ByteDance, the Chinese tech giant behind TikTok, is using OpenAI’s technology to a “very limited extent” to test and develop its own AI models, it revealed on Tuesday.
The admission comes not long after OpenAI suspended ByteDance’s account over the latter’s potentially questionable use of GPT data.
Previously revealed ByteDance documents had already indicated that the Chinese tech giant is reliant on OpenAI API in developing its own foundational LLM.
However, the ByteDance spokesperson’s statement comes as a direct admission from the company that it indeed used OpenAI’s tech to train its own AI — a potential competitor.
ByteDance Potentially in Violation of OpenAI’s Terms of Service
Generally considered to be a faux pass in the world of AI, ByteDance’s tactic is potentially in violation of OpenAI’s terms of service.
While OpenAI admitted that ByteDance’s use of its API was minimal, it will be carrying out further investigation into the matter.
Outputs from the San Francisco-based company’s AI models cannot GPT model can’t be used in the development of any AI models that might compete with OpenAI’s products and/or services, the terms state. Though ByteDance is purchasing its OpenAI access through Microsoft, the latter has the same policy.
OpenAI is used alongside other third-party AI models by the company’s engineering team during the evaluation and testing phases, a ByteDance spokesperson admitted. However, they went on to add that GPT is used to a very limited extent in this regard.
GPT finds use in powering ByteDance’s products and features outside the country — the spokesperson stated. While OpenAI’s services aren’t available in the country, ByteDance has launched Doubao, a ChatGPT-like AI tool that’s available exclusively in China.
However, the company claims to use only a self-developed AI model of its own to power Doubao, which offers a range of services including chat robots, English learning assistance, and providing help with writing.
In case ByteDance’s usage of the OpenAI API is indeed found to be in violation of the latter’s policies, OpenAI might ask the company to make necessary changes or even terminate the account altogether.
All API customers must adhere to our usage policies to ensure that our technology is used for good.OpenAI
Internal documents from ByteDance relied on ChatGPT’s API during almost every phase of developing its own foundational LLM, which it codenamed “Project Seed”.
Back when the company rolled out Doubao in China a few months ago, ByteDance reportedly asked its staff to stop using GPT-generated text.
According to the ByteDance spokesperson, the company sent out an internal instruction ordering employees to refrain from adding any GPT-produced data into the training datasets of its own AI models.
The engineering team also underwent compliance training in regards to using GPT and adhering to its terms of service.
Several Chinese tech giants including Alibaba, Baidu, and SenseTime have already demonstrated the capabilities of their own ChatGPT-like AI tools.
Experts feel that these new ChatGPT-style tools from Chinese firms will further fuel the rivalry between China and the US. As the AI rivalry continues to ramp up, it remains to be seen if ByteDance indeed violated OpenAI’s terms of service.