Why Viral AI Research ‘Hacks’ Often Fail Students (Full Transcript)

[00:00:00] Speaker 1: My 58 page chat GPT prompt, my prompt is 10 times bigger than your entire paper.

[00:00:08] Speaker 2: There is no plagiarism and there is no AI score.

[00:00:12] Speaker 1: All right guys, um, this, this is just, uh, this is going too far.

[00:00:17] Speaker 2: This might have worked then, but it won't work anymore.

[00:00:20] Speaker 1: Taking academic advice from a masked figure, you might need to check your priorities in life. You're kind of exposing yourself for the rest of your life to risk if somebody for some reason decides they have a bone to pick with you and come after you. I'm starting to despair at the state of humanity right now. Hey, Professor Stuckler here. So I've been seeing a lot of viral AI hack videos going around and some of them look a bit dubious, shall we say? So what I wanted to do today was take a look at them, just raw, unedited, how I see them through my eye, having been a professor for several decades at institutions like Harvard, Oxford, and Cambridge, and I'll just give you my no fluff, no BS response. One kind of caveat here, our goal at FastTrack and on this channel, if you're new here, is to just give you the honest unfiltered advice that you can use to stay on the side of angels when doing your research. So I'm going to not aim to critique the people, but the ideas. I haven't looked at some of these in depth. I've just glanced and saw that there were viral hacks. So you're just getting my raw, unedited responses here. So let's go. Okay, so this one is saying it's going to do a research paper in hours with ChatGPT. Let's take a look.

[00:01:35] Speaker 3: Are you ready to write a full academic research paper in just one day using AI? In this video, I'll show you how to use my 58-page ChatGPT prompt.

[00:01:47] Speaker 1: My 58-page ChatGPT prompt. My prompt is 10 times bigger than your entire paper. Okay, let's see where this is going.

[00:01:57] Speaker 3: A fully automated framework designed to help students and researchers write research proposals, literature reviews, methodology, data analysis, and... I'm just kind of curious.

[00:02:10] Speaker 1: Let me look at the comments. I'm curious what people are saying. I want to write research article. I want to write. I want to write my research project. I want to write my thesis. Okay, I wonder if these are bots.

[00:02:20] Speaker 3: Okay. Even references all in just a few hours. Whether you're working on an academic research...

[00:02:26] Speaker 1: Okay, look, I'm fast. I can write a paper in a day. I've actually got a video where I show a workflow where I had a person deadline and had to do it. But a few hours? Right, this is already red flags. If something's going to write an entire full research paper in a few hours, that's a massive red flag.

[00:02:46] Speaker 3: ...research project, thesis, dissertation, or journal article. This prompt works seamlessly with ChatGPT, Claude, Gemini, and other AI tools. I'll walk you through every chapter from intro...

[00:02:58] Speaker 1: Okay, this is not actually saying where it is. Let's see if we can find this. I want to see what this actually is. So there must be in the description somewhere. Okay, I see there in the reply, they're linking to a website. Let's take a look at this website and see what these guys are actually doing. Okay, so they're using an old GPT. They recommend GPT-5 rather than GPT-2. Okay, so they want you to get... I see. They want you to get membership to their program. Okay, so they want you to look at this My Thesis GPT, it looks like. So let's see what they're actually trying to do here. Okay, so they want a custom GPT. Guys, I wouldn't be buying a custom GPT. Firstly, we actually offer one for free that's trained on all of our data and research. They're wanting you to pay for this. They're wanting $10 to get research humanizers, plagiarism shield, and all this stuff. Guys, I'm not going to go through and actually test this. But I don't see anything validated of how it's trained its data, or see a sign of why this is going to be useful or anything beyond just a regular GPT. And just if your selling point is that you have a 58-page prompt, that doesn't necessarily correlate to quality. In fact, sometimes longer prompts lead to lower quality. So this one's definitely going to be a fail. All right, let's go to the next one. Okay, now we've got the Turnitin Bandit, who's going to give us some AI advice. Okay, this one's going to be fun. Just as a side note, if you're taking academic advice from a masked figure, you might need to check your priorities in life.

[00:04:46] Speaker 4: Every single one of these AI detectors are completely useless because 69% of universities use Turnitin.

[00:04:54] Speaker 1: But luckily for you... That's true, that is true. Turnitin is kind of the main standard used.

[00:04:59] Speaker 4: I made a website where you can check your work before your professor using Turnitin.

[00:05:04] Speaker 1: And if you don't... It's not a bad idea to check your work with Turnitin. Just make sure it doesn't deposit it inside Turnitin because then it will show up and look like you plagiarized yourself.

[00:05:14] Speaker 4: Believe me, just look at this and this. And keep in mind, any document you upload is completely anonymous. Thank me later.

[00:05:21] Speaker 1: Okay, well, look. Okay, Bandit aside, being able to check your work for plagiarism is not a bad idea, especially if you have been drawing on work online. Maybe you've been doing some copy pasting and paraphrasing. We do recommend running through a plagiarism detector. So this actually, despite this guy being an AI bandit, doesn't get a fail. So one thing to note, these AI detectors have a ton of false positives. I actually loaded in a paper I wrote in 2017 before the AI age fell upon us. And it actually marked a high proportion as AI written, when that was just completely implausible. So it sometimes does mistake good writing for a high portion of AI. Just bear that caveat in mind. Also, these detectors are evolving rapidly, but I suspect for a while Turnitin will remain the gold standard of the not so good AI detectors out there. All right, let's go to the next one. Okay, simple steps to beat Turnitin AI percentage manually. Okay, I can see I had scrolled forward in this one initially scanned.

