I'm doubtful that the results from the experimental process can be used to draw such generalizing conclusions about the utility of Copilot. The task (implementi

I'm doubtful that the results from the experimental process can be used to draw such generalizing conclusions about the utility of Copilot. The task (implementing a HTTP web server in JS) is already something that would be over-represented in public codebases, this is like asking participants to implement two sum or a sudoku solver, of course the language model would have essentially memorized these types of problems.