Episode Transcript
[00:00:00] Foreign.
[00:00:05] Welcome to ChatGPT Curious, a podcast for people who are, well, curious about ChatGPT. I'm, um, your host, Dr. Shantae Cofield, also known as the Maestro, and I created this show to explore what ChatGPT actually is. Really, though, are the files in the computer, how to use it, and what it might mean for how we think, work, create, and move through life. Whether you're skeptical, intrigued, or already experimenting, you're in the right place. All that I ask is that you stay curious. All right, let's get into it.
[00:00:38] Hello, hello, hello, my curious people, and welcome to episode 34 of what podcast. I'm still working on the name as, uh, it currently stands. It is chatgpt Curious, but I still need to change it. And I will, y', all, the lift the lift of this name change and the domain and all the things it will happen.
[00:00:58] But for right now, we're just gonna say episode 34. I am your grateful host, the Maestro, and today we're talking about AI patterns that are a dead giveaway. To be more specific, eight AI writing patterns that are dead giveaways. You, you just know when you see it, you all have seen it. And I want to use this episode simply to name those patterns. That's literally it. I want to give you language for the things that I know you 100 see and feel when you're reading threads and captions and email. Just like this was 100% written by AI. My goal is not to be able to have you, you know, be able to spot it, cuz you can already spot it, or to give you special prompts so that you can, you know, use them and, uh, make AI better sounding when it generates stuff from you or. Excuse me, for you.
[00:01:52] No, I literally just want to give you words, but for those immediately recognizable AI patterns. AI writing patterns. Why? Because having words for things makes my personal brain feel better. Kind of like when you're like, what is that actor's name? Or like, what movie? That shit happens to me all the time. You're like, what movie was that person in? And you're like, from 1983. And then like, you go on IMDb and you see it, like it just feels good in your brain.
[00:02:18] So that's what I feel. And I'm hoping that perhaps giving you these words, these names, will make your brain feel like that and, you know, do the same for you. So let's dive in. Eight AI writing patterns that are dead giveaways. Number one. Yeah, we're jumping right in, folks. Number one, that antithesis structure. It's not X, it's Y. Lord Jesus. These drive me crazy. Right? It is, right? That structure, it looks to create a feeling of insight and wisdom without actually ever delivering any actual insight or wisdom. People don't speak like that, right? We do not speak like that in real life, which makes it feel hugely performative. Uh, which it is. Don't do it. Next one. You already know the M dashers. Jesus Christ. The M dashes. And I say this because you can't make it go away, right? I feel bad for anyone who used M M dashes before, and I know that people did. I literally never used one before, ever. I never liked them. I will use the out of parentheses.
[00:03:22] Love me some commas. Not so good with the semicolon, right? It's not really my jam. But I never use an M EM dash. So now to see them everywhere, it's just like, oh, my God. So I feel bad if you use them before because now you, like, can't use them because it was like, you know, the original hallmark of AI AI writing, right? It. It uses that illusion of a, uh, of a punchline that's never going to arrive, right?
[00:03:48] We see him used to, like, interrupt a sentence and into restating what the sentence was already gonna say. It's just like, nonsense. It's performative.
[00:03:55] And it is a hallmark of.
[00:03:59] Of AI writing.
[00:04:01] Number three, parallel structure overload to the max. This is a big one, right? Everything comes in threes. It's balanced, it's perfectly symmetrical, and it is jarring as fuck. Why? Because real writing is uneven or why?
[00:04:18] Because thinking is uneven, right? This is not a poem.
[00:04:22] This is. This is not how. It's not how we speak, it's not how we think. And so when it's written like that, you're just like, dude, a. A robot wrote this.
[00:04:30] Number four, very similar metronomic cadence, right? Meaning it is if we. If we revert back to the unevenness that is naturally human, right? Human writing and speaking, it speeds up, it slows down, it breaks off. Breaks off. Fun fact. If you ever listen to, like, actual music from, like, back in the day, September, that is a good. A good example. The song September.
[00:04:57] If you were to put a metronome to it, you would see that by the end of the song, it is off from that metronome. It gets faster. That's why I labeled this one Metronomic cadence, right? That song gets faster because humans were playing it and you get excited and you move forward with things and like, things come faster.
[00:05:16] AI doesn't do that. It maintains that steady, measured pace throughout. And we hear it in music too, right? If you have electronic music, it's just the same forever and ever and ever. Nothing changes, nothing lingers, nothing goes slower or faster. There is no little to no variation. It reads and feels like a metronome. It. It reads and feels like a robot. Because guess what it is.
[00:05:42] Number five. This one drives me crazy. The illusion of consideration. And these are, in my opinion, the new telltale kid on the block, right? EM dashes came first.
[00:05:52] That was a super easy tell. Then it was that antithesis structure. And now you'll see, and especially like if you're on threads, you'll see these kind of transitional phrases of here's what they don't tell you, or this is what most people get wrong. And it's actually the fucking worst, right? It sets up, uh, a reveal and it delivers nothing.
[00:06:13] Number six. Adjectives stacking. It's giving keyword stuffing, my friends example. Robust scalable solutions. Thoughtful, nuanced approaches.
