I'd even argue that if you are given a bunch of code of which 20% is incorrect, it's going to take more than 20% time to fix that. If you've written 100%, you know what the code is doing, what kind of data the variables are supposed to hold etc. In the 80-20% case you've got to build the internal model, test the code, reason and think about it, all of which you would have done during writing.

There are so many nuances to these kinds of statistics (same as the github copilot claims) that they give me a little marketing nausea every time they are claimed as truth.