[00:06:30] Speaker 5: But let's take a look. Okay, I guess all these should have been 100%.

[00:07:06] Speaker 1: Okay, yeah, I think this is some of the viral AI hacks people are searching for to try to, okay, I had AI do everything, but maybe in a second step, I can humanize things. Let's see.

[00:07:22] Speaker 5: GPT 0 became 0%, GPT 0, all right. Number two, zero GPT became 9.78%. Number three.

[00:07:34] Speaker 1: I mean, this precision is funny, 0.78%. Good thing it was not 9.79%, but okay.

[00:07:41] Speaker 5: Wheelbot became 24%, 24, number four, Turnitin AI shown 81%, 81%. After the recent Turnitin update on August 27, this is the article, all humanizing tools became utter total rubbish.

[00:08:04] Speaker 1: This is a good point. And this is one of the worrisome things that in a video I did here, I looked at how Harvard's president got accused of plagiarism, and it came back to haunt her 30 years later. The point is, if you do some dodgy AI stuff now with your essays, and the detectors evolve, and maybe even if they can't detect something you did now that was on the wrong border of the ethical line, they could in the future. And you're kind of exposing yourself for the rest of your life to risk if somebody, for some reason, decides they have a bone to pick with you and come after you. Let's see where this is going.

[00:08:41] Speaker 5: Rubbish. Since humanizing is not working, AI generators can be detected, humanizing also can be detected. What is the only solution? Human-based paraphrasing.

[00:08:52] Speaker 1: Oh, so you mean like learning actually how to write properly. But I think what he's doing is still saying, take a full AI document and let's humanize it, but just do it as a human, not as a human AI tool.

[00:09:07] Speaker 5: This is the tracked file after the humans paraphrased, rewrite the whole thing.

[00:09:15] Speaker 1: Okay, I see what's going on. So they are taking, that's what they're doing. I'm going to fast forward a bit and see how they're actually doing it, because I think they discussed this a bit in here. I think I have seen this somewhere before. One of those things when you're scrolling around at two in the morning and you should be sleeping. Maybe let's go, okay, here, this looks like a segment.

[00:09:35] Speaker 5: That had high 2018 AI percentage. So now let's take this particular part from the problem statement section. Technique number one, sentence total reconstruction. So let's understand the sentence first. There is literature, right? Then cognitive impairment, breast cancer, and specifically cancer and chemotherapy. Now, all these parts, all these keywords cannot be changed. Cannot be changed. So now let's reconstruct a new sentence. First thing, delete the whole sentence because we are going to reconstruct everything, but leaving the keywords. Now we paraphrase. As reported in the past studies, cancer survivors, particularly breast cancer patients, experienced cognitive impairment as a side effect of chemotherapy process. Look at the paraphrased version now. Across the literature became as reported.

[00:10:28] Speaker 1: So yeah, you can change up nouns and shuffle things around, subject, object in the sentence. That is kind of standard paraphrasing. We've got a whole video on how to paraphrase in the right way here in case you want to check that out. Just be very careful that you haven't lost the precise scientific meaning when you do this because even in the sentence, there are some subtle differences that can come out and scientists are very, very careful with their word selection. But make sure you don't lose the citation as well in the process. Okay, let's keep going here.

[00:11:05] Speaker 5: In the past studies, breast cancer survivors were brought forward and expanded into cancer survivors, particularly breast cancer patients. If you realize there, I'm not removing the context.

[00:11:18] Speaker 1: Yeah, so this is an example where there is some important subtle nuance that can get lost here and you don't want to do it in this way because if the research is being very specific about a narrow subset cognitive impairment among breast cancer survivors, you actually are changing this meaning to say cancer survivors, in particular, this subset. That is actually, it's subtle, but it's a different meaning. So that's what you have to be careful with and I don't... Human paraphrasing, this is the right way to go, but you shouldn't do it on a fully AI written text ever. But taking out some snippets, if you're going to say, here's a citation, I don't want to verbatim have that citation and I want to paraphrase it a bit, that is okay. You just, this is actually an example of where you could lose important meaning and precision. Well, I want to see other things they do.

[00:12:11] Speaker 5: Only elaborating the context. Okay, don't change the word breast cancer patients. That will change the entirety of the context. This is what humanizing tools will screw your life. This is what they will do. Be careful.

[00:12:23] Speaker 1: That is also very true. The humanizing tools do not get this right at all. So stay away from them.

[00:12:28] Speaker 5: So human, we don't do this. We paraphrase very carefully. Next, linked to chemotherapy, but totally changed to as a side effect of chemotherapy process. This way turning in AI cannot detect at all because the entire origin had been broken down. We have broken the whole thing down. Technique number two, active to passive sentencing.

[00:12:52] Speaker 1: Whoa, whoa. Okay, wait, active to passive sentences. So it's like saying active voice is better. It's just clearer overall. The classic example is the pitcher threw the ball. And if you turn that passive, it's the ball was thrown by the pitcher. The second one takes more words. It's harder to understand. It's just less clear. If his solution is you're going to humanize by making writing worse and less clear, you're going down the wrong path. Let me just, let's just watch to make sure I didn't misunderstand.

