On December 16, some media reported that ByteDance has been secretly using OpenAI technology to develop its own big language model (LLM). Niko Felix, a spokesman for OpenAI, said that the ByteDance account has been suspended.
Jodi Seth, spokesman of ByteDance, responded that the data generated by GPT was used to annotate the model in the early development of the "Project Seed", and had been deleted from the training data of ByteDance around the middle of this year.
The statement said: "ByteDance has obtained the license to use GPT API (Application Programming Interface) through Microsoft. We use GPT to provide power for products and functions in non Chinese market, but use our self-developed model to provide power for Doubao AI, which is only listed in China."
The foreign media said that a document obtained from ByteDance learned that in the big language model project code named "Seed Plan", ByteDance relied on OpenAI API for development at almost every development stage, including training and evaluation models.
The "Seed Plan" was launched about a year ago and currently focuses on developing two products, one of which is the Doubao that has already been launched in China; The other is a chatbot platform aimed at commercial users, which is currently under development.
After the report was released, OpenAI spokesman Niko Felix said: "Although ByteDance uses our API very little, we have suspended their accounts, and we will further investigate. If we find that their use does not conform to the rules, we will ask them to make necessary changes or terminate their accounts."
The above "rules" refer to a clear provision in the OpenAI service terms that the model capabilities provided by OpenAI are not allowed to be used to "develop any AI models that compete with its products and services.". ByteDance purchased the access rights of OpenAI through Microsoft, but Microsoft also formulated the same terms as OpenAI.
Microsoft spokesperson Frank Shaw stated in a statement: Microsoft's AI solutions, such as Azure OpenAI services, are part of our limited access framework, which means that all customers must apply for and obtain Microsoft's approval to access them. We have also established standards and provided resources to help our customers use these technologies responsibly and comply with our service terms. We have also established a process for identifying abusive behavior and stopping businesses from violating our code of conduct Our visit. "
In this year's "Hundred Model War", ByteDance was regarded as "slow half beat" by the outside world, and the volume was not large. In June, ByteDance Volcano Engine released Volcano Ark, a large model service platform. The operation mode is to provide multiple models in one platform, so that enterprise users can try them synchronously, and select a model combination more suitable for their own business needs. The general large model "skylark" developed by ByteDance itself was not revealed in the list of the first large models that passed the Interim Measures for the Management of Generative Artificial Intelligence Services until August this year. On August 17th, the AI dialogue product "Doubao" based on the Skylark model officially began external testing.
Doubao AI's official WeChat account introduces that Doubao AI can provide functions such as chat robots, writing assistants, and English learning assistants. It can answer various questions and engage in conversations, helping people obtain information. It supports web platforms, iOS, and Android platforms. Doubao can provide various types of assistance, including natural language processing, knowledge comprehension, dialogue, information retrieval, sentiment analysis, and machine learning.