Openai seeks to make the upcoming AI “Open” better in its class

Near the end of March, Openai Declare Its intention to issue the first “open” linguistic model since then GPT – 2 At some point this year. Now details about this model have started to finish the company’s sessions with the artificial intelligence developer community.

Sources tell Techcrunch that Aidan Clark, Vice President for Research in Openai, leads the development of the open model, which is in the very early stages. Openai aims to make an early summer version and aims to make the model-style model model O-Series-which leads the measurement between other open thinking models.

Openai explores a very tolerant license of the model with a little use or commercial restrictions, for all techcrunch sources. They were open models like Llama and Google’s Gemma It was criticized by some in society To impose hard requirements – Openai’s criticism seems to be avoided.

Openai faces an increasing pressure from its competitors such as the Chinese AI Lab Deepsek who adopted an open -ending approach to launching models. Unlike Openai’s strategy, these “open” competitors make their models available to the artificial intelligence community of experience, and in some cases marketing.

It has proven to be a very successful strategy for some clothes. Meta, who has invested extensively in the Llama family of AI Open, said in early March that Lama had it Made more than a billion download. Meanwhile, Deepseek quickly collected a large user base around the world and Attract the attention of local investors.

Sources tell Techcrunch that Openai plans its open style, which will be “a text, a text outside”, to operate advanced consumer devices and may allow developers to switch and stop “thinking”, similar to thinking models Recently released by Antarob And others. (Logic can improve accuracy, but at the expense of increased cumin.) If the launch is a good future, Openai may follow with additional models-it is possible that they include smaller models.

Previously General commentsThe CEO of Openai Sam Altman said he believed Openai was on the wrong side of history when it comes to opening the sources of his technologies.

“I personally think we need to) discover a different open source strategy,” Altman said during a question and answer in January. “Everyone does not participate in Openai this opinion, as it is not our current top priority … We will produce better models (to move forward), but we will maintain less progress than we have done in previous years.”

Altman also said that the open model from Openai will be placed in the red category and evaluated for safety. Sources tell Techcrunch that the company intends to issue a model card for the model – a comprehensive technical report that shows the results of the internal and external standards and safety test from Openai.

(B) Edition before that, we will evaluate this model according to (to) our willingness, as we do in any other model, “altman He said in a post at x last month. (A) We will do additional work, given that we know that this model will be modified after the release. “

OpenAi angered some of the ethics of artificial intelligence It is said Hurry to test safety from modern models Failure to launch models cards for others. Also A defendant stands From Openai CEOs of Officer Officer on Model Safety Reviews before him The overthrow is short In November 2023.

We have contacted Openai to comment and we will update this piece if we hear again.