Ben, I am not a believer in AI. Having spent a large period of my life working on the very problems that the generative folk claim have been solved I consider that snake oil. Be that as it may, I would like to suggest that you include examples (specific) that show when you do x, you will get y. You have described some examples but I have found some is not enough. The more examples you have, the more it becomes a learning opportunity for the reader and I take it that is your goal here. Just a suggestion from an old guy. Randy.
My experience with Grammarly is pretty much exactly equivalent to yours. It catches a lot of my genuine mistakes but when I’m trying for a certain style - my own - it has difficulty following my meaning. In particular I love wordplay and cultural references and Grammarly is all at sea tracking my train of thought.
I ignore a lot of its suggestions.
As an editor, my rule of finger is four errors before I return a piece to an author with advice to run it through Grammarly or similar. I’m not going to spend my day correcting typos.
Here’s an experiment, though. Get a current AI model to generate some text and run *that* through Grammarly. There will be no mistakes. Indeed, it is one of the ways people pick AI writing. No mistakes and none of the “style” that generates words that might be seen as mistakes and their awkward corrections.
Run Shakespeare or Steinbeck through Grammarly and you'll see a sea of incarnadine.
Compare AI to the best of human writers and you'll find it falls short. Compare AI to the vast bulk of humanity - execrable stylists and banal with blandishments - and AI is clearly a better communicator than most. Especially those who have never learnt English.
AI is improving all the time. It’s evolving so fast that any story you publish is out of date. The twin drivers of massive investment and massive feedback are galloping it ahead.
As an editor, if someone submits AI that I can pick, I suggest they submit it to our specialist publication that handles AI. If I can’t pick it, I’m not going to do that. I’m going to publish it. It might read a little awkward but so what? A lot of human writing reads awkward. Am I really going to try to identify someone's racial or ethnic ability and suggest to them that one of their sort can’t write English proper?
Right now, AI writing - at least the stuff I can pick - is an accent. Too much and it’s obvious. Too little, and maybe the writer has crafted their prompt well, maybe they have massaged the output, and maybe it’s just a fallible human being who don't write good to start.
Or maybe it’s an AI model I can’t pick.
I don’t really care. If it’s good enough to pass as human, to communicate the intended message, and not contain any howling hallucinations, it does the job and it’s up to human readers to decide if they are entertained or informed enough to give it a thumbs up or down.
AI is in a journey and its potential to evolve is greater than ours. It’s evolving right now and I’m watching it with combined wonder and worry for what lies ahead.
You are on to something with needing to know the language. The deeper truth here is that you are a writer and care about the language. Someone who is not a writer implicitly would probably accept most suggestions without a second thought.
I use AI, and Grammarly. The latter regularly, and the former occasionally. For example if I write a sentence that just doesn't flow as it should, at least in my mind, I might submit it to ChatGPT and simply ask it to keep the original language, but to reorder it to improve the flow, as I was blocking on that. For such things it does a marvelous job, most of the time. I see it a tool like any other. You have to learn how to use it, before you should use it. Loved your article.
Ben, I am not a believer in AI. Having spent a large period of my life working on the very problems that the generative folk claim have been solved I consider that snake oil. Be that as it may, I would like to suggest that you include examples (specific) that show when you do x, you will get y. You have described some examples but I have found some is not enough. The more examples you have, the more it becomes a learning opportunity for the reader and I take it that is your goal here. Just a suggestion from an old guy. Randy.
My experience with Grammarly is pretty much exactly equivalent to yours. It catches a lot of my genuine mistakes but when I’m trying for a certain style - my own - it has difficulty following my meaning. In particular I love wordplay and cultural references and Grammarly is all at sea tracking my train of thought.
I ignore a lot of its suggestions.
As an editor, my rule of finger is four errors before I return a piece to an author with advice to run it through Grammarly or similar. I’m not going to spend my day correcting typos.
Here’s an experiment, though. Get a current AI model to generate some text and run *that* through Grammarly. There will be no mistakes. Indeed, it is one of the ways people pick AI writing. No mistakes and none of the “style” that generates words that might be seen as mistakes and their awkward corrections.
Run Shakespeare or Steinbeck through Grammarly and you'll see a sea of incarnadine.
Compare AI to the best of human writers and you'll find it falls short. Compare AI to the vast bulk of humanity - execrable stylists and banal with blandishments - and AI is clearly a better communicator than most. Especially those who have never learnt English.
AI is improving all the time. It’s evolving so fast that any story you publish is out of date. The twin drivers of massive investment and massive feedback are galloping it ahead.
As an editor, if someone submits AI that I can pick, I suggest they submit it to our specialist publication that handles AI. If I can’t pick it, I’m not going to do that. I’m going to publish it. It might read a little awkward but so what? A lot of human writing reads awkward. Am I really going to try to identify someone's racial or ethnic ability and suggest to them that one of their sort can’t write English proper?
Right now, AI writing - at least the stuff I can pick - is an accent. Too much and it’s obvious. Too little, and maybe the writer has crafted their prompt well, maybe they have massaged the output, and maybe it’s just a fallible human being who don't write good to start.
Or maybe it’s an AI model I can’t pick.
I don’t really care. If it’s good enough to pass as human, to communicate the intended message, and not contain any howling hallucinations, it does the job and it’s up to human readers to decide if they are entertained or informed enough to give it a thumbs up or down.
AI is in a journey and its potential to evolve is greater than ours. It’s evolving right now and I’m watching it with combined wonder and worry for what lies ahead.
Britni
You are on to something with needing to know the language. The deeper truth here is that you are a writer and care about the language. Someone who is not a writer implicitly would probably accept most suggestions without a second thought.
I use AI, and Grammarly. The latter regularly, and the former occasionally. For example if I write a sentence that just doesn't flow as it should, at least in my mind, I might submit it to ChatGPT and simply ask it to keep the original language, but to reorder it to improve the flow, as I was blocking on that. For such things it does a marvelous job, most of the time. I see it a tool like any other. You have to learn how to use it, before you should use it. Loved your article.