Speaker 1: The three monsters of AI are now available to everyone. They are ChatGPT, Google Bird, and Bing by Microsoft. So what are they like for research? Clearly we are really far away from a general AI for research, but let's have a look at the strengths and weaknesses of each. And it will surprise you, I think, what these can actually do. And we'll be looking at where the strengths and weaknesses are for each one by comparing them side by side. So here they are in my window. We've got ChatGPT, Bird, and Bing. So let's go through the sorts of things that researchers would typically use AI for. And those are researching papers, summarizing papers, working with words in theses, in papers, and we're gonna stick to kind of like, you know, the language things, and also transforming papers and research into outputs that are not just academic. That could be PowerPoint presentations, that could be blog posts, and that could be press releases. So I've got a load of different prompts, and the first one we're gonna do is have a look for papers. Let's see, now typically these have been very, very bad at finding papers. But what we're gonna do is enter the same prompt in all of them and see what happens. Acting as an expert scientist researcher, find the latest papers on new materials for organic photovoltaic devices. Now obviously, Google has got a slight advantage because they've got access to the internet, as does Bing, whereas GPT-4 has no internet access. So, hi, editing Andy here. You'll never believe it. As soon as I went to edit this video, I was given access to ChatGPT with internet browsing. So if you have a look, I've run a few tests, and essentially you can get real references from the internet. You end up with this little box here, and it tells you what it's done, where it's gone. It's like sending out a little agent into the world to do some tasks for you based on what it can find on the internet. So it's really powerful. Expect a video in the future talking about how you can actually use this, but just to let you know that this was very bad timing. And it now does have access to the internet, and you can see here that I've tried lots of different ways to get papers, and you end up getting real references, so it does exist. Bear that in mind while watching this video. All right, back to the edit. So clearly, because ChatGPT doesn't have access to the internet at the moment, it will not tell us anything about the most latest papers. So in this sense, at the moment, this is not doing so well, but it does tell you how you can find papers using different databases. Here, we've got some latest papers, so 2023 apparently from Nature Energy, and we need to go and be a little bit suspicious about this because we'll need to check to see what it's actually kicking out. Here's some more here. I found some recent advances on materials for organic, foltaic devices. So this is Bing, and it has provided some links. So let's open up and have a look to see if it's actually provided us with real answers. So, renewed prospects for organic photovoltaics. This is from 2022, and I think overall, it's not done too bad. It's only found two, but that's okay because we can use a tool such as Connected Papers, Litmaps, or Research Rabbit to start to expand our search from there. But you can see it's actually not done too bad of a job for finding 2023 papers. So that's a massive leap forward from where we were with just Google and ChatGPT. Now, I've tried this a little bit on my own before recording this video, and I can tell you that not all of these are going to exist. These papers describe the new developments, so let's see if it can provide me with a link. So, provide links. Let's see if it can provide us with the links, and it's got Google it down here, so maybe it's going to help us in that. I am text-based AI and that is outside of my capabilities. Fine, no problem. I'll go and Google it myself then. Okay, now let's have a look at summarizing PDFs. Now, clearly you can't put PDFs in directly to ChatGPT yet or Google BARD, but we can use a text prompt splitter to put a paper in, so that's exactly what I've done. Let me just copy and paste across this paper. Now, one thing I like about this is that ChatGPT really does know when you've put stuff in and it tells you, okay, I'm waiting for the next stages. That's something that Google BARD doesn't do at the moment, so I'm just going to let that run out. It's providing me with a little bit of a summary, which is no problem, but I'm going to do the same thing with Google BARD to see if I can actually put stuff in. So, here I'm saying, look, this is the total length, and it says, I understand, please send me the first part of your content. So, then I'm going to send it the first part of my content. Why does it not go down the bottom there? Look, I can't assist you with that. That's all right, let's just put it in the next one. I don't quite understand why it says that. So, it's struggling with the input stage. So, it doesn't really like receiving lots of text or at least struggles to understand what I really want from it, despite saying I'm going to be posting in loads of text to you. But once it's in there, I hope that I can do fun things with it. So, if we're talking about PDFs or putting stuff in, like ChatGPT and BARD, it's a pain in the ass. But all I have to do with the Bing chatbot is open up the PDF in a tab, and then I've instantly got access to the information in there. So, now that we've got the information sort of across the three, it's all in there, let's see what we can actually do with it. So, I've got a prompt that I'll put into each one, acting as a scientist researcher, summarize this paper into the most important five bullet points. Let's have a look to see what each of them does. So, first off, BARD has just put out a load of information in one go. So, here it's got the five most important bullet points, graded active layer, blah, blah, blah. It hasn't got any facts or figures in there, and I'm a little bit suspicious that it's quite high level. Whereas, I think ChatGPT has kind of captured all of the most important information a little bit better, and it's sort of little bit more verbose, which helps me sort of like trust what it's saying. It's also going into the nitty gritty of what was actually found, which I kind of like. Something that BARD didn't do, and over here, Bing is pretty good at doing this, but it's not a huge kind of amount of information, and it didn't pick up on the fact I wanted five bullet points. So, overall, I reckon that ChatGPT is a winner in this case. Okay, writing. So, let's say we've got this paper, and now we want to turn it into different things, something that's now becoming increasingly important in science and research. So, I want to turn this into a press release. Let's copy and paste all of this