship. Zero-one's software uses the company's source code but does not label the source, making others look like this part of the content is their own development, indeed suspected of infringement of authorship rights, that is, suspected of plagiarism. Even if they were negligent, they did not violate the right of authorship. As for the legal consequences, we have to switch perspective and look from LLaMA's copyright holder META company to know what rights they have been infringed.

 1. Rights of the copyright owner META

Although META company disclosed the code of LLaMA2, it made a number of reservations in the software agreement. Here are some points for everyone to see:

  1. To produce and distribute Llama materials or any derivative work, if available to a third party, a copy of this Agreement shall be provided to such third party. And advertise the rights logo: "LLaMA 2 Licensed under the LLaMA 2 Community License, All rights Reserved (c) META Platforms, Inc. All rights reserved."
  1. Llama may not be used to improve any other large language model other than LLaMA 2 or its derivative works.
  1. META has reserved the license to its 700 million monthly users. If the Licensee uses the software products developed by LLaMA 2 with more than 700 million monthly active users, the Licensee shall apply to META for a code license license one month prior to the license, and META shall have the right to authorize the license at its sole discretion, and the Licensee shall have no right to exercise any rights prior to the license.

Are the above terms valid? Of course it works. The software copyright of Llama 2 belongs to META, and the rights are reserved when the source code is released, as long as it does not violate the law, it is also its legal right. In addition, the author checked that the OpenAI user agreement Claude user Agreement contains a clause that does not allow its service to improve competitors' products, but the Google Bard and Xinyi user agreement does not have such a clause.

Although META has released the Source code for LLaMa 2 because of licensing restrictions, Open Source does not consider this practice open source, and they have specifically issued an article to explain that because open source licenses should not discriminate against individuals, groups, or fields of work. META's licensing of the LLaMa model and code restricts access to developers with more than 700 million monthly users and to other large language models that use LLaMa to improve, and therefore does not meet open source standards.

What legal responsibility does the company have

I have checked the user agreement of the company and it does not contain the limitations of META's LLaMA 2 software agreement, nor does it disclose that its large model is adapted from the LLaMA 2 code (which has been partially modified). Therefore, first of all, the alleged infringement of the right of authorship has been mentioned above. Second, the software agreement for LLaMA 2 is a prerequisite for licensees to publish the rights identifier: "LLaMA 2 is licensed under the LLaMA 2 Community License, All rights reserved (c) META Platforms, Inc. All rights reserved."

If there is no publicity, the licensee may be deemed to have failed to meet the license conditions, so the secondary development and release are suspected of infringement. The secondary development infringes META's software modification right, and the release of the modified source code and compiled target program infringes META's software information network dissemination right.

At the same time, if the company does not recognize the shell, it will not be a derivative of LLaMA 2 and will violate the user agreement that "Llama may not be used to improve any other large language model, except LLaMA 2 or its derivative works".

  Third, the legal risks of those who use zero-everything code to develop derivative works

If third-party developers do not understand the situation and use the code hidden by the company to develop again, they will also face the risk of infringement because of the involvement of the company. Qualitatively, all companies have legal risks that third-party developers have, but if they do not know, the nature of the infringement is not so serious.

Redevelop the infringement code of the company (including three times development, four times development...) If the code of the work contains LLaMA 2 code, it will first violate META's right of authorship, and then META's right of software modification. Releasing the modified source code and compiled object program will violate META's right of software information network dissemination.

"LLaMA" refers to a cute animal called a llama, which is also commonly known as a "grass-mud horse" by Chinese netizens. Although the Chinese commonly known as a bit vulgar, but LLaMA's software protocol although there are restrictions on the use of, but the provisions are not strict, if like the author introduced before "with GitHub open source code training artificial intelligence illegal?" As mentioned in the article, the legal risk is greater if the code under the GPL is not identified. In short, our attitude towards open source software should be more rigorous, to use other people's code, you should comply with other people's open source protocol, otherwise, not only their own infringement may be involved, but also the development of derivative products with your product.

The author: You Yunting is a senior partner and intellectual property lawyer at Shanghai Dabang Law Firm. Tel: 8621-52134900, Email: yytbest@gmail.com The opinions expressed in this commentary are solely those of the author.

(Disclaimer: This article represents the views of the author and does not represent the position of Sina.com.)

translate engine: Youdao

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments