Artificial intelligence systems are unlikely to make scientific discoveries some of the leading laboratories that scientists hope for face,

Talk to luck At Viva Technology in Paris, the co-founder of Huging Face said that although the LLMS models showed a great ability to find answers to questions, they shorten when trying to ask them to the correct models-something that the wolf considers the most complex part of real scientific progress.

Wolf said: “In science, asking the question is the difficult part, he does not find the answer.” “Once the question is asked, the answer is often completely clear, but the difficult part is to ask the question really, and the models are very bad in asking great questions.”

Wolf said that he had reached a conclusion after reading a blog post widely circulated by the CEO of Anthropor Dario Amani called Love machines. In it, Amodei argues that the world is about to see the twenty -first century “compressed” in a few years as artificial intelligence greatly speeds science.

Wolf said that he initially found an inspiring piece but began to suspect the perfect Amodei for the future after the second reading.

He said: “It was among the artificial intelligence to solve cancer and will solve mental health problems – he would even bring peace to the world, but I read it again and realized that there is something that seems very wrong with it, and I do not believe that.”

For a wolf, the problem is not that artificial intelligence lacks knowledge but lacks the ability to challenge our current knowledge framework. Artificial intelligence models are trained to predict potential continuity, for example, the following word in the sentence, and while today’s models excel in simulating human thinking, they lack any real original thinking.

“The models are only trying to predict the most likely thing,” explained Wolf. “But in almost all cases of discovery or great art, it is not the most likely artistic piece that you want to see, but it is the most interesting.”

Using an example of the GO game, a tablet game that has become a milestone in the history of artificial intelligence when DeepMind’s Alphago was defeated by world champions in 2016, Wolf has argued that during the mastery of Go’s bases is impressive, the biggest challenge lies in the invention of this complex game in the first place. He said that the equivalent of inventing the game is to ask these original questions really.

The wolf first suggested this idea in a blog post entitled Einstein AI modelPublished earlier this year. In it, he wrote: “To create Einstein in a data center, we do not only need a system that knows all the answers, but instead he can ask questions that no one else thought or dared to ask.”

He argues that what we have instead is models that are behaving like “Yes-Men on Servs”-which is well acceptable, but it is unlikely to challenge assumptions or rethink basic ideas.