prompt into each one, and we'll ask, acting as a science journalist, take this paper and turn it into a press release. Here, this is interesting that BARD, once again, is super quick off the mark, but you can see that it's got, for immediate release, your name, and then a study published in the journal Nature Energy has found the graded, so no, it's not in Nature Energy. At least I don't think it was published in Nature Energy. That's like too good of a paper. No, it was in Organic Electronics. So, here, it's got some of the facts wrong, but overall, it's given us a nice structure. Let's have a look to see what ChatGPT has managed to do. So, once again, it's got, for immediate release, breakthrough in solar cell efficiency, researchers innovate with graded nanoparticle. So, I think ChatGPT has done a much better job at capturing the essence of the paper, but also putting it into a press release format. We can use this as a good base, and I think it's done a better job than the other three, because if we go over to Bing, you can see that it's come up with a press release. It didn't follow the strict press release structure like the other for immediate release and that sort of stuff, but it has provided us with subheadings and the body and conclusion. So, it's kind of missed out a little bit on the subtext of what a press release is, but I'm sure that you could use it. But once again, this is a win for ChatGPT. So, I used to write for a company called Science Alert, and I would take articles like this and turn them into blog posts, about 800 to 1,000 words, nothing crazy. Let's see if ChatGPT, Google Bud, and Bing can do the same thing. So, here I'm going to say, acting as a science journalist, take this paper and turn it into a blog suitable for publications like Science Alert. So, we'll send that one off, we'll send this one off, and we'll send this one off as well. Okay, I think once again, ChatGPT has done the best job. It's kind of really understood the brief. Here we've got making solar energy more efficient, a deep dive into graded nanoparticle layers, and it's done its best. It's still writing, and they understand the brief. Whereas, Bud kind of got there. What are OPV devices, limitations, graded? So, it's more formal, and it's more about like an exploratory kind of blog, rather than the actual kind of impactful blog that I was expecting it to write. So, here we can see that ChatGPT, using the GPT-4 model, has done a much better job. It's using language that is appropriate for a blog, whereas the other two just really missed out. Bud did okay, but I wouldn't be able to publish that or send it off to an editor in a publication like Science Alert. And here, it's just missed it completely. So, Bing doesn't really understand what I'm asking it to do. So, it's just saying, sure, writing a science blog, it's just telling me how to write a science blog, rather than acting like a science journalist and creating a blog. If you can get the information into ChatGPT, there's no doubt that this is the most powerful language model at the moment. It understands the nuances of what you want. But what if you're starting from scratch? Let's have a look to see if we don't put in a paper, and we want to generate the headings of a literature review, for example. One of my areas of research was transparent electrode materials. So, let's see if it can start me on that process of discovery in the right direction. So, here I've got a simple prompt, where I'm simply saying, acting as a PhD researcher, write an introduction for a paper about transparent electrode materials. Include references where possible. Now, that's going to be a little bit of a stretch, I think, for all of them, but let's see what happens when I send them all off. Now, I think they've all done a pretty good job, really. I'm not going to trust the references, but overall, you can see that it's understood the brief, and each of them have done a pretty good job. So, here we've got ChatGPT, and you can see that it's tried to put in stuff, and it understands that indium tin oxide have been predominantly used, and it's got some references. Like I said, I'm not going to check those now, but I'm going to guess that a lot of them are fake. Whereas over here, we can see that BARD has not done too bad a job either. So, here it's looking at transparent electrodes. It knows that it's mostly indium tin oxide, and it knows that there's other materials, such as carbon nanotubes, metal nanowires, and all that sort of stuff. Graphene, carbon nanotubes. So, overall, it's done a really good job at sort of like just a little grab of where the field is. Now, I don't trust the references, but you can go to see if it's given you proper references. Whereas here, you can see Bing has done a really good job at producing what I think is probably the best one of the three. Not only does it understand the format of a paper, but also it's given me references. It's given me three, four references that I can go and I can use as seed links and find other things in other tools. So, overall, I think that's a win for Bing. So, those are all the different language-based tasks that I think AI tools can potentially help researchers with. However, we're at a stage at the moment where we're in this kind of like scrambly stage where no one tool is perfect for everything. If I was going to choose one tool, I think I would choose Microsoft Bing. I think it does the best job at the hardest parts of research, which is finding references and producing the first steps towards a research project or paper. Whereas, if I was working with a lot of language and I could copy and paste it into an AI bot, I would choose ChatGPT. Now, BARD is, I think, for research, not very useful at all. It struggles not only with referencing, but it also struggles with understanding the nuance of what I want. So, at the moment, I really feel like if you are working with language, it's ChatGPT with the GPT-4 model. Whereas, if you are wanting references, if you want to sort of like interact with PDFs and you want to sort of do a little bit of stuff where it searches the internet, Bing is the place to go. BARD hasn't quite got it. So, there we have it. There are a comparison of all of the juggernauts of AI tools at the moment. Let me know in the comments what you think. And also, there are more ways to engage with me. The first way is to sign up to my newsletter. Head over to andystoughton.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 podcasts I've been on, how to write the perfect abstract, and more. It's exclusive content available for free. So, go check it out 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 the Insider Forum and a blog that's growing out. And it's all there to help you make sure that your PhD in academia works for you. All right then, I'll see you in the next video.