AI prose is mediocre right now. Too verbose, indirect constructions, passive, etc. That being said, it's actually a great editor and can pick out all those issues consistently.
My workflow right now is to use AI for rough draft and developmental editing stages, then switch AI from changing files to leaving comments on files suggesting I change something. It is slower than letting it line/copyedit itself, but models derp up too much so letting them handle edits at this stage tends to be 2 steps forward 2 steps back.
That's my main criticism as well. Even before we get to the ethical implications of AIs communicating on your behalf without a disclaimer, LLM writing is just poor and making me read through it is disrespectful of my time.
I recently had a colleague send me a link to a ChatGPT conversation instead of responding to me. Another colleague organised a quiz where the answers were hallucinated by Grok. In some Facebook groups I'm in where people are meant to help each other, people have started just pasting the questions into ChatGPT and responding with screenshots of the conversation. I use LLMs almost daily, but this is all incredibly depressing. The only time I want to interact with an LLM is when I choose to, not when it's forced on me without my consent or at least a disclaimer.
> I recently had a colleague send me a link to a ChatGPT conversation instead of responding to me.
I find this kind of thing interesting anywhere someone is being paid more than minimum wage: a really good way to make your boss think that they can replace you with ChatGPT is for you to perform it at ChatGPT’s level. I do give them points for not trying to hide it, but it really seems shortsighted not to consider that each time you do that, you’re raising the question of why they shouldn’t cut out the middleman.
> I recently had a colleague send me a link to a ChatGPT conversation instead of responding to me.
I honestly would rather this than my colleague sending me text that is obviously from chatgpt, but not stating it upfront. Or even the "I asked chatgpt and it said this.." along with pasting 10 paragraphs of stuff they didn't even read to confirm it could be relevant.
I've always used AI as an editor, which is to say I give it the my efforts and ask it to highlight mistakes. They invariably find a few. They occasionally miss a few too, so they aren't particularly reliable editors.
They aren't reliable at anything I guess, but for English I have nothing else, and they are better than nothing. I do wish they would use a more effective way of highlighting their suggested changes, such as italics for new text and strikeout for deleted text.
Unless you are paid by the word, I struggling to think of why you would use an AI to create new text. The facts will be wrong, the tone won't be yours. "If I had more time, this would be shorter" is a truism here - AI can spit out an enormous amount of text in a very short time, text could be cut down to a fraction of the size with a bit of effort.
You can prompt AI to focus on writing prose that is optimized for impact and clarity, and its output is MUCH better.
AI prose has been mediocre since the release of ChatGPT. My layman's interpretation is there's just no strong creativity / humor / etc signals to train on, as compared to say math or coding. Current models are "smarter" so when asked to produce eg a joke they think harder, but the end result always misses the mark just the same.
There's a difference between AI being bad at prose and storycraft. Good prose is totally achievable and it's just that it hasn't really been a priority for the tech shops, and I think they also often don't understand what makes really good prose so they're not good at optimizing for it anyhow. I expect given people's aversion to slop that the big laps will start to push hard on it soon and get their act together though.