Speaker 1: ChatGPT can speak, listen, and now it can see. Hey ChatGPT, say hello to my YouTube viewers. Hey there YouTube viewers, it's great to be part of your day. Hope you're all doing fantastic. Perfect. And I've been absolutely blown away by what it can do when you put in images and figures such as schematics from peer-reviewed papers. Now, to really understand its power you have to understand the peer-reviewed paper writing process. Here's a quick overview of what I do. So when I start writing a peer-reviewed paper, I actually start with data analysis, obviously, but then I take that data and I put it into figures, schematics, and tables. I then arrange those things in order of a story that I want to tell. Then I expand on that by putting bullet points underneath each schematic, table, or figure about the things I want to talk about. And then I literally go in and I expand on those bullet points. Super easy and you'll be amazed that ChatGPT now can essentially do all of that stuff. Once you have the figures, you can actually use ChatGPT and its image input to create the peer-reviewed paper outline for you. It's not perfect, but this is what I found. Now here's a schematic from one of my papers. I was using carbon nanotubes and silver nanowires to create a transparent electrode, but I've asked it here, this is a schematic from a science paper, can you understand what is going on? And importantly, I did not give it the caption. With the caption, this is even more powerful, but it says yes. The schematic appears to depict a process for preparing a substrate with specific material. Here's what's happening. It does really well at picking up not only what's going on, but the numbers and the order in which it goes, which can get a little bit confusing when there's lots of different types of arrows all over the place. But here it knows this is A, B, and C, so it says these two solutions are combined, yes. Pass-through filter paper, yes. The material that remains on the filter paper is of a certain thickness, no, no. That was stacked filter papers so that we ended up with a certain pattern on the filter paper. So it didn't get it 100% right, but that was the only thing it got wrong, and a quick edit will solve that, incredible. So I think it did a brilliant job. I wanted to know whether or not it could handle really complicated graphs without the caption. So I put in one of the most complicated things from that paper. You can see there's lots of arrows. The two different graphs share axes, and there's all of these different squares and whatnot going on. There's circles and arrows, there's everything going on. There's stuff up here, some information, and once again, I did not give it the caption, which actually holds a lot of the information that would help it work out what is going on. So it says, of course, this graph presents data on the relationship between the weight fraction of single walled carbon nanotubes, which it had to sort of like work out for itself, and two key properties, sheet resistance and the average spectral transparency. It's reading information that's up this way as well. I love this. Axes, plots, data points. Now it really does sort of like pick out all of the important bits. Importantly, it's explained the graph to me. Now, that's pretty good. It did get some things wrong, but ultimately it did get all of the important trends correct. So that means when you're producing a peer-reviewed paper or a thesis, whatever, you can put your figures in and just say, hey, can you make sure that you understand this and give me some sort of key trends that you've picked out from this table or graph? Excellent. So it seems to do pretty well with little tiny dots on a graph. Does it deal well with photographs? This is a photographic sort of experiment I did about the structural stability of these transparent electrodes at different weight fractions of single walled carbon nanotubes. And you can see here I included the figure caption, which I think improves its power. So writing the figure caption is really important. It says, of course, it can explain it. Of course, it's easy for it. It doesn't even bother asking me anymore. I can do everything. Chat GPT, I love you. Figure five showcases a series of color photographs taken in a structural integrity test. Boom, it knows immediately because I put in the figure caption. Column A, column B, column C. It does it perfectly. Now, this is where I wanted to expand on what I think this can do. Because of course, if you've got figures, you do want to write about that figure for a paper or thesis or experimental report. That's the word. All right. So turn this into a paragraph suitable for including in a peer reviewed publication. Certainly. Certainly. No problem for me, mate. In figure five, we observed the time resolved impact of acetone on the structural integrity of various film compositions. The pure silver nanowire film depicted in column A demonstrate a rapid degradation. Now, this is one thing I wanted to point out up here, that it does show that degradation takes place. It knows that this is degradation and this doesn't change. I think this is just so incredible because it did it so quickly as well. So here it knows what degradation is and then it says, films with 20% single walled carbon nanotube weight fraction presented in column B display noticeable but limited structural changes. So, yeah. It does do that. Incredible. So there you have got a first draft for a paper and it's not always perfect. In fact, there are some little tiny things that it really does sort of like get wrong but that is fixed by a five minute edit. You'd no longer have to create all of the information for your first draft of your