Post #4506723

I think a lot of folks seem to argue for or against "AI" based on how good it is at what it does. This seems to surface in the, "it has hallucinations", "it can't think", "it makes mistakes", "AI slop" line of reasoning. I think those arguments are facing challenges in the light of quite reasonable performance For example, I've been working with Sonnet 4.6 recently on a code base that some might consider legacy. With rules and guidance, it does a reasonable job. It gets you 80% of the way to completing tasks with minimal effort So maybe the argument now has to move from the polar opposites of can it / can't it to the much tricker: should it / shouldn't it I've seen most value over recent customer projects by starting with the problem and working out the solution, of which AI is one potential answer - factoring in its trade offs. Some of these are easy, such as non-determinism, but some much harder yet more important (imo), such as environmental or social