[00:13:22] Speaker 5: In a review by Skagen and Woffel, 2013, 70% of related studies reported that 17% to 78% of breast cancer survivors treated with chemotherapy, experienced cognitive dysfunction. Now let's paraphrase these two. Let's apply active to passive sentencing. This is the outcome. That's the original. This is the paraphrased version. In Skagen and Woffel's 2013 review, 17% to 78% of chemotherapy treated breast cancer survivors were reported by 70% of the relevant.

[00:13:57] Speaker 1: Maybe this was just a mistake. Where did 17 to 78% of the survivors, okay, something funky must have happened there.

[00:14:06] Speaker 5: These two experienced cognitive dysfunction. You see, we have broken down the whole AI writing. It sounds huge.

[00:14:13] Speaker 1: Okay. And these are actually two cases in trying to do this. Subtle errors have crept in, and that is also part of the danger. And the writing has become, in this case, the loss of clarity switching to passive isn't as bad in this summary. So I'm not saying avoid passive at all times. So spirit is right. Execution, a little bit flawed. Let's go on to the next video. Okay. All right. We just got a couple more. This one's going to be fun. How to bypass, turn it in AI detection every time. I wonder when this was made, seven months ago. So the funny thing here is, like we saw in the earlier video, this might've worked then, but it won't work anymore. But it's a short one. Let's take a look.

[00:14:56] Speaker 4: Hello, everybody. I hope you're doing well. Here's a quick tutorial on how you can bypass turn it in literally every single time using these three tools. And for the sake of convenience, I'm going to open all three tools now so that I don't have to later. So to start, you're going to want to open your Google doc with the AI generated content that you got from ChatGBT or wherever it might be.

[00:15:17] Speaker 1: Starting point. Yeah, you don't want a whole written document, like I said, from ChatGBT, but okay.

[00:15:24] Speaker 4: This entire document is AI generated and it's very obvious if you take a quick look.

[00:15:28] Speaker 1: Yeah, it is very obvious. As a professor, you can snip out right away.

[00:15:33] Speaker 4: With that, we are now going to go to this website right here. It's essentially going to let you check your work before your professor using turn it in, like the exact.

[00:15:41] Speaker 1: I think this might've been our bandit guy from earlier.

[00:15:44] Speaker 4: Exact same algorithm, but in this case, it's completely confidential. So we're going to go ahead and open up the dashboard, click upload a file and upload that file. And just to save you guys some time, I'm going to speed this part up until we get our results. Okay, now that our essay is done uploading, we're now going to click view results and download both the AI and similarity report that it provides.

[00:16:04] Speaker 1: Okay, that was just turn it in checker. So they're saying, okay, it's high AI, I guess. Yeah, okay, not surprising, cool.

[00:16:12] Speaker 4: And we're going to ask, can you please?

[00:16:17] Speaker 1: This just strikes me guys, I don't know if this is actual turn it in or the updated version of turn it in. So do be careful of using that as the gold standard. If it's not actual turn it in, you don't necessarily know what you're getting.

[00:16:31] Speaker 4: Rewrite this essay, so it doesn't seem or flag as AI. And we're going to aim for 700 plus words. And again, for the sake of convenience, I'm going to speed up this process until the essay is generated.

[00:16:56] Speaker 1: All right, it looks like chat to BT is And you're going to spot this right away with the long dashes here. The way the hyphens are here. It's going to be written almost too well. If you're doing this for a course, definitely wouldn't work for publishing. But let's see if he's saying this is good.

[00:17:16] Speaker 4: Is done providing us with an essay here. So what we're going to want to do now, and I'll probably end up speeding this process up too, is we're going to copy this entire essay that it provided. We're going to paste it into our new copy document. And we're going to make some changes. And I'm going to speed this up just so you guys don't have to watch.

[00:17:35] Speaker 1: Okay, cool. So he's doing some human formatting as well, which one of the earlier guys was advocating.

[00:17:50] Speaker 4: All right, now that I'm done editing this document, I'm going to go ahead and download it. And once I'm done downloading it, I'm going to make my way back to the same website.

[00:18:01] Speaker 1: So the principle here was, again, just he tried to humanize it by being a human and take out the GPT signatures. And I guess it's going to go through their churn it in detector and say it's okay. The problem is that this is then not. And you're creating this professional liability for yourselves. This gets a big thumbs down. Don't do this, guys.

[00:18:20] Speaker 4: Make sure I can meet out CX and up. So now what we want to do is download the AI report and download the similar similarity report similar to what we did at the start. And we'll put these side by side, like so. So as you can see, 20%.

[00:18:39] Speaker 1: Okay, yeah, they're trying to say that this is going to work. Yeah, don't recommend this. Okay, okay. So this is how to write a PhD thesis with AI like chat GPT without AI plagiarism. Let's check it out.

[00:18:54] Speaker 2: Hello, intellects. I am glad to have you here. I always like being flattered. Great. Thanks. Here once again. I have written and coached many students through their theses and master's projects, helping them achieve A plus results. So I am going to try helping you too to write an excellent research project. Whether for a PhD, a master's, or even for a first degree with or without AI. I will also show you how to achieve no plagiarism.

[00:19:26] Speaker 1: Okay, using some AI old generation graphics. Let's see how they're doing this with or without AI.

[00:19:34] Speaker 2: My AI score in your projects, despite using AI. I will not only be providing extensive prompts for each section of each chapter, but also show you how to write it.

