The lack of new training data also makes it bad at projects that are still maturing because it will suggest outdated code - or worse it will mix/match old and new syntax and generate something completely broken.
I worry that the lack of new examples for it to train on will self-reinforce running old syntax that has bad patterns.
If the "AI" could actually store its mistakes and corrections from interactive sessions long-term I think it would greatly alleviate this problem, but that opens up another whole set of problems.
I worry that the lack of new examples for it to train on will self-reinforce running old syntax that has bad patterns.
If the "AI" could actually store its mistakes and corrections from interactive sessions long-term I think it would greatly alleviate this problem, but that opens up another whole set of problems.