Openai has allegedly claimed that Deepsek may have distilled its Artificial Intelligence (AI) model to manufacture the R1 model. According to the report, San Francisco -based AI firm stated that there is evidence that some users were using their AI model output for a competitor, suspected to be deepsek. In particular, the Chinese company released the Open-SOS Dipsec-R1A AI model last week and hosted it on the molding and hugging face. The logic-focused model overcame the capabilities of the Chatgpt-Maker’s O1 AI model in several benchmarks.
Openai says that there is proof of foulplay
According to a financial time ReportOpenai claimed that its ownership AI model was used to train the model of Deepsek. The company told the publication that it had seen evidence of distillation from several accounts using the Openai App Programming Interface (API). The AI ​​firm and his cloud partner Microsoft investigated the issue and blocked their reach.
In a statement to Financial Times, Openai said, “We know [China]-The manual companies -and other -are trying to remove models of American AI companies. “Chat-Maker also said that it is working closely with the US government to protect its frontier model from contestants and opponents.
In particular, the AI ​​model distillation is a technique that is used to transfer knowledge from large models to a small and more efficient model. Here the goal is to bring the small model equal or large models, reducing computational requirements. In particular, Openai’s GPT-4 has about 1.8 trillion parameters, while Deepseek-R1 has 1.5 billion parameters, which will fit the details.
Knowledge transfer is usually using a relevant dataset from large models to train small models, when a company is creating a more efficient version of its model’s in-house. For example, Meta used the Lama 3AI model to create several coding-centered Lama models.
However, this is not possible when a contestant, in which a model does not have access to a dataset, wants to separate a model. If Openai’s allegations are correct, it could have been done by adding early injections to its API to generate a large number of outputs. This natural language data then converts to code and is fed into an base model.
In particular, Openai has not publicly issued a statement about this. Recently, the company’s CEO Sam Altman praised the Deepsek for creating such an advanced AI model and increasing competition in AI space.
