Network Security Internet Technology Development Database Servers Mobile Phone Android Software Apple Software Computer Software News IT Information

In addition to Weibo, there is also WeChat

Please pay attention

WeChat public account

Shulou

Huawei's new patent can protect the copyright of the AI model and automatically generate watermarked reply messages.

2025-01-30 Update From: SLTechnology News&Howtos shulou NAV: SLTechnology News&Howtos > IT Information >

Share

Shulou(Shulou.com)11/24 Report--

CTOnews.com November 8 news, at present, the major Internet manufacturers are developing their own AI model products, which involves a lot of copyright disputes, and the competition is also extremely fierce.

According to App, the patent for "a language model protection method, device and computing device cluster" applied by Huawei Technology Co., Ltd. was announced on November 7.

According to the patent document, developing a system based on the large language model requires high machine and labor costs, so the large language model itself is the core asset for the company to build AI competitiveness. At present, the research has proved that through the model theft technology, the reproduction of the function of the large language model can be realized at a very low cost, which leads to the infringement of the intellectual property rights of the large language model. Therefore, it is necessary to effectively protect the large language model, avoid the attack of model theft, and realize the identification of the existing infringement.

CTOnews.com learned from the patent summary that Huawei's application methods include:

Get the request text input by the user, input the target instruction and the request text into the language model for processing, get the first reply information with watermarked words and output it, wherein, the target instruction is used to instruct the language model to add the watermark to the processing result. In the case that the category of the request text does not belong to the target category, the request text is inputted into the target language model for processing, and the second reply information is obtained and output.

In this way, when dealing with specific types of requests through the language model, the reply information with watermark can be generated automatically through the language model, which can improve the copyright protection ability of the language model without damaging the text quality of the model as much as possible.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.

Views: 0

*The comments in the above article only represent the author's personal views and do not represent the views and positions of this website. If you have more insights, please feel free to contribute and share.

Share To

IT Information

Wechat

© 2024 shulou.com SLNews company. All rights reserved.

12
Report