Google DeepMind researchers say they have made the first real scientific discovery using a large language model, suggesting that the technology behind ChatGPT, Bard and other chatbots can generate information beyond human knowledge — rather than just teaching them, as information feeds are repackaged.
The AI system called FunSearch is based on Google's large language model, PaLM 2, which is equipped with a fact-checking function. The system provides solutions to various mathematical problems in the form of computer programs and then automatically ranks them according to how well they perform. The best programs are then combined and fed back into the language model, which ultimately explains the best possible solution in understandable human language.
This approach differs from previous DeepMind approaches (e.g weather forecast Or the Protein research (Revolutionary Breakthroughs), that while these models are trained for a specific task by feeding them the data required for a specific area, FunSearch is based on a large language model capable of providing general responses, so it can be used on a much broader scale.
The researchers distributed the system over two tasks more comprehensively. One of them is known in asymptotic geometry, the so-called “maximum set problem”, where you have to find the largest set of points in space where three points do not form a straight line – FunSearch has created programs for this purpose that generate new points, large sets and override the results that reach Mathematicians have reached it until now.
In the other task, they looked for a solution to how to pack objects of different sizes into a storage container as efficiently as possible. In fact, this is important, for example, in shipping and container loading, but it is a valid problem in many areas of mathematics, for example when trying to solve the scheduling of computing tasks in data centers efficiently. the Recent results published in Nature According to FunSearch, we found a way to solve this problem that avoids leaving small gaps in the containers that cannot be filled later.
“Over the past two or three years, there have been exciting examples of mathematicians working with artificial intelligence to make progress on unsolved problems. The latest results could provide another very interesting tool for similar collaborations and allow mathematicians to effectively search for intelligent, unconventional methods.” Expectations that humans can explain. Tim Gowers, a professor of mathematics at the University of Cambridge, who was not involved in the research, told The Guardian.
Read the full article on Cobbett. on the side.
Follow our page A in Facebook And the On Twitter also!
