Huawei has developed patented technology to protect the copyright of large AI models and automatically generate watermarked reply messages.

November 8 news, currently all major Internet manufacturers are developing their own large AI model products, which involve a lot of copyright disputes, and the competition among them is also extremely fierce.

Aiqicha App shows that on November 7, the patent "A language model protection method, device and computing equipment cluster" applied by Huawei Technologies Co., Ltd. was published.

The patent document states that developing systems based on large language models requires high machine and labor costs, so the large language model itself is a core asset for companies to build AI competitiveness. At present, research has confirmed that through model stealing technology, the functions of large language models can be replicated at a very low cost, resulting in the intellectual property rights of large language models being infringed. Therefore, it is necessary to effectively protect large language models, avoid model theft attacks, and realize the identification of existing infringements.

This site learned from the patent abstract that the method applied by Huawei includes:

Obtain the request text input by the user, and if it belongs to the target category, input the target instruction and request text into the language model Perform processing to obtain the first reply message with watermark words added and output it. The target instruction is used to instruct the language model to add watermarks to the processing results; when the category of the requested text does not belong to the target category, the requested text is input Go to the target language model for processing, obtain the second reply information and output it.

In this way, when processing a specific type of request through the language model, can automatically generate the reply information with watermark through the language model, realizing the Try to improve the copyright protection capabilities of the language model without damaging the quality of the text generated by the model.

The above is the detailed content of Huawei has developed patented technology to protect the copyright of large AI models and automatically generate watermarked reply messages.. For more information, please follow other related articles on the PHP Chinese website!