Self-writing code

What's under the hood for programs that write their code? What are they lacking to do the job of a programmer? How will the development industry change because of these programs, and will it change at all soon?

Machine learning-driven code generation increases the scale at which it can work effectively:

• First, the IDE had auto-completion, also known as code completion;
• Then machine learning was connected to autocompletion and it began to work more accurately;
• Then there were full-line completion tools like TabNine and Kite. They learned how to substitute parameters in method calls, write error messages, and generally guess not one word, but several;
• In the summer of 2021, OpenAI and GitHub announced Codex and Copilot. They allow you to generate a function code by passing a text description in English to the input.

Is it time to be afraid of artificial intelligence, which will enslave us all? Everyone will decide for themselves.

Nikita and Roman will try to explain how these technologies work, show them in work, highlight strengths and weaknesses. We will also learn what JetBrains is doing in the field of automatic writing.