The artificial intelligence company DeepMindowned by Google and known for developing AlphaZero and AlphaGo, has announced a major achievement in the race for equip AI with programming problem solving capabilities that are still usually considered solvable only by humans.
Thus, after simulating their participation in 10 contests with more than 5,000 participants, they affirm that their new AI —baptized as AlphaCode— has been within the estimated range of the 54% of the top participating code writers. Such contests had actually been held months before: the company put AlphaCode to the test by putting it through challenges selected by the Codeforces competitive programming platform.
However, they acknowledge that for now AlphaCode’s skills aren’t exactly representative of the kind of challenges the average programmer might face because, after all, it is not intended to solve the same kind of challenges that a programmer must face when developing a commercial programfor example.
On the contrary, AlphaCode is optimized to deal with tasks of “competitive programming”, that carried out in contests such as those of Codeforcesin which participants are given a series of long descriptions of problems and a few hours to write programs to solve them, receiving points based on the number of problems solved, which requires a high knowledge of programming logic.
ZAO, the Chinese MOBILE APP that through DEEPFAKE turns you into DICAPRIO in SECONDS
From Copilot to AlphaCode (and what lies ahead)
Oriol Vinyals, Principal Research Scientist at DeepMind, states that “in the long term, we are excited about the potential [de AlphaCode] for help programmers and non-programmers write codeimprove productivity or create new ways of developing software”.
For his part, Mike Mirzayanov, founder of Codeforces, assures that “the results of AlphaCode exceeded my expectations”, despite his initial skepticism, “because even in simple competitive problems often it requires not only implementing the algorithm, but also (and this is the hardest part) inventing it“.
The truth is that in recent years, little progress has been made in the development of artificial intelligence with programming capabilities, as evidenced by the launch of GitHub Copilot, a tool as popular as it is controversial, both at a legal level (because of the licensing problems that your training with pre-existing code represents) as security (because the aforementioned code is susceptible to the intentional introduction of rear doors).
Therefore, its main task for now is to act as an assistant capable of making suggestions to human programmersnot to replace them by automating software development.
Via | deep mind