| In short |
|
Keeping out of prying eyes, an intellectual competition has opposed a new generation chatbot to some of the best global calculation specialists. This discreet duel has highlighted a revolution in the way the machines are now tackling mathematical abstraction. Long considered as the last bastion of human intelligence, mathematics now seem accessible to machines capable of reasoning, test hypotheses and explore unprecedented solutions. This phenomenon raises exciting questions about the future of collaboration between humans and machines in the field of science.
A formidable efficiency in the face of complex problems
Among the new models of language, some are distinguished by their ability to reason well beyond the simple sequence of words. This is the case with o4-minia chatbot developed by Openai, optimized to solve abstract problems thanks to specific training. During a confrontation in Berkeley, this model found itself faced with unpublished mathematical problems, created specially for this test. According to LiveScience, Ken Ono, one of the organizers, tried to trap him with a question in theory of numbers, worthy of a doctorate. In less than ten minutes, the AI not only found a relevant solution, but formulated it with confusing insurance. AI has sent in real time which would have asked a human work for several weeks, with method and apparent logic. This technical feat questions the place of machines in mathematical research work and their ability to transform traditional methods.
“He moves like Messi on a walking truck”: this Chinese robot defies the laws of balance with amazing precision
Artificial intelligence and traditional methods
The project, baptized FrontierMathwas supervised by Epoch AI, an organization specializing in the evaluation of artificial intelligence models. The objective was clear: gauge the true potential of LLM of reasoning in the face of unpublished questions, therefore unknown to their learning base. Scientific American reports that the researchers had developed a series of events divided into increasing levels of difficulty. As the tests were linked, O4-Mini turned out to be able to solve around 20% of the most complex problems, those relating to the research level. Performance goes far beyond what traditional models were able to accomplish until then. This project highlights the capacity of machines to adopt a approach similar to that of a human researcher, exploring simplified cases, formulating hypotheses and gradually adjusting reasoning. For participants, the border between simulation and understanding has suddenly become less clear, thus raising the question of the evolution of research methods.
Tao and Kim spray human limits with IQs of 230 and 276, absolute geniuses which redefine the concept of global intelligence
The current limits of artificial intelligence
Despite the brilliance of the results obtained, the mathematicians present finally formulated ten puzzles that the machine could not resolve. This meager batch of human victories illustrates an essential reality: if AI calculates faster, it does not yet create with the same finesse. The production of good questions, the flair in the face of a mathematical intuition or the heuristic approach The reserved domain of humans remain for the moment. The researchers now evoke a double -sense collaboration where the AI excellent to demonstrate, but where humans would remain the architects of discovery. If this distribution of roles is confirmed, the function of the mathematician could evolve towards an initiator of ideas and exploration supervisor, like a mentor guiding a generation of digital brains.
“He has decoded the language of extraterrestrials”: a Chinese engineer resolves a mathematical theory that has been misunderstood
A collaborative future between humans and machines
In the era of artificial intelligence, collaboration between humans and machines seems promising. AI, with its ability to quickly process massive data volumes and formulate solutions, becomes a precious tool for researchers. However, human creativity, intuition and heuristic approach remain essential. Where machines demonstrate, humans design and guide. This complementarity opens up new paths for mathematical research and beyond. The future of this collaboration raises exciting questions: how can we maximize the potential of this symbiosis between man and machine? How far can this collaboration lead us in understanding mathematics and other scientific disciplines?
This article is based on verified sources and the assistance of editorial technologies.
Did you like it? 4.6/5 (26)
