Maybe it is
Isn’t that basically how it works?
Just add an additional group of monkeys that aggressively hires and fires the others based on performance
Tada, machine learning
asks for password hashing
gets code that looks like password hashing, named like password hashing, but, without any of the hashing
It generates the blurst of code!
I’ve been using copilot and find it’s suggestions are perfectly cromulent.
Tbh, copilot was probably the worst AI coding experience I’ve had. It actually made me less productive and made me question my competency as a programmer at the same time. Straight up did not have a good time. Use Cody or GPT-4 instead.
It is designed for other purposes than GPT models. Next time try to use copilot as autocompletion, not to generate new code. It’s excellent in that.
That’s how I thought it was supposed to be used. It’s “copilot” not “autopilot”. I don’t need nor want it to write whole functions for me.
Company ran a trial for it, and it worked really well for generating boilerplate code following our existing system design. Sometimes it makes mistakes, but during the trial it was a rare occurence
The company is giving it to us all for free next year, hope it doesn’t negatively affect hiring though…
That’s how I was using it; I ended up spending as much time as I was saving going around and cleaning up after it and/or second guessing myself. Basically, because it only operates in the context of the file you’re working in, it will suggest garbage half the time if you have to work with resources from other files.
But the propaganda from GitHub said it was making devs 80%+/- more productive!
How could this have happened? /s
Because we are the apes that wrote the code that copilot read.