So, it is loosely the same as copilot? I understand that approach is a tad different, but result of converting natural language descriptions into code-changes should be comparable.
And both are trained on large corpus of github sources
Is there a way to test it somehow? Public API maybe?