Attached: 1 image Here's a fun AI story: a security researcher noticed that large companies' AI-authored source-code repeatedly referenced
Via @mostlysignssomeportents, this story about how AI code completion tools in software development hallucinated a library that didn’t exist (ie a set of software routines called as an external file to your own program), and then someone made a real library to match that call, which could have contained malware.
This is why AI code completion doesn’t save time, because it doesn’t understand what it’s doing. It isn’t intelligent, it has no knowledge of good code practices. It requires human developers to double-check what it does, and you might as well have hired competent developers instead.
It is, as I have said multiple times, just autocomplete on steroids, filling in what it thinks us statistically likely to follow what you’ve already typed.