Facial co-founder says it is difficult to say what company has won in a race

• Thomas Wolf, the face says that this is more difficult to say that AI model Traditional AI criteria are the best as saturated. It is said that Wolfe, the AI ​​industry can relate to two new correspondence approaches – agency-based and relating to the use.

Thomas Wolf, co-founder and chief Scientist while hugs the face, We think we can need new ways to measure AI models.

Wolf told the audience Brainstorm ai As a more advanced in London, it becomes more difficult to say that this is the best of it.

“It’s hard to say what the best model is,” said the nominal differences between Openai and the last releases Google. “It all seems, in fact very close.”

“The evaluation world has developed a lot. We thought that this very academic criterion was the most famous MMLU (massive multitask multitask multitask multitask multitask multitask multitask multitask multitask multitask multitask multitask language), the model of the model should respond to the model,” he said. “These are basically all satiated.”

In the last year, the academy, industrial and this general AI criteria, the academy, industry and politics, claiming that it is a real world program, has been an increasing choir.

In a study published in February, a joint research center was published in the joint research center of the European Commission “Can we trust the AI ​​criteria? A Lake consideration of current problems in an AI evaluation” “System defects in the current benchmarkment” “existing items”, “existing items”, construction, construction failures, game and data pollution game.

The forward AI industry said that the two main criteria going to 2025

Hugging his face is already working on the last one.

The company’s new program, “Your bench” helps users determine which model for a specific task. Users have several documents for users to see the best users to see the best for users.

“These models are not all the same as it works the same in this academic criterion, not all,” he said.

Open Source ‘Chatgept moment’

Founded by Wolf, Clément Delangue and Julien Chaumond in 2016, hug face was a long source champion.

The company presents an open source platform that allows the company, company learning Github, company, developers, researchers and enterprises to build machine learning models, supplied models, data and applications on a scale of data and applications. Users can view the models and information that others have uploaded.

Wolfe, Brainstorm AI, the “work model” is really open source “and the company’s” goal is the number of people involved in the open society and participating in the models. “

Wolfe predicted open source AI to develop EU in the open source after the success of Deepseek earlier this year.

After the start of last year, the Chinese-made AI model DeepSEek R1 sent Shockwaves through the AI ​​world when the testers are appropriate or even superior to American indoor AI models.

Wolf, Deepseek’s open source is a “chatring moment” for AI, he said.

“It was just the moment DeepSseek, who has just discovered the AI ​​of the world like Chatgpt, was the moment that the whole world was a kind of open society,” he said.

This story was first displayed Fortune.com

Source link