The creator of the chatbot ChatGPT made this statement in an interview with the Financial Times newspaper. OpenAI has found signs of a distillation process - improving the performance of small models by using the results of larger ones, but at a lower cost.
Representatives of OpenAI reported that they found some evidence indicating that DeepSeek models were trained in this way. In the fall of 2024, Microsoft security specialists revealed that individuals allegedly involved in the DeepSeek group were able to obtain a significant amount of information through the use of the OpenAI interface.
Photo: ru.freepik.com
According to FT, distillation is a common practice in the industry, but using mechanisms to create your own competing model is contrary to OpenAI rules and violates intellectual property rights.
The day before, Sam Altman, the head of OpenAI, promised to accelerate the release of new company products and introduce significantly improved artificial intelligence algorithms, commenting on the success of the Chinese startup in the market.
