OpenAI, the corporate behind the GPT/o1 collection of fashions, and Microsoft are investigating whether or not Chinese language AI startup DeepSeek obtained unauthorised information outputs from OpenAI’s fashions.
As reported by Bloomberg, safety researchers from Microsoft consider that people probably linked to DeepSeek are “exfiltrating a considerable amount of information” utilizing OpenAI’s API.
Microsoft is OpenAI’s largest investor, and, as per stories, it notified OpenAI of the suspected exercise, which violates the corporate’s phrases of service.
Furthermore, a number of DeepSeek customers on social media speculate that the mannequin shows comparable tendencies to OpenAI.
Am I lacking one thing? Did @deepseek_ai copy/paste @OpenAI docs and simply overlook to vary some references? Or is it some customary in docs that I'm simply not aware of
https://t.co/Rmts3RT9Q6 pic.twitter.com/jSM4vschS3
— Benita (@NirBenita) January 26, 2025
OpenAI additionally informed the Monetary Occasions that it had seen “some proof of distillation”, which is a method to enhance the efficiency of an AI mannequin through the use of outputs from one other one.
A person on Reddit additionally noticed the DeepSeek mannequin attempting to generate a solution that complies with OpenAI’s phrases of use.
DeepSeek’s newest reasoning mannequin, R1, has outperformed OpenAI’s o1, the corporate’s strongest mannequin obtainable for public use. R1 scored greater than o1 on a number of benchmarks involving logic, reasoning, coding, and arithmetic.
Just lately, DeepSeek’s official app dethroned OpenAI’s ChatGPT and different competing AI apps within the ‘Prime Charts’ on the US App Retailer for iPhone and iPad.
Just lately, round $589 billion was worn out from GPU large NVIDIA’s market cap. This was possible as a result of DeepSeek was constructed with little computing and capital, elevating considerations concerning the demand for GPUs and different AI assets to construct state-of-the-art fashions.
For example, one in all DeepSeek’s earlier fashions, the V3, used nearly 2048 NVIDIA H800 GPUs to realize efficiency higher than most open-source fashions. It additionally solely took $5.5 million to coach the mannequin.
Andrej Karpathy, former OpenAI researcher, stated the DeepSeek V3’s stage of functionality is “presupposed to require clusters of nearer to 16,000 GPUs”.
DeepSeek’s mother or father firm, Excessive Flyer, is a Chinese language hedge fund firm. Whereas the corporate was based in 2015, the DeepSeek mission was began in 2023.
US President Donald Trump stated, “The discharge of DeepSeek AI from a Chinese language firm ought to be a wake-up name for our industries.” He added that he views DeepSeek producing an AI mannequin utilizing cheaper strategies “as a optimistic”.
DeepSeek has additionally introduced Janus Professional, an AI picture era mannequin, which is claimed to supply higher outcomes than OpenAI’s DALL-E 3.
The publish Microsoft & OpenAI Examine if DeepSeek Obtained Information from OpenAI appeared first on AIM Media Home.