[00:19:47] Speaker 1: Some people say, okay, so in spirit, this principle is good. I want you to know the fundamentals of how to write each section of a manuscript. They do follow a formula. And once you get that formula, it's actually very fast. You don't need AI at all. So hopefully I'll teach the right way and then see how he enhances it. That's kind of what we advocate is AI enhanced workflows.

[00:20:07] Speaker 2: Say, hey, all you need to have a good AI to write your project like thesis AI. You don't need prompts. The truth is that like I always say, AI can only be as good as the person prompting it. If you don't understand how to write your background or statement of the problem, for example, and you just tell AI to write your chapter one for you, the AI will produce something bland that does not deserve an A grade. So watch.

[00:20:35] Speaker 1: I 100% agree with this point. I mean, it's kind of like you put garbage in, AI is just going to accelerate garbage out and produce more of it. It's our theory of AI error is that it's making it faster to do bad research, easier to do bad research with AI than to do good research. So when people don't have the fundamentals, they start using AI. They're often getting past the early frictions that would normally detect errors along the way. And those errors are showing up much deeper in the process that are later and more structural and lead to a radical surgery to the paper. Where before without an AI free world, errors would hit, they'd get stuck earlier in the process before really going down a rabbit hole. But let's keep going. I want to see what they recommend.

[00:21:23] Speaker 2: This video without skipping any part. Okay, we're probably going to skip some parts. Trust me, you won't regret it. Typically, a research project, be it a systematic review or an empirical study, must have five or six chapters. However, in this...

[00:21:40] Speaker 1: You don't really call in a systematic review. I've never called my components of a systematic review a chapter. Okay.

[00:21:47] Speaker 2: Video, I will focus on the five core chapters. Chapter one, which is typically the introduction. Chapter two, literature review. Chapter three is the methodology section. Chapter four.

[00:22:02] Speaker 1: Okay, we're not going to go through this. This is a little bit confused. Let me just go through. I just want to see where they're recommending to use AI. Think of your study setting stage. Okay, so there's some decent stuff with... It's okay. Explaining the background here. Let's just scan through. Okay, now we're seeing some stuff that looks like a prompt.

[00:22:23] Speaker 2: Humanizer and detector. You can check my channel for some of my videos on Ryan AI. Look at this. Very rich with relevant real stats. Very critical and analytical.

[00:22:39] Speaker 1: Whoa, whoa, whoa. I think he maybe AI generated something. Let me come back.

[00:22:44] Speaker 2: I used the Ryan AI GPT-5 model to write the background using our prompt, sample topic, and study objectives. But you can use chat GPT or any AI you are comfortable with. I love Ryan AI because of its AI humanizer and detector. You can check my channel for some of my videos on Ryan.

[00:23:06] Speaker 1: I glossed past that too quickly. This looks like they're... This is... No, no. You don't want to do this. Let's come back to this. Look at this. Very rich with relevant... Look at this. Very rich. Fake stats. You got to check these citations carefully that they're not made up. People know about this hallucination. That's going on. So first, I just don't recommend AI to do this. This is way too heavy a hand. Way too heavy AI generating, replacing thinking. So at the beginning of this video, they were talking about AI enhanced workflows, knowing the right way to do things. But this is having AI write it for you. Not a good idea.

[00:23:46] Speaker 2: Relevant real stats. Very critical and analytical. Next is the statement of...

[00:23:55] Speaker 1: Analytical. Yeah. Critically and analytically wrong. Don't do this, guys. All right. Let's see if I can find what they're actually prompting. Okay. Chat GPT deep. Okay. Here's a prompt for a literature review. So all they're saying is put in postgraduate dissertation. Put in the research topic. Begin with a short introduction that outlines purpose of the chapter. Organize subheadings. It's got this long prompt. But this prompt isn't even really doing... Purpose of a literature review is to justify what's going to come next in your study. And this isn't doing this. Well, okay. No, I take it back. The screen's glitching here. Clearly identify the research gap your study addresses. 2,000 to 3,500 words for academic dissertation. That's going to be short for a literature review. And how's it going to know what research gap your study addresses? I guess you're going to insert your aims and objectives here. But it kind of puts the cart before the horse in a way. How are you going to know your aims and objectives very effectively if you haven't even done any review of the literature yourself? And so your literature review needs to funnel and narrow down to get to this gap. And so it's saying highlight points of agreement, disagreement, gaps in existing literature. Okay. But these AI tools don't really do a good job of searching existing literature because they don't have access to the complete repertoire of what's out there. They're limited to open access sources. And they can't discriminate the quality of the sources that they're dropping in. So you end up building a, like, you put junk into your literature review, you're going to get junk out. So this, anyway, don't do this. This isn't going to work. I like their original idea of, say, learn the basics of the fundamentals first and then use AI to enhance and accelerate. This is just having AI do it for you. I want to see, how do they think that this is going to be no AI plagiarism? Some of these guys are promising things that they can't deliver on. So beware of these false profits. This stuff will get you into hot water. And we've seen students getting kicked out of universities for exactly stuff like this. I'm just going to scan and see if there's some humanizing step that I'm missing. And I don't see that. And I just want to see in the comments, what do people actually say about this stuff? Really, I found it deeply convincing. This is what worries me. I've emailed my details requesting support. These prompts are helpful. Most helpful 20 minutes of my academic journey. Going to try to, I don't know if these are bots putting their hair. This is going to save me so much time. We do get bots commenting on our channel as well. So yeah, this is, guys, be really wary of stuff like this. The idea of prompts can be good. Once you have your draft and use it as a stress test, ask it instead. Did I miss a study? That could have helped flesh out this point and check if it's a real study. And stress test your document. You can also get it to clean up the grammar, clean up the English. That's perfectly okay. Make sure you master the fundamentals before you start trying to AI enhance, or in this case, do an entire literature review for you. We have a whole other video on how to use AI in the right way to automate and speed up your literature review without crossing over into ethically gray territory. Check it out up here.

