I use it a lot for coding in R and it's often very helpful - I don't have to write code completely from scratch but can start with its code.
How do I know it's right?
I run it on test examples where I know the answer.
Also, I can read R code (just to lazy to generate it), so I read the generated code and might go "Oh, cool, that's a neat way to do it", or alternatively "What the hell are you doing?"
If it generates code that doesn't execute, I don't even read it, I'll send it back and tell it to debug it. I've found that if it can't debug its own code after three tries, then I need to read it and debug it myself (sometimes AIs can be very dense).

In the next reply, I'll post a short summary about a new result some AI researchers got about GPT-4o that is fairly alarming.
And - these things are getting pushed out the door as fast as possible in various ways to monetize them. Capex of hundreds of billions of dollars per year in development cost can't go on forever.