Replying to messages can automatically generate watermarks! Huawei language model’s patent for protection has been announced.

New Huawei language model patent automatically adds watermarks to message replies for enhanced protection announced!

[Mobile China News] Recently, Huawei Technologies Co., Ltd. has released a patent for “A method, apparatus, and computing device cluster for protecting language models.” According to the information from the Enterprise Check App, this patent was applied for on June 19, 2023 and publicly announced on November 7, 2023. The inventors are Wu Chuhan, Meng Xiaojun, Dong Zhenhua, and Tang Ruiming.

According to the patent abstract, this is a language model protection method, which includes: obtaining user input request text; if the category of the request text belongs to the target category, inputting the target command and request text into the target language model for processing, obtaining the first reply information with added watermark words, and outputting the first reply information. The target command is used to indicate the target language model to add a watermark to the result of processing the request text; if the category of the request text does not belong to the target category, inputting the request text into the target language model for processing, obtaining the second reply information, and outputting the second reply information. This way, when processing specific types of requests through the language model, the language model can automatically generate reply information with watermarks, thereby improving the copyright protection ability of the language model while minimizing any impact on the quality of generated text.

• HarmonyOS 4 Upgrade Device Quantity Exceeds 100 Million, NEXT Always Has Surprises
• iPhone 16 Pro concept map exposed Netizens say they only guess the name right every year
• The foldable screen smartphone market is booming, and Huawei deserves the credit for the perfect software ecosystem.

By understanding the patent abstract and the schematic diagram, we can see that when dealing with responses from the language model, the addition of watermarks for protection is determined based on the target category, thereby enhancing the copyright protection ability of the language model.

We will continue to update Phone&Auto; if you have any questions or suggestions, please contact us!