[00:06:22] The adjectives stuffed the fuck in there. And they are always complementary, meaning they complement each other, like C O, M, P L E, complimentary and additive. And they're always fake. Um, number seven, no risk, no risky sentences. Everything is predictable and fits the pattern, right? It's like when you're watching a poorly written movie and you can guess exactly what the character is going to stay going to say, despite the fact that you've never seen the movie or the TV show before, right? The di. The dialogue in that case is very formulaic. There's no genuine opinions and reactions because this is what you see in the writing that doesn't exist, right? It's formulaic. No genuine, genuine reactions or genuine opinions because there's no human on the other end to generate them, right? This, you already know what the thing is going to say when you're talking to her. Quote, unquote, talking to it, right? When you're writing to Claude, you have a very good sense of what it's going to say. Why, it's because the writing is produced by math instead of. And yeah, I'm gonna get a little woo quote unquote here. It's not coming from the heart, or any heart for that matter.
[00:07:30] All right, last one. Number eight.
[00:07:33] Hollow opens and tidy closes.
[00:07:37] This is the worst. They're all the worst. But this is. You just like, makes you roll your eyes and you're like, jesus Christ, get to the point. Especially with the openings, right?
[00:07:46] Typically, see here, the responses that are generated by AI, they Start with context setting, and they will typically end with, like, a nice little bow every single opening. It's, you know, what we'll call, like, metaphorical robot robotic throat clearing. And it just wastes your time and your tokens. Worth noting, you can give AI your AI instructions to not do this, and it will generally respect those parameters pretty well.
[00:08:12] Um, not fully. We know with any of this, like, when you say no em M dashes, you always have to repeat yourself. And same thing with, like, don't give me any throat clearing or any of these openings. Just give me an answer. I have that in my instructions, and for the most part, it's pretty good. But, like, every now and then it does slip up. But this is not what real writing does, right? Real writing, it can start and stop anywhere, and it's largely dependent on the individual's writing style. I know that for me personally, and this is something that actually I had A.I.
[00:08:47] um, what do you want to call it? Assess my writing.
[00:08:50] Right. I had that. I, uh, spoke about this in the past episode. We're talking about it, having it create a voice anchor, and it highlighted this. And I was like, yes, that is exactly what I do. And this is part of the reason I'm making this episode. Because sometimes, sometimes what's nice with AI is that because it is a pattern recognizer, you can ask it to analyze things and then it'll give you words for things that you do. And you're like, yes, that's it. And you can feel that closure that dopamine hit. And you're like, yes, that's why I made this episode. Because hopefully in saying this, you can, like, feel that closure on things. You get a little dopamine hit, and you're like, yes, like the dopamine hit from. From knowledge acquisition. And you're like, yes, that's it. So for me personally, I end my paragraphs and I end my writing, whatever. It's the email, things like that. When the feeling is resolved, not when the pattern requires a period.
[00:09:40] AI doesn't do that. It's just like, here's the pattern, here's the set, here's the structure, here's the formula, and then it's done. Um, so within those eight patterns, my guess is that you have absolutely seen or felt literally probably all of them. And like I said in the intro, I just want to give you some names and words for them.
[00:10:00] I will spend a few minutes here, a minute or two on as to, you know, answering the question that maybe you have as to. As regarding why AI writes like this, right. Why does it do these things?
[00:10:11] It is because. Insert drumroll.
[00:10:14] Can you hear that?
[00:10:15] It's a computer program that was trained on human writing and now simply does math to generate the answer. We have been talking about this, folks, since episode one, right? Yes.
[00:10:29] Real talking. This is worth noting.
[00:10:31] The human brain is also doing this. Right. We're trained on other people's work. Absolutely. And we are doing math and mathematical computations at lightning speed.
[00:10:44] Right. At a much higher level in order to produce an answer and to produce what we're going to say. Absolutely. But we also have free will, we have sentience, we are self aware, and we have a soul.
[00:10:57] Right.
[00:10:58] This whole discussion about sentience and being self aware, this is something that I've spoke. I spoke about this in one of my very early episodes and it's definitely a topic that I want to dive deeper into. Any more fucking time know I can't even change the name of this podcast. I need more time.
[00:11:13] But it's exciting to me and it's going because it's like the confluence of like multiple disciplines, Right. To really think about this, it's leaning into fields like cognitive science and philosophy and neuroscience and psychology, and most notably, hearing from brown people, black and brown people and women. Because all we got right now is like white dudes primarily. You got Asian dudes in there too. Right. But it's still dudes.
[00:11:42] And that's an incomplete discussion, but there's definitely something. There's so much there and I want to dive into it.
[00:11:51] Uh, but right now I don't have time for that. Right?
[00:11:55] So just to take it back a step, to answer the question of why AI writes like this, it's because it was trained, yes, largely illegally, on, um, basically all the text and writing that exists on the Internet. And those patterns that we previously noted, those eight patterns, they appear a lot and typically in a positive light. So AI reproduces these patterns, right? Statistically speaking, these patterns look like good writing. If you're sitting there and you're like, what the writing is it pulling from? Because I don't recall reading any of that nonsense and reading, you know, writing that sounds like this. You're right.
