Like the article mentions, it's only good at boilerplate code at the moment, and can't really do architecture very well. I guess that's why it's "Github Copilot" and not "Github Pilot".

Going forward, who knows? We fundamentally don't understand why LLMs work.