I Can Spot AI Writing Instantly - Bypass ChatGPT Detectors for FREE

Think you can spot AI writing? I can instantly. And by the end of this video, you'll be able to spot it, too.

Oh, and by the way, I'll also show you how to fix it so your work will never get flagged by AI detectors ever again. Let's check it out. So, the first thing we need to think about is my presentation.

AI detection bypass. How to spot AI writing and pass AI detection. I love a good presentation.

And this is where it starts. So, when you generate AI text, what happens? The AI spits out some stuff, but what is it really spitting out?

It's really spitting out the most likely form of content that you want. And because of that, and because it's just like a prediction machine, there's algorithmic tells that AI detectors look for. So, it's stuff like this.

This is the algorithmic cues, first of all. Oh, I'm going to use my little pointer. Let's turn on the pen.

So, first of all, is perplexity. So, how confused is the AI detector by your content? Is it looking at and going, "Oh, I didn't expect that.

" If it's not expecting a particular phrase or sentence, clearly it hasn't been generated based on probability. So, increased perplexity is a win. Um, and then we've got burstiness, the variation in length and structure.

Now, AI is very formulaic in the way it writes, and we'll have a look at some examples throughout this video. So we need a lot of variation in terms of the sentence structure, the sentence length, the sort of rhythm that's inside the writing. That's what it's looking for.

So we want to increase this so-called burstiness. Otherwise, the AI detector is going to be like this is AI. It sounds like a robot essentially sentence opening.

So um AI detectors know how AI writes and it is very lazy when it comes to writing. It uses a lot of the same sentence structures. So here it's got like this study it is important in conclusion therefore nonetheless these patterns it knows and it's looking for.

So we want to make sure our writing doesn't include that stuff. And the last thing is syntactic repetition. So same length sentences repeated structures symmetrical rhythms.

This uniformity in writing is detected by AI detector algorithms. And we want to sort of make sure that our writing, even if we write it ourselves and we happen to have some of these things written like an AI, we want to make sure that we are not submitting stuff that could be flagged with AI detection. Okay, let's now have a look at some other things that are not algorithmic based, but just sort of like as you read it, you go, "Oh, yeah, that's clearly AI.

" Here's the first one. So, these are the red flags in syntax and style. We've talked about uniform sentence structure.

So, if we're rewriting stuff, we want to make sure that we've got short sentences, long sentences, and everything in between. We want a nice kind of like rhythm to our uh writing just like we would be talking or just like we would write naturally. Short sentences, long one.

Give the sentence as long as it needs to be, not just like, okay, this is a long sentence. The next one needs to be long. The next one needs to be the same length.

That is an AI tell. And then we've got here lack of variation, identical sense and starters, and then no tangents or flow breaks. This is important.

AI stays very much on task. Whereas when you're talking or writing in an academic sense, sometimes you're pulling from information over here that maybe is a little bit of a tangent, but is somewhat related and we're just constantly pulling as we go. AI doesn't pull very much.

It stays in its lane. So that is a red flag. If I see just sort of like completely on topic, no variation in a long bit of text, then I think to myself, okay, this is clearly AI.

Then we've got language and tone, overly polished. Now, I am not a good writer. I've never been a good writer.

So, I always have mistakes, grammatical errors. In fact, one of my supervisors would look at my writing and put red all in the margins and be like, uh, don't do this, don't do that. I had no idea what they meant because I'm not like academically good at writing.

And so I would just go, "Okay, yeah, how would you suggest I change it? " They used all these kind of really sort of like detailed linguistic terms. I was like, "I don't know what that means.

Uh, we'll just change it until they stop complaining. " So yeah, overly polished. That is a sign of AI.

No hedging. In academic writing, you often and should always be using this hedging. like it could possibly, it appears to, it may.

All of that kind of like academic hedging is important because a lot of the times what we're talking about and presenting isn't set in stone. AI doesn't do that. It doesn't hedge.

It just sort of like plows through of like here's the information. And then generic vocabulary, frequent use of terms like important, robust, enhanced. We don't use these very much in our day-to-day language.

The thing is in academic writing we tend to write you know quite um generic vocabulary because that is just the way academic stuff is presented. So sometimes we're in this gray zone where we're like oh we need to use this word but that's exactly what AI would use. Oh no I hope it doesn't get flagged.

