Talking points for yall's reference:
General idea:
Use external resources not as a source of ground truth to encode into your head, but a giant toolbox to give you an overview when you know nothing about your question.
Example Time!
When you are a newbie and you get a lot of conclusive statements from external sources, it’s not which side is right or wrong, if there’s not enough factual that can help you to judge what’s the assumption in both side of the argument.
Be more questioning or hesitant because it will propel you to more action to draw your own conclusion from facts you collected from your own experiments.
It might be some what wasting time to verify some conclusion that is widely accepted, but it really does not hurt you because over time, you will find the ability to use code/action to draw your own conclusion is much more useful than the ability to acquire information in the field of machine learning and many other fields.
Practical example:
The kind of question that gets you a tutorial loop:
Why does this concept work on this task/model? (Why question)
The kind of question that drives you to code and experimentation:
Why does not the other concept work on this task/model? (Why not question)
Trust your intuition, implement bold questions into experiments (hence more action), and test your idea with experiments instead of theory proofs from external resources.
Loop:
Search for code, run the code, debug the code, modify the code, think about what to do next (reading papers/books)