I knew I wasn’t interested in A.I. for a while now, but I think I can finally put it into words.
A.I. is supposed to be useful for things I’m not skilled at or knowledgeable about. But since I’m not skilled at or knowledgeable about the thing I’m having A.I. do for me, I have no way of knowing how well A.I. accomplished it for me.
If I want to know that, I have to become skilled at or knowledgeable about the thing I need to do. At this point, why would I have A.I. do it since I know I can trust I’m doing it right?
I don’t have a problem delegating hiring people who are more skilled at or more knowledgeable about something than me because I can hold them accountable if they are faking.
With A.I., it’s on me if I’m duped. What’s the use in that?
Maybe I’m just too particular about things.
I cannot imagine an LLM world write the way I want it to be written.
Maybe not letters to your grandma, but to send out 1000 “your benefits have changed” in 600 subtly different ways.
I would argue that’s actually the last situation you’d want to use an LLM. With numbers like that, nobody’s going to review each and every letter with the attention things generated by an untrustworthy agent ought to get. This sounds to me like it calls for a template. Actually, it would be pretty disturbing to hear that letters like that aren’t being generated using a template and based on changes in the system.
TBH I don’t have much experience with it, because of the myriad other issues that plague LLMs, but style and tone is generally considered the thing that they’re good at.
If you want to sound like a mewling quim they are perfect!
I almost did not believe the words mewling and quim existed in real life language and had to look it up to ensure you didn’t write that comment with an LLM AI
I have some experience on the letter receiving side to share. I have a work colleague who recently decided it was a good idea to answer inquiries in MS Teams or email with LLM generated text. It was very obvious because the wording was too business-polished polite, was too verbose and did not sound like anything you would answer to a colleague ever. While the content was technically fine, the tone was missed by a mile. Also the generous use of the infamous em dash and unnecessary exclamation marks gave it away immediately.
That poses a problem. If you do that to a person you’re working with and they immediately know you’re serving them AI slop because you’re too lazy to be bothered with basic human interaction they WILL be offended. Same goes for customers if they know you personally or expect a human on the other side.
Humans are getting better at identifying AI garbage faster than LLMs improve. Because humans are still excellent at intuitive pattern recognition. Noticing that something is off intuitively is an evolutionary advantage that might save our ass.
Style and tone MIGHT be something they can mimic, but they are phenomenally bad at nuance. The LLM model loses information when it is constructed, and it similarly loses detail when it’s asked to elaborate on a point.