Speaker 1: The first thing we're doing wrong is using non-specific prompts to get what we want. We've just gone through this crazy sort of like few months where only a few months ago we had to do everything ourselves and now I feel like I know I did. I expected AI to read my mind and know exactly how I want things laid out, presented, the information I need in a single kind of two sentences. It's possible but you have to use the right prompts. There's two things I think are very important when it comes to prompt engineering. The first one is the simple format. Follow this format and it will make your life so much easier. So first of all you have a role that you want the AI to kind of take on. The second thing is the task that you want it to do and the third thing is the format in which it should present the information or the results of the task. So for example as a scientist outline a good literature review in this field, in organic photovoltaic devices and give it to me in headings. And now you've just given it everything it needs to do the best job for you. The second thing you need to really understand is that linked prompting really works. So quite often you know we just want to put create a literature review on this when in reality we should be asking it step by step like what we would do as a researcher without AI. So for example linked prompting means first of all asking the AI to produce headings, secondly asking it to produce subheadings, thirdly asking it to produce a hundred words under each subheading with citations using the right prompt structure as a role, task and format and then you slowly build it up. And at any point the AI kind of just doesn't seem to understand what you're asking. You can stop it and ask it to do it again with a different prompt. So link prompting is very powerful especially when you use prompts that follow role, task, format. That's how you use it. Stop expecting it to read your mind. Stop using the base data that comes with ChatGPT. I have found that the most powerful way that it can be used is when you are giving it your own data and input. This is why I really like HeyGPT. I pay for that because it interacts with PDFs, with websites, with Google. It just really means that you are getting data from beyond the base model of this large language model which is ChatGPT or any other AI large language model systems. Now the thing is that it can take a little bit of time to generate your own little database within ChatGPT. So it's like read this, read this based on this information, now grab this. You need to feed it information before you start asking it questions. But I have found that it's well worth it because the information you get not only is unique and specific to what your work's about but also it really sort of acts as a personal assistant for working with the language and information you put into it. So stop using the base model and just going up and being like hey do this for me. Say hey here's some information, now this is what I want you to do with it. Using it in this way is just so much more powerful. Trust me it's worth a little bit of effort. Stop using it for everything. There are some things it excels at and there are some things it doesn't do very well at the moment and you need to go find the specific AI tool that does that thing better. ChatGPT is very good at language, at giving it information, formulating really sort of like concise and specific paragraphs of text. It does that outstandingly well. It doesn't do literature searches very well. That's where you need specific tools like elicit.org, like litmaps, like connected papers, like ResearchRabbit. So you need to find tools that are designed by researchers for researchers because that is where the real sort of like power of AI comes in. There is no general AI for science and research at the moment. Therefore you do need to go and pick and choose the strengths of each tool and combine them on your own. I'll do a video about that later, maybe like the ultimate review of all of the AI stuff so that you know at what section you should be using which AI tool during your research. So that's to come. Remember to subscribe and you'll get that video. If you're doing your literature review, there's a bunch you can use. If you're writing up, there's a bunch you can use. If you need sort of like submission guidelines and checks, there's a bunch you can use. If you head over to academiainsider.com and have a look at my recommended app page, that's where I will list out all of the things I've personally tried that I personally think are valuable and that's where the specific areas of research and the AI tools that you can use are really kind of like aggregated. So go check it out. You are still responsible for everything that the AI kicks out. If you are submitting it to peer review, even just for a course and coursework, you are responsible for everything it kicks out. You need to read over the stuff it gives you. You need to change it to make it specific for you. Remember that this AI is a tool. It is not an assistant that will just give you perfectly sort of formulated scientific and research pros for you to submit. You need to change it for your own use case. You use what it gives you as a base model for changing to meet your needs. You need to make sure that every reference and citation that it gives you actually sort of like is the appropriate reference. I know that even without AI, people were just referencing random science that actually had nothing to do with what they were using it to cite. So it's very important that you check specifically what you're citing, how you're citing it, and you change the text that comes out the other end of chat GPT. You have to remember that this is a tool. Any AI that you use during your research is a tool and it's not to be used without going through your editing, without going past your checks because you're going to get yourself into a lot of trouble if you don't check the stuff that any AI tool generates for you. So there we have it. There's everything you need to know about the ways you shouldn't use AI during your research. Let me know in the comments what you would add and also remember there are more ways that you can engage with me. The first way is to sign up to my newsletter at andrewstapeton.com.au forward slash newsletter. The link is in the description. When you sign up, you'll get five emails over about two weeks, everything from the tools I've used, the podcasts I've been on, how to write the perfect abstract, my TEDx talk, perfect daily schedule and more. It's exclusive content available for free. So go sign up now and also head over to academiainsider.com. That's my new project where I've got my eBooks and my resource packs as well as a blog that's growing out, my insider forum. Everything's happening over there to make sure that academia and your PhD works for you and not your supervisor. Well, it can work for your supervisor as well but mainly it makes sure that you are informed about the insider workings of academia. There we are. That was long, wasn't it? I'll see you in the next video.