Amnesty International Dream of Google DeepMind, algorithms, beyond human experience

Main question in artificial intelligence It is the number of times that the models exceed mere renewal and re -enhancing what they have learned and really producing new ideas or visions.

New project from Google DeepMind He explains that with some smart modifications, these models can exceed at least human experience that designs certain types of algorithms – including those beneficial to progress in artificial intelligence.

The latest project of the company’s artificial intelligence, is called alphavolveIt combines coding skills for its Gemini AI model with a way to test the effectiveness of the new algorithms and the evolutionary way to produce new designs.

Alphavolve has reached more efficient algorithms for several types of account, including a method of accounts that involve the improvement of a method called a 56 -year -old STSSSEN algorithm. The new approach improves mathematical efficiency by reducing the number of accounts required to produce a result.

Deepmind also used alphavolve to reach better algorithms for many problems in the real world including task schedule inside databases, drawing computer chips design, improving the design of algorithms used to build large language models such as Gemini itself.

“These are three crucial elements of the modern Amnesty International’s ecosystem,” says Pushmeet Kohli, head of Amnesty International Science at DeepMind. “This supernatural coding agent is able to assume certain tasks and much exceeds what is known in terms of solutions to them.”

Matej Balog, one of the research in Alphavolve, says it is often difficult to know if the great language model has truly reached a part of the new writing or code, but it can be shown that no one has reached a better solution to certain problems. “We specifically showed that you can discover something new and new,” says Balge. “You can really be sure that what you found could not be in training data.”

Sanjeev Arora, a scientist at Princeton University specializing in the design of algorithm, says the developments presented by Alphavolve are relatively small and are only applied to algorithms that include research through an area of ​​potential answers. But he adds: “Research is a general idea that is applicable to many settings.”

The coding, which works with the artificial intelligence system, began to change the way developers and software companies write. The latest artificial intelligence models make beginners to create simple applications and web sites, and some developers with artificial intelligence are used to automate more of their work.

Alphavolve explains the possibility of completely new ideas through continuous experimentation and evaluation. Deepmind and other artificial intelligence companies hope that artificial intelligence agents gradually learn to show more general ingenuity in many areas, and may eventually generate brilliant solutions to a new work or visions when giving a specific problem.

Josh Al -Mam, an assistant professor at the University of Colombia works to design the algorithm, says Alphavolve seems to generate new ideas instead of re -confusion that you learned during training. “You should do something new and not just renew,” he says.