peer reviewed paper. Incredible. It does pick up on the bubble like surface protrusions and it just has done such an amazing job. I am almost speechless but not quite. I then wanted to know whether or not it was able to actually take different types of information like tables. Now this is a pretty low resolution table. It's got the title and it's got some other sort of like different information. So can you give me a summary of what's happening? And it does a really great job but with tables and this low resolution stuff here, it really sort of like doesn't do incredibly well. So I think by including a simpler table, it would do even better. So for example here, it says that the RS and RP values are slightly increased compared to the unannealed sample and if we go there, we can see one of them is not only just slightly increased but significantly increased and also this one is not increased, it's decreased. So overall, it has got some stuff wrong but also let me remember this, that this is picked up with a simple edit. All of this would have taken me, I don't know, half an hour because it's got all the numbers and stuff. This is already done for me. And then I say, can you turn this into a paragraph suitable for peer-reviewed paper just like I did before and it does a really great job. I think if you were to correct it and say, no actually, in the table, I noticed this, this and this and then you asked it to do this, it would just write it out perfectly. That is because this is a conversation. Remember with ChatGPT, you need to go backwards and forwards. It's not just a one-shot solution. So have that conversation, tell it where it's going wrong and I feel like this is a real incredible step for speeding up some of the boring parts of writing for science and research. This is how I think you can use it. First of all, you can use it if you don't understand someone else's graph. Plug in a screenshot of that table and it will explain it to you. Brilliant. The second way I think you can use this is just like we've done. If you've got figures for your own peer-reviewed research or your own thesis, put it in and say, hey, explain this for me, tell me the trends and then also put that into a paragraph. Make sure it gets that trend thing right first, correct it and then say, hey, give me a paragraph. I think that's the most powerful way you can use it and finally, if you're doing a presentation or even if you're just trying to refine your figures, put them in and see if it can actually pick up on the important information. If it can't do that, maybe there's something wrong with your figures. This could be a way for you to self-edit and make figures better by making sure that something like ChatGPT can actually work out what's going on. If ChatGPT can do it, it's almost certain that the people reading your paper will also have a really easy time figuring out what each of your figures mean. Brilliant. Use it for those ways. I can't be more impressed at the moment. Now, that is with ChatGPT Plus, the one you need to pay for. Now, I know a lot of people don't have that money but I'm going to tell you why it's so important to spend that little bit of money. There are free versions. There's this one called Lava and what I did is put in that same schematic and I said exactly the same prompt, this is a schematic from science paper, can you understand what is going on? And you can see it does not such a great job. The image is a schematic from a science paper illustrating the process. It misses out on all of the details, it's the details that we're interested in. Therefore, free models aren't quite there at the moment. I think they'll get there over the next few years but at the moment, ChatGPT is just smashing it when it comes to getting the details out of the figure, getting the details correct and then able to talk about the trends. Just incredible. So, I think ChatGPT Plus is going to become something that is invaluable for science and researchers going forward. The free models, the open source stuff is slowly catching up but it is far behind where it needs to be at the moment to be absolutely useful. ChatGPT Plus, it's able to listen to you, tell you what's going on and now see what you're creating is just mind-blowing. I really wish I was in science right now because I feel like I'd be pumping out those papers, becoming that academic superhero we all want to be. Let me know in the comments what you think. Have you tried it and is there something else that I should know about that is better than this? I always love to find that out. If you like this video, remember to go check out this one where I talk about epic ChatGPT prompts that work for science and research that will speed up your paper writing process, that will speed up your experimental process. It's all there to make sure that you are more efficient with your time and ChatGPT. I include all of the prompts that I use in there, so go check it out. So, if you love that, remember there are more ways to engage with me. The first way is to sign up to my newsletter. Head over to andrewstapleton.com.au forward slash newsletter. The link is in the description. When you sign up, you'll get five emails in about two weeks, everything from the tools I've used, the podcast I've been on, how to write the perfect abstract, my TEDx talks and more. It's exclusive content available for free, so go and sign up now and also go check out academiainsider.com. That's my project where I've got e-books, I've got resource packs, I've got a blog forum and courses are coming out soon. I can't believe it. So, go check it out because it's all there to make sure that academia works for you. All right then, I'll see you in the next video.