Speaker 1: As AI tools become more powerful you may be tempted to use them just to generate a literature review. You may be tempted just to generate a little bit of text for that assignment or that peer-reviewed paper you're writing but be careful because you can easily detect AI content and I'm going to show you all of the ways people have tried to get around AI detection and plagiarism but only one of them now works. It is a game of cat and mouse and at the moment the AI detection tools are winning. None of the older stuff seems to work and this is what I mean. So I asked ChatGPT to generate 500 words on the current state of the field of organic photovoltaic devices and this is what it did and I must admit it did a pretty good job and so what I've done is I've used two different ways of determining the plagiarism and the originality and AI detection. So the first one is Unicheck and I look for similarity and plagiarism and the second one is originality.ai where I've used it to look for AI detection and this is what you can see. This is the first thing to come out of ChatGPT. It's got 100% AI detection and so what I've been doing is putting it into this table. This is AI generation GPT-4. You can see that Unicheck gave it 3.85 and originality saw it as 100% AI. That probably doesn't make sense. Originality AI detection. There we are. That's a bit better isn't it? That's a little bit less ambiguous. The first way I've seen people say you can get around AI detection is by using synonyms and also asking it to retain domain specific details. So that's exactly what I did. I went over here and I said hey rewrite this with blah blah blah and retain domain specific words. So I posted that into originality. I put it into uncheck and guess what? It came back still as 100% detection and a Unicheck plagiarism was even slightly higher. So that's not a good start. The second way is using a tone. They're saying that you can change the way that AI detection works by asking the AI to generate it in a tone. So I asked it to rewrite it in the tone of Albert Einstein. Now you can't really use this for anything because it's a little bit weird but I put it into Unicheck. I put it into originality and you can see that we ended up with 100% AI and also 0% plagiarism because Albert Einstein definitely didn't write this. You can see here with Albert Einstein if I go to results it's 100% AI. So not a good start. People have been saying that you can use power phrasing tools such as Quillbot. I did that. I put the same thing into Quillbot. I did it on the free version. I did it paragraph by paragraph and ultimately you can see that we ended up with 100% AI detection and 0% plagiarism which isn't too bad. Okay, manual power phrasing. Surely this is where it should all change. If I go in and actually just go through and change each individual sentence to reflect how I would say it, surely that should help. So that's what I did. I went through painstakingly and changed all of the sentences power phrasing to retain the original meaning but then adjusting it to just my own way of writing and I'm going to reveal to you now that I did that and we ended up with a 97% AI score. So apparently I am 97% AI and I think it just goes to show how advanced these things are becoming and how hard it is now to go from AI generated content to change it to make it your own. So really you should be starting if you want to get around these just by writing it out on your own. That seems to be the only way but there is a little one. Stay to the end of this video because I'll show you the way that is currently working and it is pretty good. Other people are also saying re-sequencing so taking the information that you've already got and just sort of like changing it up. So that's what I did. I put it into Unicheck and originality AI detection and guess what? 100%. So just changing around the paragraphs and the sentences and the information no longer works. Other people have been saying what about putting more details in the prompt. I went over to ChatGPT and I said hey acting as a scientist please write 500 words but make sure you include information about fullerene blends and encapsulation issues, work from Stapleton et al, that's me, so I should be in here and include references where possible. I thought maybe that would help so I put it in and it did an okay job although it didn't get my study right. I've never done anything with encapsulation so never mind it's still in there. Nonetheless it's factually incorrect but does it pass the plagiarism and AI detection? So I put it in and take a guess. Doesn't work. It does not work. Putting more details in the prompt and sort of like asking it to do certain things doesn't help with its AI detection so that's 100%. And a last little thing I want to talk about is this idea of perplexity and burstiness and that's just something that humans tend to use in their language when they're writing and it just makes it less robotic I guess so I did that with the last one. I said please increase perplexity and burstiness in this response. This is what it's done. I didn't quite like what it wrote but nonetheless it could be a way of getting around plagiarism and did it? No. Well it did 1% apparently it's only 1% efficient so AI detection told me that it was 99% AI generated and that's because it was. It was actually 100% so a little bit of a win who knows and then it got me thinking well actually what can we do if we don't want to write from scratch? What tools are there out there? And I decided to try undetectable.ai so here we've got readability you can select where you want it, you've got your purpose, you can select what the purpose is and then you post your words in and I did pay for this and this is what I ended up with. So I put all of this in down in the original submission and this is what it's given me and so I was really tempted. It's saying it's going to pass all of these different output checks but does it really? So I put it into originality.ai and let's go to it undetectable.ai here we are big reveal. Does it actually work? It does. This is the only thing despite going through all of the different ways people have told me to get around AI, this is the only tool that actually works at the moment. Quillbot no longer works, paraphrasing, changing things around, manual review. If your text is originally AI generated it can be detected apparently and using a tool like this undetectable.ai seems to be the only way to get around it. So I have got a link to that in the description and I'd be interested to know whether or not that actually works for you and if I go here and we can reveal what it goes, there we are, it says 29% AI detection and only 2.19% of plagiarism. So I think that currently is the only real way that you can get around AI detection and of course there's loads of reasons you shouldn't be doing this. Clearly you've got to be writing your own stuff from the get-go and using AI only as a tool for trying to sum up your work or to change your work in some way. But I think we've got to feed information in by using our own stuff and then asking the language models to work with that language instead of relying it for the generation side of things. Now I know the world will change quickly because of all these tools and in science that is not going to be an exception. So there are ways of actually starting to own up to using AI and I'd like to thank one of my subscribers who actually sent me this. It's a statement they put at the end of their papers which says that GPT-4 has been used and I think we'll see this more and more. So here we can see we've got a conflict of interest statement which is like no, no conflict of interest but there's this extra statement now at the end which is the LLM statement. As the author is not a native British speaker he utilised the GPT-4 model as an editing tool to do these things and that it did not correct graphs and mistakes or linguistic inconsistencies by producing plagiarised wording. The manuscript has been submitted to three different checkers. The result showed a similarity score of below blah blah blah. So I think we'll be seeing this sort of stuff much more in the peer-reviewed literature. I think we'll be seeing it in theses as well and as long as you kind of also say this is what I've used, this is how I've used it and these are the results I got from it, I don't see a reason why this shouldn't be part of a normal academic workflow. So there we have it. There's the only way you can get around AI detection right now. Let me know in the comments what you've tried and if it works in other ways. I'd love to know it but also remember that this isn't a game that you should be playing at all. You should be generating your own information and using large language models to help you edit, to help you check for inconsistencies, make things more concise. Use it as a tool rather than a generation device and I don't think you can go far wrong. But undetectable.ai is currently the only way that I found that you can actually get around originality's AI detection. That's really interesting and I think it will continue to evolve rapidly over the next few months. So let me know in the comments what you think but also head over to academiainsider.com. That's my project where I've got my two e-books, the Ultimate Academic Writing Toolkit as well as the PhD Survival Guide. I've also got the blog as well, a forum and a resource pack if you're actually applying for a PhD or grad school. And also remember to sign up to my newsletter. Head over to andrewstoughton.com.au forward slash newsletter. The link is in the description and when you sign up you'll get five emails over about two weeks. Everything from the tools I've used, the podcast I've been on, how to write the perfect abstract and more. It's exclusive content available for free so go sign up now. Alright then, I'll see you in the next video.