This is also naive. In that case you won't be able to publish your results in prestigious journals and your career will die out.

That said, you do learn not to act 100% confident. You need to say "This suggests that" instead of "This proves that", etc.

Anyway, the solution to reliable data is simple: Have multiple competing groups try to repeat the work as closely as possible.

Now that you have something to hang your hat on, you come up with some theory/model to explain the phenomenon. Even better, come up with multiple competing such models. From these you deduce some precise prediction that is checked against future data.

All of this was well known a few generations ago...