Saman said that Openai will issue an “open -weight” model of artificial intelligence this summer

Photo of author

By [email protected]


Sam Taman Today I revealed that Openai It will export open weight artificial intelligence Model in the coming months.

“We are excited about issuing a strong, strong, open -weight language model while thinking about the coming months,” the CEO wrote. On x.

This step is partially in response to fleeing success from The R1 model from the Chinese company DeepseekAnd as well as popularity Llama meta models.

Shortly after the release of the Deepseek model in January, Altman said that Openai was “on the wrong side of history” with regard to open models, indicating a possible shift in the direction. He said on Monday that the company is considering launching an open -weight model for some time, adding, “Now it is important to do it.”

Openai may feel the need to show that it can train the new model at a cheap price, since the Deepseek model has been trained on a small part of the cost of most artificial intelligence models.

“This is amazing news,” said Clement Dylangoy, founder and CEO of Hugingface, a company specialized in hosting Open AI models for WIRED. “With Dibsic, everyone realizes the power of open weights.”

Openai currently offers Amnesty International available through Chatbot and across the cloud. R1, Llama, and other open -weight models can be downloaded for free and adjustment. Old weights indicate the values ​​inside a large nervous network – which is set during training. Open weight models are cheaper for use and can also be designed for sensitive use situations, such as dealing with very confidential information.

Stephen Heidel, member of the Openai technical team, Re -spread The advertisement added, “We launch a model this year you can run on your own devices.”

Johannes Heidik, a researcher who works in the safety of artificial intelligence in Openai, also republished the message to X, adding that the company will conduct strict tests to ensure that the open weight model is not misused. Some artificial intelligence researchers are concerned that open weight models can help criminals to launch electronic attacks or even develop biological or chemical weapons. “Although the open models bring unique challenges, we are guiding with our willingness and will not issue models that we believe are catastrophic risks,” Heideckke wrote.

Openai today also published a web page Invitation to developers To apply for early access to the next model. Altman said in his position that the company will host events for developers with the early preliminary models of the new model in the coming weeks.

Meta was the first major company of artificial intelligence to follow a more open approach, as it was launched The first version of Lama In July 2023 a An increasing number of open -weight artificial intelligence models Available now. Some researchers notice that Lama and some other models Not transparent as it can beBecause training data and other details are still secret. Meta also imposes a license that limits the ability of other companies to profit from applications and tools designed using Llama.

March 31, 2025, 4:21 EST: This article was updated with a comment from Clement Dylangoy, Founder and CEO of Hugingface.



https://media.wired.com/photos/67eaa7594d92599e8c2f4021/191:100/w_1280,c_limit/GettyImages-2202831486.jpg

Source link

Leave a Comment