Top White House advisor says DeepSeek may have used OpenAI's models for training

DeepSeek, a Chinese AI startup, is alleged to have used OpenAI's models to train its own AI model, the R1, according to a White House advisor, David Sacks. The R1 model has caused market panic due to its efficient performance, which matches rival models but requires less computational power. Sacks suggested that DeepSeek might have engaged in intellectual property theft from OpenAI through a technique called distillation, where a smaller "student model" learns from a larger "teacher model." This has led to significant financial impacts on companies selling AI chips, such as Nvidia.

OpenAI is already under scrutiny for its training practices and faces several copyright infringement lawsuits from media organizations. The company offers its own model distillation services but prohibits users from copying its services or using its models to create competing ones. Despite the allegations, neither Microsoft, OpenAI, nor DeepSeek have responded to requests for comment. Sacks indicated that companies might take measures to prevent distillation to curb the rise of similar copycat models.

Key takeaways

DeepSeek may have used OpenAI's models to train its own AI model, raising concerns of intellectual property theft.
The Chinese startup's R1 model caused a market panic due to its efficient performance with less compute power.
OpenAI is facing scrutiny over its training practices and copyright lawsuits from media firms.
There is substantial evidence suggesting DeepSeek used a technique called distillation to mimic OpenAI's models.

Top White House advisor says DeepSeek may have used OpenAI's models for training

Key takeaways

Discussion (0)