White House expresses alarm over DeepSeek’s AI techniques

Top White House advisers have raised concerns over China’s DeepSeek using a technique known as “distillation” to potentially replicate US AI models, a method where one AI system learns from another. This could allow DeepSeek to benefit from the extensive investments made by US rivals, such as OpenAI, without incurring the same costs. DeepSeek recently made waves by releasing an AI model that rivals those of US giants, at a fraction of the cost, and giving away the code for free. US tech companies, including OpenAI, are now investigating whether DeepSeek’s model may have improperly used this distillation method.

Distillation, while common in the AI industry, may violate the terms of service of models like OpenAI’s. The technique allows a newer, smaller model to benefit from the learnings of a larger, more advanced one, often without detection, especially when using open-source models. Industry experts have pointed out that blocking such practices is difficult, particularly with freely available models like Meta’s Llama and French startup Mistral’s offerings. Some US tech executives, however, are advocating for stricter export controls and customer identification measures to limit such activities.

Despite the concerns, DeepSeek has not responded to the allegations, and OpenAI has stated it will work with the US government to protect its intellectual property. However, as AI technology continues to evolve, finding a way to prevent distillation may prove to be a complex challenge. The ongoing debate highlights the growing tensions between the US and China over the use of AI and other advanced technologies.