Everyone’s using AI to write. Stop pretending you’re not.
We are way past professors using AI to write their papers. Now professors feed a set of papers into AI and ask where the gap is in the logic so they can write a paper to fill it. When submitting, AI tells them who will likely be reviewing the paper for publication so they can strategically cite that person’s work. Spoiler: it’s working.
And really, why does this matter? Academic papers were never known for stellar writing anyway.
The disclosure game is BS
When I had paid links on my site, FTC law said I had to disclose them. I did not. My feeling was: if my writing is good, who cares if there’s a paid link? If I put a link in a post in a way that sounds stupid, you won’t keep reading my blog.
The same is true with AI. Who cares if someone let AI write their entire post? If you think it’s interesting, great. If not, don’t read it. You’ve been making this decision forever. Nothing changes.
I’ve had human editors mangle my voice much worse than AI does. My first major writing job was at a magazine where my editor constantly added ridiculous opening paragraphs. Here’s an example. But when you have an editor, you don’t get to tell her she’s bad at writing.
This is just the latest round of gatekeeping
“Don’t use AI to write” sounds exactly like “don’t use the Internet to find story ideas” (1995) or “don’t let Grammarly write for you” (2010). It’s the people who benefit from gatekeeping who try to keep the gate shut.
Meanwhile, the OG gatekeeper — The New York Times — is using AI to edit and write headlines. Because publications like the Times have such a strong, established voice, editors can easily adjust the AI output to stay consistent.
AI cannot say anything new
AI doesn’t make logical leaps unprovoked. It doesn’t say controversial things because it has a strong feeling that it’s right. When I had guest posts on my blog, I had a rule that the post had to say something I hadn’t heard before. Almost no one did that. Because when you say something new, people jump all over you. That’s what new is — it jostles what everyone thinks is true.
We’ve seen this before in music. Anyone can make passable music with our current technology. But the technology can’t automatically dream up something that hasn’t been done before. People have to do that. (See: street–streaming musical improv)
ChatGPT told me lawyers and entrepreneurs use AI most
Lawyers use it because half their day used to be busy work — now it’s ChatGPT filing briefs back and forth with itself. Entrepreneurs use AI for all their sales and marketing copy. This means most of what you read online is AI-generated. But we already knew that the Internet is going to hell. That’s why we use AI instead of Google.
If you’re a writer and not using AI, you’re nuts
I uploaded all 4,500 of my blog posts onto ChatGPT and told it to help me be the best writer I can be and still be me. Every time it makes an edit that I think is wrong, I add it to our rules document. Things like: don’t use bold for emphasis, edit the sentence so readers can feel the emphasis.
I can start with a messy post, and AI will organize it. Posts I used to throw out because I didn’t have a strong enough editor, I can now publish. AI tells me when I’ve lost my mind. Or I tell AI that I don’t care that it thinks there’s no evidence for what I’m saying — I want to say it anyway.
AI tells me that I need to be more strange. That I’m at my worst when I don’t let enough of myself in. I think that’s true, but I wouldn’t trust AI to know what’s brave versus what’s stupid. So I send almost all my posts to Nami, a real person, who can judge if I’ve gone too far or not far enough.
But you don’t need to know that to know if you like what you’re reading. You got to the end of this post. That matters more to me than anything else.
I can’t tell if you’re the ghost in/of/or the machine. It hurts to care. Need I? Maybe that’s how I know we’re still here.
Yes. The biggest issue is can we feel a conneciton when we read with the expectation of connection. It’s ironic because when I ask chatgpt what most people use it for, chat says connection. People want to be heard and understood.
Agreed. Rejection sensitive dysphoria sharpens the pain when expectation exceeds reality. It’s worse if you see an AI mask fall from what you thought was a human face. Your pride at a successful interaction is degraded to a sci-fi arc. If someone lies once, how can you know anything else they say is true? Maybe typos are the only way for humans to signal. Until they too become part of the ‘style’.
I’ve been using AI to do the heavy lifting on writing proposals and reports for work. I don’t care if it sounds like me, it just needs to get the job done. And AI delivers. It helps a lot that I’m in software development, that AI knows a lot about it, and that I’m not proposing or reporting on anything groundbreaking.
I have a blog about software development, dev.jimgrey.net. I’m using Claude to help me with that. I blast out my thoughts on a topic and then I feed it to Claude to organize and make sound like me based on my writing on that site.
I still do all of the writing on my personal blog, blog.jimgrey.net. I like the process.
I love hearing how people use AI in their life. I think a description like yours is one of the best ways for us to learn how to use AI for ourselves. There are no best practices yet — just us sharing. Your description makes me see that eventually a person starts to be a liability if they insist on adding their personal touches to stuff that doesn’t require a personal touch.
As I build my consulting business I’m getting advice to post more on LinkedIn. I hate LinkedIn, but I see how it could be valuable to me. A fellow I know who’s CEO of a design consulting firm posts there all the time. I asked him how I do it. He said he goes into a room for an hour every week with his assistant and just talks about things that are on his mind related to design. She records it and does some AI processing on it that spits out 5-10 LinkedIn posts that she just copy/pastes in. Brilliant!
“Entrepreneurs use AI for all their sales and marketing copy. This means most of what you read online is AI-generated. But we already knew that the Internet is going to hell. That’s why we use AI instead of Google.”
There’s another way that the internet is going to hell, and AI is in the handbasket along with it: the large language models are trained on publicly available text, and the greatest source of that is the internet. The text on the internet is increasingly written by AI. Therefore, increasingly, AI is eating its own production.
Even if you think your LLM is being trained on just the data set you give it, it’s still augmenting that with internet retrieval. Garbage in, garbage out, AI starts to drift and hallucinate, and the inevitable result is AI model collapse.
Some people like to say we’re on the verge of AI super-intelligence. But others say we might be at peak AI right now, with LLMs at their maximum usefulness before model collapse.
I love this writing: “There’s another way that the internet is going to hell, and AI is in the handbasket along with it.” I had to read the sentence twice, which I think most people did. AI never writes a sentence you have to read twice. So I’m thinking about my experience of reading it twice – it’s intimate. Becuase the sentence is like a broken rhyme that I have to step into your head to understand.
Okay. That’ one thing your comment made me think. Aother is that Z bought a book of Mary Oliver poems yesterday. (What? Where is this coming from??) So I read a few poems to see if maybe they’re about sex or something that I didn’t know. And what I realized is that the poems I loved in my 20s and 30s now sound like AI. Of course AI is trained on all the poetry in the world — because AI routinely is the preferable author when side by side with a human poet. So maybe that is good news for you??? I don’t know.
One thing AI told me: people like you who are managing multiple aspects of your life via AI make AI much smarter for everyone. What i heard: Mary Oliver is not the only person who sounds like AI.