[00:27:06] Speaker 2: Ah, we got another one. Prompts for a complete research paper.

[00:27:09] Speaker 1: This seems like this is the shtick of this channel. And these guys are just going to continue to somehow make a complete research paper without AI plagiarism. You've just seen in the previous videos, we just went through that this isn't possible today, right now. And you're exposing yourself to risk in the future. So we'll just briefly glance at this, but not a great idea. I'm starting to wonder because this is faceless as well. So before we had the AI bandit, who wasn't that bad, but now we've got a faceless channel. And again, got a faceless channel giving you advice. What are these guys hiding from?

[00:27:42] Speaker 2: I wrote this research paper from start to finish using AI. It was published in a Q1 journal.

[00:27:49] Speaker 1: Somehow I have some doubts about that because I can already see from the formatting that it's not following the format for journals. But okay, maybe they gave us the citation. Let's look. No, they wouldn't do that because they know that they can't do that or it will get retracted.

[00:28:04] Speaker 2: There is no plagiarism and there is no AI score.

[00:28:08] Speaker 1: All right, guys, this is just, this is going too far.

[00:28:12] Speaker 2: Let's be honest. Anyone can tell an AI to write me a research paper, but what do you get? A bland, shallow, and unpublishable piece that no serious professor or journal will take seriously. The truth is AI can only be as good as the person prompting it. If you don't understand what each... This is the challenge.

[00:28:33] Speaker 1: I mean, some of these channels are mixing in some good points that are true and then some garbage. And if you're just starting out, it might be difficult for you to have the, forgive my language, the shit detector to detect what's okay and what's not okay when they're mixing some things that are okay with things that are clearly not, like this channel.

[00:28:51] Speaker 2: ...section of a research paper really requires, from the abstract to the conclusion, the AI won't either. True. In this video, I'll show you how to write a high quality, fast, and effective research paper the right way using any AI of your choice. You'll get powerful prompts for every section of your...

[00:29:10] Speaker 1: Let's look at the prompts here because this might be similar to what we just saw a moment ago. Okay. Oh, here's the lit review prompt again. And it looks like the same thing. It's the same thing that we just saw. Assess the overall condition. I'm looking down here. Condition of the literature review. Why it does or does not correspond to the aims of the study and how it will address the gap. Again, guys, just to repeat, never have AI do the first draft. That you've already gone off the deep end and you're on the wrong side of the ethical line if you've done that and you are putting yourself at professional risk and journals will not accept it. The actual journal policies, which are right up here, expressly state not to do this. You can use it to do light touch stuff on the grammar, but anything more than that, you'll need to acknowledge. This will not get accepted. And if this paper was real, it would have been retracted as a result.

[00:30:06] Speaker 2: So I'm not gonna waste any more time on this channel.

[00:30:08] Speaker 1: I don't recommend that you guys do either. Let's keep going. Okay, this one's how to write a review paper using AI step-by-step guide. Cool.

[00:30:16] Speaker 6: So you've got to write a review paper and you're thinking, what do I write about? Where do I find all the papers? And how do I even put everything together? If you're asking any of those questions, this video is for you. In the next few minutes, I'm going to break down how to write a proper review paper using AI step-by-step. This will... Okay, cool.

[00:30:34] Speaker 1: A lot of researchers need to write review papers and I don't know if this is systematic review or literature review, let's see.

[00:30:41] Speaker 6: Save you a lot of time and make the whole process way more easier. Let's start with the basics. What is a review paper? A review paper is not about doing new experiments. It's about reading what other researchers have done on a topic and then putting all of it together in a way that makes sense. You're basically telling the story of a topic. What's done, what works, what doesn't work and what's missing. And honestly, this can really help you make a very strong base in your subject.

[00:31:06] Speaker 1: Review papers are a good idea to do. We recommend to almost all of researchers who are just starting out to do a review paper to help you find low-hanging fruit topics, gain quick mastery of your area and a systematic review in particular, we highly advocate as a great starting point. Let's go forward though. I want to see where the AI is coming in. Okay, step one, define your review paper topic.

[00:31:28] Speaker 6: Before you can start writing anything, you need to decide what your review paper is going to be about. Now, here's a mistake that most people make. They pick a topic that's way too broad like AI in healthcare. It sounds very interesting, but honestly, it's way too much. You'll end up finding hundreds of papers and get overwhelmed. Instead, just try to pick something more specific. For example, how AI is being used to detect early signs of depression in young adults. That's much better. It's narrow, clear and easy to work with.

[00:31:59] Speaker 1: We use actually a PICO model to better define topics and that specifies the population, the intervention or exposure, a comparison group, outcome, sometimes a time period or design as optional features. There are tools for narrowing your topic appropriately and then testing, establishing and validating them. I want to see though what they're doing here.

