Re:GitHub Copilot investigation
by bluenose69 on Tuesday October 18, 2022 @05:34AM
(#33244883)
Two things.
First, it would be nice to have a copilot variant that searched only my own work, so I wouldn't need to grep through other code I've written to get a reminder of how I solved a problem in the past.
And, speaking of the past ...
Second, I am old enough to have seen slide rules being replaced by calculators. This was a great addition to the toolbox, but it also had its downside: I've seen many students who have very clouded notions of significant digits, and many more who get quite confused with where to put a decimal point, when I ask them to compute something simple by hand.
Similarly, coding has been transformed with the advent of stack-like systems. There are two communities of coders now: those who learn a language and then can solve problems based on a solid foundation, and those who shorten the learning phase and code by web-search. The latter, it seems, are in danger of creating code that is brittle, limited, or downright wrong.
To the extent that copilot amplifies this habit of searching instead of thinking, I think it may lead to unreliable code.
So, sure, there are copyright issues. I think they have been well-discussed here and elsewhere. And courts may weigh in with new ideas. But my concern is with the reduction in code quality that may ensue. I'd love to see a discussion of the groups that are using copilot. If they are working on something I don't care about, then this is just a copyright issue. But if they are working on the "smarts" behind drug discovery, the control of dangerous machines, etc., then we have another issue, besides copyright.
First, it would be nice to have a copilot variant that searched only my own work, so I wouldn't need to grep through other code I've written to get a reminder of how I solved a problem in the past.
And, speaking of the past ...
Second, I am old enough to have seen slide rules being replaced by calculators. This was a great addition to the toolbox, but it also had its downside: I've seen many students who have very clouded notions of significant digits, and many more who get quite confused with where to put a decimal point, when I ask them to compute something simple by hand.
Similarly, coding has been transformed with the advent of stack-like systems. There are two communities of coders now: those who learn a language and then can solve problems based on a solid foundation, and those who shorten the learning phase and code by web-search. The latter, it seems, are in danger of creating code that is brittle, limited, or downright wrong.
To the extent that copilot amplifies this habit of searching instead of thinking, I think it may lead to unreliable code.
So, sure, there are copyright issues. I think they have been well-discussed here and elsewhere. And courts may weigh in with new ideas. But my concern is with the reduction in code quality that may ensue. I'd love to see a discussion of the groups that are using copilot. If they are working on something I don't care about, then this is just a copyright issue. But if they are working on the "smarts" behind drug discovery, the control of dangerous machines, etc., then we have another issue, besides copyright.