OpenAI has accused newcomer rival DeepSeek of training its chatbot on the firm’s proprietary models.
There is some evidence that DeepSeek may have performed model distillation, OpenAI reportedly told the Financial Times, a technique whereby a smaller model is trained on a larger one to replicate certain responses for specific use cases.
The US AI and Cryptocurrency Czar, David Sacks, echoed the sentiment in an interview on Fox News, saying there is “substantial evidence” that DeepSeek trained its model on distilled knowledge from the US firm.
While distillation is gaining popularity as a technique in the industry, OpenAI’s concern centers around the fact it may be being used to create a rival model.
“The issue is when you [take it out of the platform and] are doing it to create your own model for your own purposes,” said one person close to OpenAI, according to the FT.
OpenAI’s terms of use make no clear reference to distillation as a practice, though it does indicate that users cannot modify, copy, lease, or distribute OpenAI services. It also states that users cannot use output to develop models that compete with OpenAI.
This is the latest in a series of shocks to US tech caused by the emergence of DeepSeek’s models, many of which are competitive with OpenAI’s technology despite costing only a fraction of the price to develop.
DeepSeek’s models are open source, causing several industry experts to herald this as an important moment in showcasing the power of open source AI as opposed to closed-source.
Shares of major US tech companies have been sliding since the release of DeepSeek’s latest models, as flustered markets react to the newcomer’s entrance.
Some reactions from big tech have been positive, though. OpenAI CEO Sam Altman called it “invigorating” to have a new competitor on the scene. Microsoft CEO Satya Nadella also seemed to welcome the competition, calling DeepSeek’s models very impressive.