But don't worry it's not looking for sort of like individual things. It's looking for the general flow and structure of the entire piece of work. So we can get around that in a number of different ways.

So stay around. Then we've got content and augmentation. Surface level claims.

If your academic writing just sort of like is wishy-washy is a big old wishwash of just like maybe this. Oh, and this and then doesn't really pull from the depths of research. Then it's no good to us.

Predictable phrasing, overuse of triplets. This is one thing you'll see over and over again in the academic AI writing generation game, which is triplets. They will always have three examples.

Mainly because to our human ears, triplets sound really lovely. This and this and this. Ooh, if I go this and this, you're like, "Oh, where's the last one?

" And AI knows this. And so it always gives us triplets. We'll have a look at him in a minute.

And the last thing is this missing authoral voice, which is no personal insights, uh, crit critiques or nuance stance. It is very rare in the academic world that we come across something where we are just like yes I will die on this hill. In reality there's always like a nuance to it and that is quite often ignored by AI.

So I generated some text to see what would happen. This is what I generated. I just said give me a paragraph that could be suitable for the beginning of a peer-reviewed paper on OPV devices.

So, this is what I ended up with. And we're going to play AI Bingo. [Music] Yeah.

Okay. AI bingo. Here we go.

So, this is the text. And these are the things I'm looking for in this text. So, the first one, rhythm.

What does it sound like to read? Organic photovoltaic devices have emerged as a promising class of solar energy technologies due to their lightweight nature, mechanical flexibility, and compatibility with, oh my, there is no rhythm. It's like a machine gun of words.

More words. It's long sentences. The rhythm.

Okay, it's definitely AI from the rhythm sense. And then even if we look down here, we can see in this context new strategies that enable precise control over the vertical phase distribution in oh machine gun of words. So yes, there's no rhythm to it.

So it's AI rule of three. Let's have a look at that. Okay, let's look at where they put some commas.

Uh, due to their lightweight nature, one, mechanical flexibility, two, and compatibility with solutionbased. Oh my god, they're doubling up. So, there's three.

And there's another three with solutionbased low temperature and fabrication process. Oh my god, rule of three. Yes, there's a rule of three in here.

And you can even see down here, despite this progress, challenges and long-term operational stability, scalability, and morphological. There's our three. There's a rule of three.

They love it, love it, love it. So, yeah, we've got that lack of variation. Well, we know you've got that.

We've read it. It's just this machine gun of words. There's no variation in the tone, in the structure, in anything.

Sentence length. Okay, let's have a look. This sentence goes from here to here.

That's quite a long sentence, isn't it? And then the next one goes from that spot all the way down to here. So, what's that?

One, two, three. One, two, three. Okay.

Okay. And the next one goes all the way down here. Okay.

That's one, two, three lines. And then the next one goes all the way down here. That's one.

Okay. There is no sentence variation in terms of length. So yeah, that's AI surface level depth.

Well, I've read through it and I can tell you, yeah, it's there's no depth in it at all from an academic standpoint. No hesitation or nuance. Okay.

Yeah, it's telling me addressing these limitations requires a deeper understanding of blah blah blah. And it's like it's not saying and but yeah. So we've got that no hesitation or nuance from an academic perspective which means also we got no hedging.

There's definitely no tangents. M dash. Oh the formatting thing is I think in my experience less important.

So an M dash is one of those long dashes. We got all the little kind of ones in here but the M dash is something that AI loves to use. We actually don't have it in there.

So no I didn't win. I didn't get AI bingo in this one cuz it's missing this one. AI generated text.

AIS also love bullet points. They also love M dashes. They also love like short paragraphs.

So, this is something to look for. But no, I haven't got bingo. Oh, that's a shame, isn't it?

Oh, do it for your bit of text and see if you get AI bingo and let me know in the comments. Bingo. You win something.

I don't know what I don't know what you win. All right. So the question is now that we know all this, now that we can look at a bit of AI text and be like, "Yes, that is AI generate.

I know about it because I can see all of these things in there. What can we do about it? " Well, I actually put it through a tool called originality.

