If you're truly honest about wanting to improve your skills then do not use AI!

Just write code. Any code! It doesn't matter. Spend the hours problem solving and debugging and banging your head against the wall. That's how we all learned and gained experience. But also reach out and seek help about specific problems you can't solve.

For a project idea, how about a program that uses the Lemmy API to scan resent posts for Python topics. It's something that can start basic but can be expanded upon. Start simple by just scanning posts from the last day. Then expand by adding a configurable date range, scanning several communities, multiple topics, presenting notifications, etc.

Working with ML libraries is a lot different than traditional software dev since those tend to be a lot more loosely typed and grounded in mathematics (linear algebra, calculus, etc).

To answer all of your questions at once: just keep writing code. Work on a lot of projects! Try reading the source code for other projects, like the libraries and tools you use.

Try to use LLMs as little as possible when writing code as well. It won't help you practice.

What do you mean by this?

when I am faced with a specific task, I tend to forget the actual implementation of the task