AI does not create new knowledge, said Thomas Wolf of the embrace

• Thomas Wolf said that AI excels in the following instructions but has trouble creating new knowledge.
• The AI ​​must question its training data and adopt counter-intuitive approaches, wrote the Face Exec on X.
• Wolf’s comments arise while Tech focuses on agency AI.

A has excelled in the following instructions – but that does not push the limits of knowledge, said Thomas Wolf.

The director of sciences and co -founder of Hugging Face, an Open Source company supported by Amazon and Nvidia, analyzed the limits of large language models in an article Thursday on X.

He wrote that the field produces “too compliant aid” rather than revolutionaries.

Currently, AI does not create new knowledge, Wolf wrote. Instead, he simply fills the whites between the existing facts – which he called the “collector filling”.

Wolf argues that for AI to know of real scientific breakthroughs, it must make more than recover and synthesize information. The AI ​​must question its own training data, adopt counter-intuity approaches, generate new ideas from a minimum of contribution and ask unexpected questions that open up new research paths.

Wolf also weighed on the idea of ​​a “21st century tablet” – a concept of an essay in October from the CEO of Anthropic, Dario Amodei, “Machine of Loving Grace”. Amodei has written that AI could accelerate so scientific progress that the discoveries expected over the next 100 years could occur in just five to 10.

“I read this essay twice. The first time I was totally surprised: I will change everything in science in five years, I thought!” Wolf wrote on X. “The rereading, I realized that a large part of that seemed to be a pious wishes at best.”

Unless AI, research changes speed, warned Wolf, we will not have a new Albert Einstein in a data center – just a future filled with “yes -men on servers”.

Wolf did not respond to a request for comments, sent outside the standard opening hours.

The rise of agentic AI

Wolf’s comments arise while the AI ​​world focuses on agency AI.

Related stories

Sam Altman, the CEO of Openai, predicted that it could be the year when the first “agents” – a set of artificial intelligence tools which can perform tasks independently – “join the workforce”.

“If 2024 was the year of the LLMS, we believe that 2025 will be the year of agency AI,” Praveen Akkiraju told Business Insider in January. He is managing director at Insight Partners, a venture capital company whose agency documents include the writer, Jasper and Torq.

Investors are betting very well on this idea. According to Pitchbook data, startups exploring the application of agents raised $ 8.2 billion last year.

Unlike AI assistants, who mainly recover and summarize information, agents can decompose complex tasks, make decisions and refine their approach according to the results.

Researchers also used AI to make scientific breakthroughs.

Oxford professor Matthew Higgins used AlphaFold2, a Deepmind IA tool belonging to Alphabet, to break the form of a key protein of malaria – something with which his laboratory had had trouble for years. This breakthrough led to testing an experimental malaria vaccine in people.

Without alphafold, “we will probably still try, to be honest,” Higgins at BI in 2023 said.