[00:32:18] Speaker 6: Don't know how to narrow it down or you are just blanking on ideas? Open the brainstorm tool on paper pal and use it as an advisor to help you with this step. Just type something like give me review paper ideas on AI and mental health and in a few seconds, it will show you a bunch of directions you can take. You can keep trying different angles until one of them clicks. Once you've got a topic.

[00:32:37] Speaker 1: Okay, so if you're just in an exploratory stage, it's not a bad idea for finding a topic space. Definitely will come up with ideas. You need to validate that there's a real need for your paper. You need to take a step and make sure you're not duplicating something that's already been done and you need to establish that it's feasibility. These are critical validation steps with AI. I've seen too many researchers find a topic from AI and AI sycophantically encourages them to go forward with it and then they come to me later and we discover in just a few minutes of doing those validation tests that it's a dead end. It's already been done before. There's not really any gap that they can fill and they've just lost months of work and I hate having that conversation. So this is kind of like a kind of yellow flag.

[00:33:23] Speaker 6: Yes and no in its use. If that feels right, something you actually want to explore, it's time to move on to the next step. Step two, find the right research papers. Okay, now that you've picked your topic, it's time to find out what's already been written about it. A review paper is all about reading and analyzing existing research. So you'll need to collect a bunch of papers on your topic that are verified and relevant. Now, usually this means going to Google Scholar or different websites, opening 15 tabs and praying at least some of them are useful. It's messy and takes forever. But Paperpal makes this part super easy. Just go to the research tab and type in questions related to your topic. For example, what is the use of AI in diagnosis of depression? Paperpal will instantly show you a list of verified academic papers along with a quick summary answer for this.

[00:34:11] Speaker 1: So the problem is that even the best AI tools right now, and this might get better, Google Scholar AI is a bit better in this space because it has Google Scholar. But these are, they have partial access to the literature. Now, this won't work at all for a systematic review, in which case you need databases like Web of Science or PubMed. Again, if you're doing this as maybe an undergraduate essay for a review or a graduate assignment, this can be okay. What I do like about here is that these are real articles that you can go to, but this will likely be incomplete. And look, I know I'm old school, but I think Google Scholar is fine. Its algorithm is going to, if you've got a subsection of your review, you're searching for papers on, it's going to bring the most relevant ones to you through its algorithm up to the top through a combination of most recent and most cited on your topic. So this is okay. I don't think it's necessarily good or bad. It kind of gets a yellow flag. I don't see anything unethical. These are not going to be hallucinated science like you were getting in the earlier videos we saw with GPT that are more risky, but it's going to be partial and might not actually be the most efficient way to go. Okay, I'm going to skip forward a little bit.

[00:35:19] Speaker 6: Your own thoughts. To speed things up, you can use the chat PDF feature on paper path. Just upload the paper and you can start chatting with it right away. Get quick summary.

[00:35:28] Speaker 1: I like this a lot. So this is a nice application. I think being able to chat with a PDF is one thing I have recommended in the past with size space. Although now you can do this chat GPT and notebook LM even better. What's great about those is now you check GPT. You can create a project notebook LM. You can dump a whole bunch of papers and start having a conversation with your literature and that can deepen your understanding. It can be especially good if you're just starting out and you don't understand something. You can get those AIs to explain it to you. Like your five. So this one definitely gets a thumbs up application. Let's see though. I suspect they're going to have some writing in here because this seems like they're promoting one of these one-stop shops AI to do everything. Structure.

[00:36:05] Speaker 6: Go to the template section inside paper pal. Click on outlines. Type review paper as the article type. Add a little info about your topic in at least 10 to 15 words. And that's it. Paper pal will generate a full outline for your review paper in just a few seconds.

[00:36:21] Speaker 1: So this outline again, I think it works for a graduate essay, but for an actual paper you want to publish, you're going to get stuff that's just generic. And I wouldn't recommend for an actual review paper to develop your outline this way. Also narrative reviews are pretty tough to publish. For a systematic review, there's going to be a whole other structure that's going to emerge naturally from your analysis. The AI is not going to be able to do that synthesis for you without having the right data in. And even so it won't do that very effectively.

[00:36:50] Speaker 6: The outline will show you what to write in your introduction. How to...

[00:36:53] Speaker 1: Yeah, so this is problematic. This gets the thumbs down. If you have aspirations to publish, don't do this stuff. Last one. How to write a full research paper using AI. You know, after a while, these all start to seem kind of the same, but let's see what we've got here.

[00:37:09] Speaker 7: In this era of AI, the biggest challenge for a researcher is that you need a different tool for every single task. One tool to search paper, another to summarize, and a different one to write articles and maybe even something else just for reference management. And switching between multiple tools is not at all practical. And in the end, you also end up paying more by subscribing to so many different platforms. So I was looking for a single AI tool that could do it all right from generating ideas to write the final paper so that I can minimize my efforts, save time, and of course, save money as well. A few days back, I came across Gatsby, an AI research assistant that promises to fulfill almost all my research needs. Now, of course, this sounds too good to be true, but we are here to test if this tool is really capable of writing an entire research article. It also has other features like suggesting novel ideas, writing literature reviews, meta-analysis, and even patents. So I want you to hold on until the end as we will find out whether Gatsby can actually deliver what it promises. Now let's check out this tool.

[00:38:14] Speaker 1: All right, let's go look at the comments for a second. People are buying this. Okay, I want to see what it says.

