Thomas Wolf, co-founder and chief science officer of Hugging Face, will speak at the opening ceremony of the Web Summit in Lisbon, Portugal on November 11, 2024.
Pedron Nunes | Reuters
Current artificial intelligence models from labs like Openai are unlikely to lead to major scientific breakthroughs, according to the technology co-founder, pouring cold water into some of the hype around technology, claiming claims by key figures on the ground.
Comments by Thomas Wolf, co-founder of the $4.5 billion AI startup hug Face, contrasts with key names for AI, including Openai boss Sam Altman and humanity CEO Dario Amodei.
When Wolf talks about scientific breakthroughs, he means novel ideas like those on the level of the Nobel Prize. Examples such as Nicolaus Copernicus, who theorized the Sun, are at the heart of the universe, where other planets circumference.
Wolf is currently explaining some issues with chatbots. First, these products, such as ChatGpt, often agree or align with the person urging it. Ask the chatbot to prompt and tell us how interesting the question is, no matter how wonderful it is.
Second, the models that underpin these chatbots are designed to “predict” the most likely to token” or “predict” the “word” in the sentence.
However, he noted two important traits of scientists. The first is that scientists who make big breakthroughs are often paradoxical and question what others are saying.
“Scientists aren’t trying to predict the most likely words. He is trying to predict this very novel thing that is surprisingly unlikely in reality, but in reality true,” Wolf said.
The co-founder of embracing face has been thinking about the topic for the past few months. His interest was sparked after reading an essay written by Amody of Mankind. He hypothesized that “AI-enabled biology and medicine will allow human biologists to squeeze progress that would have reached five to ten years over the next 50-100 years.”
It made me think about the state of AI and how, in his view, the current model harvest would not be impossible.
Wolf said these chatbots and tools are likely to be used as a kind of “scientist co-pilot” used in research to help humans generate new ideas.
To some extent, this has already happened. Google Deepmind’s AlphaFold products helped us analyze the protein structures that the company promised.
However, there are several new startups, such as Lira Science and Future House, that want to take AI a step further so that they can make scientific breakthroughs.
