OpenAI seeks to make its upcoming open AI model best-in-class

By the end of March, Openai declared intention to release the first “open” language model since GPT-2 This year for a while. Now the details on the model begin to get out of the company’s sessions with the AI ​​developer community.

Sources TechCrunch, Openai’s research says VP’s Eidan Clark is the development of an open model in very early stages. Openai aims to release early spring and model – a substantiated model along the lines of Openai’s O-Series models – Benchmark-toping between other open thinking models.

Openai explores a highly permissible license for a model with several uses or commercial restrictions. Open models like Gemma of Llama and Google Criticized in some community To apply violent requirements – criticisms Openai wants to avoid seemingly.

Openai is facing the growing pressure of rivals such as China AI Lab DeepSeek, who accepts an open approach to starting models. Unlike the Openai strategy, these “open” competitors are experimenting and in some cases, they present to the AI ​​community for trade.

Proven to have a wildly successful strategy for some outfits. Meta, which has invested a lot of open AI models to the Llama family, said that in early March More than 1 billion loaded. Meanwhile, DeepSeek, gathered quickly to a large user base in the world and drew the attention of local investors.

Sources TechCrunch Openai will be “Text, Text Leverage,” to work in a high-level consumer apparatus, “Allow” substantiator “and off-turning models” Recently released by anthropic and others. (Can improve accuracy, but with increased delay value.) If the launch is not well received, Openai can watch with additional models, including smaller models.

Previous Public CommentsOpenai CEO SAM Altman said he thought Openai was on the wrong side of history while opening his technology.

“[I personally think we need to] Determine a different open source strategy, “Altman said in Reddit Q & A in January in January. […] We will produce better models [going forward]But we will protect our leadership less than we do in previous years. “

Altman, Openai’s upcoming open model will be comprehensive red and security will be evaluated for security. Sources are intention to release a model card for the company’s model for the company – a comprehensive technical report showing Openai’s internal and external benchmarking and security tests.

“[B]Leave the Effe, we will appreciate this model [to] Our preparation frame, as we want for another model, “Altman Last month he said in an article in X. “[A]If we know that we will be released after this model, we will do extra work. “

Openai, some AI ethicologists raised It was reported to have something the security of the latest models and Not to leave model cards for others. Altman Accused stands Openai executives related to model security reviews are wrong short folder In November 2023.

We have reached Openai for comment and hear back, we will update this piece.