[00:38:24] Speaker 7: This is the official page of Gatsby, the next-gen AI research agent. Unlike most AI tools that can only be used on the web, Gatsby also gives... The tool still requires an internet connection for logging in, assessing AI features, and searching the web. You will see the core features listed on the left. Gatsby Innovator for generating research ideas. Gatsby Writer for drafting research papers and patent disclosures. And Gatsby Reviewer for systematic review and meta-analysis. Now let's see...

[00:38:55] Speaker 1: Okay, we've seen some others trying to promise these features. Let's see what they do.

[00:39:01] Speaker 7: Starting with its most powerful feature, the Gatsby Writer. As an input, you simply provide your topic and your research methodology.

[00:39:09] Speaker 1: The problem again, I can... Okay, input attach your existing materials. Let's see how heavy a footprint it takes.

[00:39:16] Speaker 7: I'm uploading the method section from one of my research paper draft as a separate document. I have also added some figures to get better results. This is essential for generating a full paper. Currently, the tool supports only the doc format with a maximum file size of 5 MB. Personally, I feel this limit should be a bit higher, especially if you want to include more information like tables and images. Before generating the manuscript, it first gives you a summary of your research so you can check if it's going in the right direction. At the bottom, there are options to specify whether it's a research paper or a patent and even set the expected manuscript length since some journals prefer shorter papers. For now, I will keep the default setting and click on write manuscript. The writing process can take several minutes, but...

[00:40:03] Speaker 1: Okay, so it sounds like this person has uploaded the kind of core of the paper, the methods and the results, and is getting it to then... It had a result set with the figures and tables lined up and is getting it to thread it into a full paper. Let's see how it looks.

[00:40:20] Speaker 7: Honestly, that's still nothing compared to the weeks I would normally spend in writing. To save time, here...

[00:40:26] Speaker 1: Well, just having gone through that very quickly, I can see very short sections in the results, very thin. So I'd have to look at the actual quality of this. This is a yellow light because you're uploading your actual data and having it write based on that. So that's not as bad as some of the earlier examples where you're just complete blank slate, write this for me. But if you've already curated your article to get the main figures and tables, I'm not clear how this is gonna save you time because it'll take you more time to clean it up in some cases than just writing it up yourself. So I'm missing the full payoff here. And it's unclear if you already had a draft of the method section, what did it change? And did that follow your journal format?

[00:41:14] Speaker 7: I will directly show you the results. Let's check out the generated output. Here's the title it has refined. And honestly, it looks good, clear, specific and suitable for this study we perform. Next is the abstract, which includes a well-defined background, the methods we provided.

[00:41:30] Speaker 1: So how is it gonna know, right, understanding this here, this is a very generic first sentence. It doesn't say what the gap is or the value out of your study. So this is the other problem that it's not achieving what the back abstract needs to do here. Again, I can't evaluate this fully because I don't know what the actual results of the study were. You know, if a student handed me something like this, I, well, firstly, I would be able to sniff out that this isn't probably the student's voice. And there's a certain lack of specificity here in saying what the results were. So like gene ontology enrichment, interactome mapping, leveraging this highlighted conserved pathways. It's not like clearly saying what the results and the findings were in a way that's coherent and creates the narrative arc or story for the paper. And it's not coming back in the abstract saying, well, how did that plug whatever gap it was supposed to fill that shows the value out of the study? So again, yellow flag, but publishing in high impact journals, not gonna fly.

[00:42:37] Speaker 7: And even the outcomes, since we didn't actually include the results section in our input, so you won't be able to see specific number or gene names here, but at least it guides you on what to write in the abstract.

[00:42:50] Speaker 1: Okay, so it actually didn't provide the data. So it just started making stuff up. That's not great either, but okay.

[00:42:58] Speaker 7: It saves a lot of your effort. After that comes the introduction. Now keep in mind, I didn't provide the tool with an abstract and introduction or the results in the input. Still, it managed to draft a complete introduction.

[00:43:09] Speaker 1: So here's getting, it's starting to go these previous studies. So it's nice that it's following the introductory structure to try to come up with gaps, but where did it get these studies from? Now, this is not a specific domain that I have subject area knowledge. I don't know where it's pulled these sites. You'd have to check these incredibly carefully and to make sure that this is accurate in justifying the gap, but I am seeing some disconnect from what is summarizing here and what was promised in the abstract. So my worry is that this is producing superficially, seemingly polished literature that when you probe deeper, doesn't make sense. So you get droves of well-written text that don't stand up to scrutiny. That's my concern here. Let's keep going.

[00:43:55] Speaker 7: I'm covering the background of the bamboo species, growth mechanism, past studies, the existing gaps, and why transcriptional profiling is important. I was actually surprised at how structured and logical it sounded.

[00:44:07] Speaker 1: Um, what he just showed a second ago, that's just kind of a standard signposting paragraph. You don't actually have this style often for biology papers. You get this more in much longer journal papers that are at say an 8,000 to 10,000 word length. So you can end up running a risk of one of the AI failure modes violating field norms here.

[00:44:28] Speaker 7: ...structured and logical it sounded. What I find really important here is that the tool adds the in-text citation to every citation-worthy statement. And this is critical because citing the right papers improves the credibility of your research and also helps our plagiarism issues.

[00:44:44] Speaker 1: Yeah, I mean, I guess you can go check the citation, but it's going to have partial access to the citation search. And he's at a key point. The right citation. How do you know this is the right citation? Let's keep going.

