DeepSeek: OpenAI's very heavy accusation (ChatGPT)

© Mojahid Mottakin/Shutterstock

Did DeepSeek use OpenAI proprietary models to train its open source R1 language model?? This is in any case what the company led by Sam Altman claims, which specifies that it has evidence of this violation.

A major accusation

According to the American company, its rival would have used a technique known as “distillation”. Concretely, developers use it to improve small models by relying on the results of larger and more efficient models. Enough to produce an effective AI at a lower cost.

As one might expect, OpenAI does not appreciate this approach. Quoted by the Financial Times, the company laments: “The problem arises when you take it off the platform and do it to create your own model for your own purposes.”. This method would also violate the terms of use of the service.

200% Deposit Bonus up to €3,000 180% First Deposit Bonus up to $20,000

For now, the firm headed by Sam Altman refuses to say more. Contacted by our British colleagues, DeepSeek has not yet responded to their request. In any case, OpenAI and its partner Microsoft had already investigated last year on accounts suspected of belonging to the Chinese startup that had used the OpenAI API and blocked them due to suspicions of distillation that would violate the conditions of use, recalls the economic daily.

The White House in reinforcement ?

When asked about the subject, sector specialists do not seem very surprised. Like Ritwik Gupta, a PhD student in AI at the University of California, Berkeley: “It’s very common for startups and academics to use the output of commercial human-centric language models, like ChatGPT, to train another model.”.

For his part, David Sacks, the new crypto and AI guy in the White House, didn’t mince his words on Fox News: “There’s a technique in AI called distillation… where a model learns from another model and sort of sucks the knowledge out of the parent model. And there's substantial evidence that what DeepSeek has done here is they've distilled knowledge from OpenAI models, and I don't think OpenAI is very happy about that.”. The latter, however, did not cite any evidence.

In the meantime, the emergence of DeepSeek and its R1 model has instilled a wave of panic in the American tech world. This Monday, Nvidia's market capitalization even fell by nearly 600 billion dollars, before the company that produces electronic chips regained some color yesterday.

📍 To not miss any news from Presse-citron, follow us on Google News and WhatsApp.

[ ]