Speaker 1: I have been a ChatGPT fanboy for ages, ChatGPT, ChatGPT, ChatGPT, but I have been told so many times that Perplexity is better. There's been loads of updates to Perplexity, to ChatGPT that I wanted to solve it once and for all. So we're going to compare ChatGPT with Perplexity and create a little scoreboard. Scoreboard. You'll be surprised, and I was actually surprised about which one does certain things better. We're going to look at the four things that sort of like PhD students normally do. That's finding literature, asking for help with explaining peer-reviewed papers, some data analysis, and also writing and editing. So here we go. The first thing, I went over to ChatGPT and I said, I am starting a research project about hydrogen production for hydrogen energy using catalysts. Can you please recommend the best peer-reviewed papers to read for this project? And I did exactly the same in Perplexity, but with Perplexity, I used Academic, all right? So if you go over to Perplexity, you do have lots of options. Let me just find that for you. Should have prepared this a bit earlier. How unprofessional of me. But if you go to Perplexity, look, you can go to Focus and you've got Academic, Writing, all of that sort of stuff. Now clearly, I wanted to search in Published Academic Papers, so for the Perplexity version of this, I clicked Academic. So let's go back to ChatGPT, and it says it searched four sites. So this is what it found, and it does sort of like present them just in bullet points. So here we've got, for your research project on hydrogen production using catalysts, I recommend starting with the following peer-reviewed papers. So I clicked through, and they are actual real papers, so we're not having to deal too much with the hallucination that we saw a few months ago, which is really great. And these are actually pretty good papers, which I like. So well done, ChatGPT. It did what I wanted it to do. It found four papers. But does Perplexity do better, considering that I can tell it to do academic searching? So exactly the same here. And then because I've got Pro for both ChatGPT and Perplexity, I paid the money. You can see that it went and did a Pro search, and it found 20 sources. So clearly, Perplexity has done so much better at finding resources. So let's have a look to see what it says. Based on the following search results, here are the most relevant, highly recommended peer-reviewed papers for your research project. So it gives me different kind of sections as well, which I really like. Catalysts for hydrogen production, electrocatalysts for water splitting, hydrogen for chemical hydrides. And here's the thing. It gives this little summary down the bottom. I would recommend starting with the review papers, three, five, seven. So you can see it's given 20, but it's actually sort of said, let's start here. So I have to say that this is a win for Perplexity. Bing. Chalk it up, because Perplexity has won this one, hands down. Sorry, ChatGPT. Sorry, mate. Perplexity has just done a little bit better. Actually not a little bit better, way better. So the next thing I wanted to test is whether or not ChatGPT and Perplexity can explain research papers. So I went to ChatGPT first, and I uploaded one of my papers, and I said, explain the key important points to me from this paper. So pretty clunky prompt, but nonetheless, here we are. It said here, memory updated. So it puts this into its memory. And then it's got introduction, experimental results and discussion. You can see it loves doing bullet points, does old ChatGPT. So it gives me loads of bullet points. I don't mind bullet points, but it does get a little bit boring and repetitive when you're doing this over and over again. You kind of like, you know, lose how many bullet points you've actually read, and it gets really sort of just boring, I think. But look, it's done what I want it to do. There's no details in there. It's pretty high level. But once again, remember that with any large language model, it's a conversation going backwards and forwards. So yeah, this is good. I would say this is good. So can Perplexity do any better? And I did the exact same thing. It's given the key important points to me from this paper. What a clunky prompt. But once again, it went out, it did research question, and it also did searching the web, which is kind of interesting because I only wanted to look at that paper. But it's found sources, and then it's gone out, and it's actually sort of like combined the sources with the paper, but it's mainly done the paper. You'll see what I mean. Here it says, okay, the answer, here are the key important points. And instead of going section through section like ChatGPT did, it kind of pulls out the most important things and categorizes it together. So it says that we've got fabrication of vertically graded active layers, which is great. And then we've got improved open circuit voltage, which is great. But then also you can see, look, it's got my paper, but then it's also got this one, morphology controlling organic solar cells. So I think I like that it's trying to sort of like add deeper understanding to this research paper. So it's a win. Sorry, ChatGPT, I'll miss you, but I think perplexity is my new best friend. Ding, chalk it up another one for perplexity. At this point, I'm feeling pretty sad for ChatGPT because perplexity is winning. But can it redeem itself with this? I want to see if it can plot graphs and do a little bit of calculation