That’s a super interesting read, like, there have been a few services I used that were inexplicably awful in the exact ways described and I was bewildered by how bad it was.

See, it isn’t new and it isn’t AI, but it’s the same line of development as modern LLMs. They’ve just rebranded existing projects and lines of development as “AI technology” to be marketable.

Might be that information about when you do and don’t use the output is helpful for training. Like, if you use the output, good sign the output is good.