DeepSeek accused of using OpenAI’s models to train its viral bot
OpenAI has accused Chinese competitor DeepSeek of using its models to train the competing system, violating its property rights.
The ChatGPT maker claimed that its new Chinese competitor employed a technique called ‘distillation’ – when a smaller model learns from its larger counterpart – to develop its new chatbot.
Donald Trump’s artificial intelligence adviser, David Sacks, said there was “substantial evidence” that DeepSeek distilled OpenAI’s output.
DeepSeek’s global takeover
DeepSeek recently gained global traction with the release of this new model, which was compared favourably to leading systems like OpenAI’s generative AI model, ChatGPT.
It shocked the market with its claims to deliver similar performance at a fraction of the cost of smaller rivals.
It was reportedly developed for only $6m (£4.8m) compared to the $100m (£80m) usually spent developing AI models.
However, as first reported by Bloomberg, the startup is believed to have used large amounts of OpenAI’s data to train its new large language model (LLM).
IT and software developers can pay for certain licences to access its programming as a means to integrate information.
While common in IT and AI development, this practice breaches OpenAI’s terms of service when used to train rival technologies.
A breech of terms
Last year, OpenAI and its partner, Microsoft, investigated accounts believed to be involved with DeepSeek that were using its application programming (API).
Suspecting illegal distillation, they blocked these accounts’ access to their APIs.
OpenAI is now going through the relevant measures to protect its intellectual property.
Sacks said: “I think one of the things you’re going to see over the next few months is our “leading AI companies taking steps to try and prevent distillation”
“That would definitely slow down some of these copycat models”, he added.
The tech titan is collaborating with the US government to safeguard its advanced models.
OpenAI has been approached for comment.