for me. So I put in some unstructured data that I've got from my PhD and postdocs, and I said, this is IV data from solar cell testing. Can you plot the IV curve and calculate the efficiency of the device? So it says what it's going to do, which is great. And then it gives me all of this stuff that I don't really understand. But this is a load of, I think it's Python code, look at that. And then it's calculating efficiency, it knows how to calculate efficiency. It really does such a great job, ChatGPT, I love it when you're doing data analysis. So this is what it gives me where blah, blah, blah from the file, and it knows where to find the information. And then it says, okay, let's implement this in code and see the results. And then it goes backwards and forwards a little bit. It didn't extract the data initially, but it tries and tries again. And then it comes up with something to correctly pass the data. So it's kind of corrected itself, which I really like. Anyway, let's get to the most important thing, which is this, it managed to do an IV curve, which I really like. You can see it's an interactive graph, you can click around and do things, you can switch to a static graph if you really want. That's really great. So I would pull this out, for example, for any sort of little reporting I'm doing to my group or my supervisor for my supervisor meetings. I'm dribbling again, I'm getting old, just dribbling, just leaking from my mouth. And yeah, so we'll go back to interactive graph. You can see we can hover over it, love it, love it, love it. And then it says, yeah, the efficiency of the device is calculated as zero. It was, yeah, that device failed. So well done, it did it. So can perplexity do any better than that? So this is exactly the same thing that I put into ChatGPT. It's understanding the question, it's reading one file, and then it's saying, plot this. This code will generate an IV plot with the average, so I need to go and use this somewhere in MATLAB or in Python to actually plot this. It didn't do a graph, fail. And it says the efficiency of solar cell is already provided in the metadata. So here it's looking at the metadata, it hasn't gone and calculated it itself. So I think this is a win for ChatGPT. You've redeemed yourself, ChatGPT. The last cheeky little test, ooh, cheeky, cheeky little test that I wanted to do was have a look to see what it's like for writing, comparing ChatGPT and Perplexity AI on which one is best for writing and editing scientifically. So I said rewrite this paragraph to make it more suitable for peer review, and I took one of my unfinished review papers, took the first massive paragraph, copied it in, and said make this better. And you can see here, transparent electrodes have become integrated. It does a pretty good job, ChatGPT, of just taking what I've got and just making it a little bit better. But you can see that it hasn't tried to reference anything, it's removed the references. But it has really sort of like tightened up that academic language. So once again, good. But, oh, Perplexity, you've done it again, you cheeky son of a gun. Okay, I tried not to swear through any of that. Can't demonetize me. Now I've got this exact same thing here. Rewrite this paragraph to make it more suitable for peer review. I've got that massive paragraph, and I like it, oh my god, I like it because it's gone out searching the web and it's found peer reviewed papers to put in to the references. So not only do I have better references, potentially I'd need to check them obviously, but it has tried to reference where it needs to be referenced, like here even. So how to perform peer, oh, how to perform peer review. You've received an invitation, oh I don't know why that one's there. Don't be cruel, how to write fair peer, okay, alright, okay, we've delved a little bit deeper. It's a bit rubbish with the references. So it's done an okay at creating the text, I think it reads better. We've got peer review, how to write a peer review, okay, it's completely misunderstood the references. When I first looked at this, I was like, oh my god, it's my references. I would say that because of this, ChatGPT wins. I think because it's just not giving me anything useful, peer reviewed examples. This paragraph summons the key points of the paper with references cited at the end, but it's the wrong references mate, sorry perplexity. Oh my god, that means ChatGPT wins, bing, it's a draw, it's a draw. So that means to me that you really have to trial this for the ways you want to use it. If you want to do much more data analysis, it makes no sense to use perplexity. You should be using ChatGPT, it does a far better job with data. But if you're looking for references, if you're looking for more academic kind of searching online, clearly perplexity is the best. So no one size fits all and unfortunately, it's going to have to be a draw here. I'd probably use both. I'd probably know the strengths of each and say perplexity, you're better at searching the internet, getting me references and starting that literature review, whereas ChatGPT, you're probably a little bit better at academic writing, although equal there maybe, and you're a little bit better or you're much better at data analysis. So there we are, perplexity AI versus ChatGPT, both, what a cop out on my part. If you like this video, go check out this one where I talk about using perplexity AI in research and how terrifyingly smart it has got. You'll love it, go check it out.