- ByteDance sacks intern for sabotaging AI model training
- Company disputes reports of $10m damage to AI system
- Incident affects Doubao ChatGPT-like generative AI model
ByteDance, the parent company of TikTok, has announced the dismissal of an intern for allegedly “maliciously interfering” with the training of one of its artificial intelligence (AI) models. The company has, however, dismissed claims regarding the severity of the disruption caused by the intern, asserting that such reports “contain some exaggerations and inaccuracies.”
The incident, which came to light over the weekend, involved ByteDance’s Doubao, a generative AI model similar to ChatGPT, recognized as China’s most popular AI chatbot.
In an official statement, ByteDance clarified that the individual in question was an intern within the advertising technology team and lacked experience with the AI Lab. “The individual was an intern with the [advertising] technology team and has no experience with the AI Lab,” the company stated. “Their social media profile and some media reports contain inaccuracies.”
ByteDance emphasized that its commercial operations, including its advanced language AI models, were not impacted by the intern’s actions. The firm also refuted claims that the incident resulted in damages exceeding $10 million (£7.7 million) due to the disruption of an AI training system reliant on thousands of powerful graphics processing units (GPUs).
In addition to terminating the intern in August, ByteDance reported that it had notified the intern’s university and relevant industry bodies regarding the matter.
As one of the leading technology firms globally, ByteDance operates several widely used social media applications, including TikTok and its Chinese counterpart, Douyin. The company is regarded as a pioneer in algorithm development, thanks in large part to the engaging nature of its platforms.
Like many of its competitors, ByteDance is heavily investing in AI, utilizing the technology not only for Doubao but also for various applications, including a text-to-video tool named Jimeng.