[00:44:56] Speaker 7: When you click on an individual reference, it actually redirects you to the original source. From there, you can read or download the full...

[00:45:04] Speaker 1: Oh, that's nice. So it is linking to the paper. So you know you're getting real papers, but again, it has a partial set. So it's only going to be tapping open access data. So you're limiting your lit review to a small fraction of the science. I mean, your paper can be completely redundant or duplicating something else here. And you wouldn't know it because you haven't searched all the literature that's out there. Using this method. But it is good that it goes that way. I want to see what else they're saying it does.

[00:45:26] Speaker 7: Another thing I noticed is that the tool also adds equation to the paper. For example, it includes a formula for calculating sequence identity, IP cam values and others. In this draft alone, it generated around 23 equations. Now, depending upon your study, many of...

[00:45:41] Speaker 1: 23 equations. Why are you generating 23 equations? That's quirky. Oh, this is not... Like again, I think this is hitting the failure mode if you're a beginner, you could be deceived by this and think that this looks great. Oh, 23 equations. Without having real understanding, you just get this cosmetically, seemingly good text. That's not going to hold up in peer review or when you get expert reviewers on it. So maybe as an exercise for learning, but for actually drafting a paper, this doesn't pass muster, at least not yet to my eye. But let me keep going. PHP reviewer.

[00:46:21] Speaker 7: I will use it to create a systematic literature review on topic effect of mindfulness on cellular aging. It first generates an outline covering the background, objectives, methodology for literature search and data extraction results and conclusion. You can regenerate this.

[00:46:37] Speaker 1: Okay.

[00:46:37] Speaker 7: Or add instruction to refine it further. At the bottom, it lists the papers it has referred to, but note it only analyzes papers.

[00:46:47] Speaker 1: Yeah, exactly. So it's only doing stuff that has full text for reviews with full access.

[00:46:53] Speaker 7: If you want to include specific studies, you can upload those full text yourself. Once the manuscript is generated, you can view it. Just like a research paper, the literature review comes with proper academic language, citations, and well-defined sections.

[00:47:06] Speaker 1: You can see the studies included for eligibility. I mean, just going back over this for a second here. Let me see if I can pause this.

[00:47:13] Speaker 7: You can view it. Just like a research paper, the literature review comes with.

[00:47:17] Speaker 1: Okay. And see, this is weird because it's not doing the actual search. It's actually making this up. This is subtle and this is dangerous. Hang on. I'll see if I can pause this again. It is making up the search strategy including a combination of keywords, boolean, operative, and or not. This is completely made up.

[00:47:34] Speaker 3: This is completely made up.

[00:47:36] Speaker 1: So this is not reproducible. It's just kind of simulating what a systematic review is supposed to have. It searched five databases. No, it didn't. And Google Scholar, if you use Google Scholar, it should be in the Prisma. And the Prisma would then have a right-hand branch where Google Scholar comes in. Databases, N equals 491. It's not citing. You would say how many came from PubMed, how many came from Scopus, how many from Web of Science. Duplicates removed. Let's see if the numbers at least add up. Okay. I can't see if the numbers add up. Reports not for retrieval. Five in the review. You can't get a systematic review with just five studies. This is again, violating field norms. There's nothing to synthesize. This is on the low end of what you could do if you were doing a meta-analysis and that would only be in an exceptional situation. And it's saying it's doing stuff on publication bias. You have too few studies to properly evaluate publication bias. This is, yeah, this is slippery. This is the kind of thing where the AI tool has gotten better and it's going to fool beginners who think that this has some automated sophistication that is flimsy and completely falls apart when somebody who knows what they're doing looks at it. So guys, this tool, this is actually getting ethically really worrisome for me in that how people can be beguiled by this stuff. And you're just going to, the irony is if you try to use these in this way, you're going to lose more time because you're going to eventually hit a wall and have to fix this. So if you do want to publish an actual systematic review, just learn how to do a systematic review, have that superpower, have that skill that you carry with you the rest of your life because fake it till you make it. Just doesn't work as an academic. Okay.

[00:49:18] Speaker 8: Last one guys.

[00:49:32] Speaker 1: I'm starting to despair at the state of humanity right now. I hope by this point, guys, that we're past this as a research strategy and approach and you're not going to be doing this. Guys, I have to say, I feel a bit shell-shocked and deflated by the state of what's going on and also the comments where researchers seem excited about this. I don't know if I can call them researchers and using this stuff and hoping they're going to get it and slip under the detection of a professor like myself or even worse, put it into a journal which raises even more deeper lasting ethical quandaries and complications. You do not want a paper retracted and have that stain on your record forever wherever you go. Guys, if you enjoyed this, let me know. Send me AI videos you want me to review and I will help you filter the wheat from the chaff and give you the unfiltered, no-nonsense look at what makes sense and what doesn't make sense and what's above board. But the reality is, you don't need a complicated tech stack. I often see people when they come to me and they're using five, six different tech tools. They're glorifying tools and thinking tools are going to solve a problem that's probably a thinking problem. Our tech triad that we recommend includes very simply Grammarly, Zotero for references and either ChatGPT or Notebook LM right now to be a critical voice and not a sycophant to agree with you, especially for AI peer review. And if you want to see more on how we recommend using AI and research workflows, you're not going to want to miss this video right up here.