“New GitHub Copilot Research Finds ‘Downward Pressure on Code Quality’ – Visual Studio Magazine”
No shit. I’m so surprised /s https://visualstudiomagazine.com/articles/2024/01/25/copilot-research.aspx
@baldur This seems like a gigantic DUH! to me. AI doesn't learn what's right or wrong because statistical models can't comprehend right and wrong. They can only copy and derive. AI will freely replicate horrifying security issues and because you catch it, that doesn't mean the other thousands of programmers who use AI will. If you ask it just right, AI will bury them in subtlety too.
But hey, look t the future where every dev is just reviewing generated code till no one knows how to program.
@RileyNorman @baldur Agreed. I've been saying from the start that novice programmers shouldn't use it because it reinforces bad habits.
But for those of us with good habits, it learns those habits from us and applies it to our future work. It's really a great tool when used by someone with adequate domain knowledge.
@RileyNorman @baldur Yeah, it's not a tool well-suited to bureaucracy, I agree. But for those of us who are independent contractors who can exert full control over our codebases, it's a huge boost to productivity.
@LouisIngenthron @baldur I don't think you're accounting for real world situations. Learning from novice programmers anywhere pollutes the entire model for everyone. AI is a 24/7 language derivation machine incapable of learning good habits from bad because it doesn't actually learn, another misnomer. It can only derive.
I have zero faith that investors and executives will effectively police who is adding to the models, particularly if they think they can fire you and save a nickel.