OpenAI Suspends TikTok-Parent ByteDance After It Used GPT To Train Its Own 'ChatGPT Of China'

Sam Altman-led OpenAI has suspended TikTok-parent ByteDance for using GPT to train its own competing AI model to create the "ChatGPT of China."

What Happened: ByteDance admitted to using OpenAI's GPT to train a competing AI model in China, violating the developer license of Microsoft Corp. MSFT as well as OpenAI, reported The Verge.

This comes after ByteDance publicly admitted to using a GPT-trained AI model in non-China markets. "ByteDance is licensed by Microsoft to use the GPT APIs," said ByteDance spokesperson Jodi Seth.

"We use GPT to power products and features in non-China markets but use our self-developed model to power Doubao, which is available only in China," Seth added.

TikTok's algorithmic "For You" feed has been a massive hit. Its parent ByteDance had claimed that it is coming up with an even "more powerful" model than Alphabet Inc.'s GOOG GOOGL Google Gemini, which narrowly beat GPT-4.

“Uncertain about GPT-5, but a super-strong model (more powerful than Gemini) is expected to arrive anytime now,” said Quanquan Gu, director of AI Research at ByteDance.

See Also: Watch: This 20-Minute News Segment Is Made Up Completely Of AI Anchors

This is also in violation of both Microsoft's as well as OpenAI's developer licenses, which specifically prohibit customers from doing what ByteDance has done. OpenAI has a clause that prevents licensees from using GPT output "to develop any artificial intelligence models that compete with our products and services."

"While ByteDance's use of our API was minimal, we have suspended their account while we further investigate. If we discover that their usage doesn't follow these policies, we will ask them to make necessary changes or terminate their account," OpenAI spokesperson Niko Felix said.

Why It Matters: ByteDance using another AI model to build its own could worsen the hallucination problem that plagues these models.

At present, these AI models rely on the training data given to them instead of generating their own data. Training another AI on the output of another AI could reduce the quality and also lead to even more fictional output.

Check out more of Benzinga’s Consumer Tech coverage by following this link.

Read Next: This AI Helmet Tries To Turn OpenAI CEO Sam Altman's Predictions Of Telepathic Communication To Reality

Photo via Shutterstock