Not sponsored, just one of the best ones that I've tested in the past. And you can see that likely AI, 100% confidence. Yeah, I am not surprised based on our AI bingo.

I'm not surprised by that at all. But the question is, how do we get around it? Well, there are a few things you can do.

Firstly, you can manually go through and be like, okay, I need to change up the rhythm. I need a rule of three. I need to bust that.

I need to add variation to the length. I need to add nuance or hesitation about what I'm saying. I need to add more depth.

I need to vary the sentence length. Okay? All of these things you could go through manually, sentence by sentence, and change it.

And if you're in the early stages of your career, you probably should do that so you understand what good academic writing is. This is also a great way to train you to become a better writer. But if you're short of time and you don't want any of that kind of like uh manual interference with the text, you can come up with a prompt.

I generated this prompt and stay around because you're going to be amazed at how well this works because here um I just generated this text based on what I know about AI generated text and how it needs to be sort of manipulated to be turned into a more human sort of way of communicating. I just sort of like created this. I have a block of AI generated text I want to revise.

Please improve the following. Uh please improve it. That's better in it.

Please improve it using the following guidelines. So you can see we got sentence structure, hesitation, nuance, critique. Um, avoid using like delve, robust, innovative, like generic AI vocab.

Use more natural transitions. Replace vague examples with specific or realistic details. Adjust tone.

If relevant, include first person voice or light anecdotal tone. Break repetition of how sentences begin and preserve the original meaning and context. but improve the naturalness and authenticity.

And then you paste the AI text underneath. And drum roll please. This is what we got.

And you can see it's much longer, but if we read it, you can see that it's not great, but it's better than the other one. So, organic photovotic devices have attracted attention for years. That is a pretty broad uh childlike way of starting a peer-reviewed paper.

um largely because of their potential to offer something that traditional silicon. Okay, so that sounds more like um more human I guess because this is how I would talk to someone about this. And then we've got here the idea of producing solar films with roller roll methods such as newspapers is compelling especially for application.

So you can see it's got a little bit more kind of nuance, a little bit more variation. It's not just this machine gun of words that's just like fired into your face as you're reading it. So, now you're asking yourself, Andy, you've produced this text, but how does it stand up to scrutiny by an AI detector?

Well, 100% original. So, this is the thing. You can use AI to generate the text.

And this was chat GPT40, by the way. And uh then you can just use that simple prompt, which I'll put in the description. And then you get this, which is an 100% confidence that this is original work.

It's not original work. It's AI. But you can see that uh yeah, it's passed the uh detection really well.

Really, really well. So, obviously, you now need to look at this from an academic standpoint and be like, does it have depth? Does it really sort of like represent the field or what I want to say accurately?

That is where you come into it and where AI cannot help you um in terms of molding this text. You know, other tools like consensus, like elillicit, like size can sort of like help move you down that journey much better than a general large language model. But this is never been easier to get around AI detection.

And uh yeah, that's that's crazy. So I did it with another thing because I was like, surely this can't be this good and true. So, I took something that I wrote in 2024 with AI and it came up with 100% confidence that it was likely AI.

This was still using um originality or I forget what it's called. Anyway, the detector thing. That's good.

Um and then I put it through using exactly the same prompt to see if I could humanize it. And here we are. This is what it generated.

And it was 100% confidence that it was originally uh it was likely original. So, uh, I need to go through this and check, you know, to see if it actually is good from an academic standpoint, but from a passing AI detection, it's a massive win, isn't it? Isn't it?

Use this power wisely. Don't try and like, you know, do yourself out of any learning that needs to be done. But if you're in a situation where you just can't get past AI detection and it's frustrating, you've done everything you can manually, try something like this because universities are cracking down.

The problem is is the universities are doing it I don't know in like a h a really haphazard way where they're like punishing people that are writing original stuff but it doesn't pass AI detection because they automatically write as if they you know with less burstiness or with um decreased perplexity and maybe they put in like sentence starters that are typically used in academia in their field over and over again. So, you can accidentally get flagged as AI detection. And this is how you can get around it by either using AI or doing it manually.

There we are. Let me know in the comments what you think. If you like this video, go check out this one where I talk about AI bypass detection using other tools.

I think you'll love it.