share_log

OpenAI宣称DeepSeek违规“蒸馏”,但没有给出证据

OpenAI claims that DeepSeek violated "distillation", but has not provided evidence.

cls.cn ·  Jan 30 00:56

① OpenAI claims to have found evidence indicating that DeepSeek used its proprietary model to train its own model, violating the service terms; ② "Distillation" refers to the process of using large models' knowledge to train smaller models to reduce costs, which is common in the industry but prohibited by OpenAI; ③ USA officials have labeled DeepSeek as "theft" and are conducting a National Security investigation.

On January 29th, Financial Association reports (Editor Zhao Hao) that the American AI company OpenAI claims to have discovered evidence proving that DeepSeek used OpenAI's proprietary model to train its own model.

OpenAI told the media that the company has found some evidence of "distillation." "Distillation" refers to extracting knowledge from larger, more powerful models into smaller models, which perform better and are more cost-effective.

Looking back at the cause of the incident, last week, China's Algo giant, Huansheng Algo, launched a new model called DeepSeek-R1, which performed comparably to OpenAI's strongest reasoning model, o1, in key areas such as mathematics, programming, and reasoning.

More importantly, the technical report indicates that the training cost of DeepSeek-R1 is only one-thirtieth of that of OpenAI's latest model. Following this news, the NASDAQ index in the USA dropped over 3% on Monday, with NVIDIA falling nearly 17%.

Distillation is quite common in the industry, but OpenAI's service terms do not allow such practices, stating that users cannot replicate its services and "output development of models that compete with OpenAI." OpenAI has not further commented nor provided details of the evidence.

A source directly informed stated that OpenAI and its partner Microsoft investigated accounts that used the OpenAI API last year and restricted access to these accounts on the grounds of suspected violation of service terms.

On Tuesday, David Sacks, the White House AI and Cryptos overseer nominated by Trump, stated, "There is substantial evidence showing that DeepSeek extracted knowledge from OpenAI's model, and OpenAI is not pleased about it."

Earlier this week, several officials in the USA stated that DeepSeek is "theft" and are conducting a National Security investigation into its impact. However, just on Monday, Trump referred to DeepSeek as a very positive technological achievement.

Editor/lambor

The translation is provided by third-party software.


The above content is for informational or educational purposes only and does not constitute any investment advice related to Futu. Although we strive to ensure the truthfulness, accuracy, and originality of all such content, we cannot guarantee it.
31
Comment Comment 49 · Views 100.4k

Comment(49)

Recommended

Write a comment
49

Statement

This page is machine-translated. Futubull tries to improve but does not guarantee the accuracy and reliability of the translation, and will not be liable for any loss or damage caused by any inaccuracy or omission of the translation.