As part of the Think 2021 conference, IBM presented an artificial intelligence-based algorithm called CodeNet. It is able to solve the problem of incompatibility of programming languages, acting as a translator.
The main feature of the complex generative system CodeNet is the ability to create separate sections of code and even entire projects by directly translating data from one programming language to another, says 4PDA. For example, the user knows one or more of the older languages used in highly specific scenarios. Now they do not need to retrain, because the AI will do the translation for them.
IBM officials say CodeNet has been trained in 14 million code snippets, or 500 million lines, and is capable of translating 55 old and new programming languages, from COBOL and FORTRAN to Java, C ++ and Python. It is not reported how effective the AI was, and it is not yet known whether the algorithm will be able to optimize new languages after converting to old code without increasing the size of the final files.
NIX Solutions notes that in addition to the translation function, it is possible to extract metadata and check generative models for correctness by launching individual parts of the program. This will allow you to control the translation by directly comparing code sections.