[00:12:33] You are right. And uh, you're right because your reality, AKA your good taste, does not speak to everything that exists on the Internet, right? AI is trained on everything, not just the actual good stuff. So think about all of the whack ass self help Instagram accounts out there with millions of followers. Yes, it is trained on all of that and all of Those captions.
[00:12:57] All right, so statistically speaking, there's a high frequency, uh, of those things. And so it notes that and then it spits that back out.
[00:13:07] Additionally, it's those statistics and that subsequent probability. Remember, LLMs are probabilistic models. It's choosing the one token at a token at a time, the token that has the highest probability of being correct, AKA based on what it was trained on and what it saw the most of. And because of that, that is what prevents AI from writing, you know, these, these risky sentences. That was number seven that I wrote. No risky sentences, Right. That everything is predictable and fits the pattern.
[00:13:39] You're not going to see a quote, unquote risky sentence because it would be statistic, like a statistical anomaly to do so to create that. Right. I know that I'm speaking a lot of math here, so if you've tuned out, that's okay. If you're with me, love it. Um, the next point here, the last point here is, is there anything you can do about this? Right? Is there anything you can do to have your preferred AI sound less like AI when it generates writing? And y', all, I did a whole episode and it said, you know, AI will never write like you. And I think that's a good thing. Right. But I want to circle back to that within the context of this episode, since we've outlined eight specific things and a specific patterns that we see. So is there anything you can do to have it less. Do less of that? Yes and no. Right? Uh, yeah. You can prompt it. And this is what we talked about in every episode. You can prompt it, you can give it your writing samples. It's the best thing you can do is, if you want it to sound like you, give it more writing samples. You can provide instructions to avoid certain patterns. And now, you know, I've identified those patterns for you, but you are always going to be fighting against its underlying tendency to want to do these things.
[00:14:35] This is why, yes, editing. If it's generating things for you, editing is always helpful. But honestly, your best bet is just you simply write that shit yourself.
[00:14:45] Right. And. And for what it's worth, like, I understand the appeal of having AI write things for you. I get it. Particularly at times when you're just like, I just want this thing be done. I just need another piece of paper. I get it. Yes.
[00:14:57] I'm not above this.
[00:15:00] There's also that slot machine appeal and that dopamine hit of what might this thing generate. It's fun, it's exciting. I get that.
[00:15:08] Right. But flip side of that is that this is why if you have to correct it enough times and you get mad and you're like, you what? This is not that hard. Eventually you just say it and you're like, I'll do it myself, right? M. Why? It's very much akin to, you know, you're, uh, you're in Vegas and you continue to lose at the specific slot machine and you're like, fuck it, I'm getting up. I'm fed up. I'm walking away because I know what's going to happen next. This next pull is going to be something terrible that I don't even want. Same thing for AI. You're like, this next thing that it generates, it's going to be something terrible that I don't even want. I'm wasting my time in Vegas. I'm wasting my money. I'm getting up, I'm leaving, right? So again, folks, my goal in sharing these eight telltale, dead giveaway AI writing patterns was simply to give you words for what you're seeing, right? If you want to use a list to audit your own AI generated content, so be it, right? But I think that perhaps this list might show you that there is a lot that needs to be edited when you're having AI write for you. So you might as well just write it yourself. All right, let's hit the final section and. Sounds like. Let's hit the. Dougie, let's hit the final section and then we will wrap this up. How I used Claude this week. So if you're not familiar new to the show, welcome each episode I share a quick example of how I used AI that week. This week, I used it to vary all capital letters very quickly, fix the background color and the link color for a page on my other website that doesn't need to change names, right? For my Movement Maestro website, I have talked many times about using AI to generate code. Um, so that's not really what I want to highlight. What I want to highlight is that in this instance, I knew exactly what to ask for. I knew that I needed css and I knew already how to inspect the page and to be able to target the element that I needed to change the link. Um, color for. I didn't want to change for the whole entire page. I just wanted to change for one section and that's. You just need to target an element. I am pointing this out because that is what happens.
[00:17:08] Because what. What's happened over here, rather, is that over time, I have used AI to learn about web development. In this case more specifically front end development.
[00:17:19] Right? So yes, folks can throw out there that AI makes you lazy or makes you dumb or makes you dumber. And I'm not gonna argue with them because I don't argue with anybody. But I will use my own platform to stand 10 toes down on the fact that AI can also Absolutely, absolutely help you learn things you very likely otherwise wouldn't have learned. All right? That is fucking dope to me, all right. Looking at the time, that is all. And also looking at my outline and that is all that I got for you. Hopefully you found this episode helpful if you did consider leaving a rating or a review. I love. I love reading them. I really do. Don't forget I have a companion newsletter, the Curious Companion, that drops every Thursday. That is basically the podcast episode in text format. So if you prefer to read or you just want a written record of the things, join the newsletter fam. You can head to chatgptcurious.com forward/newsletter for now or check out the link in the show Notes. As always, endlessly, endlessly one more time. Endlessly appreciative for every single one of you. Until we chat again next Thursday, stay. Mhm. Curious.
