Ryan Eakin Live on Shally’s Alley

Shally Steckerl shows how to build a custom Google search engine for quantum transduction research using ChatGPT-4o. She walks through generating domain lists, importing them into a search engine, removing duplicates, and adding query refinements. The paid GPT-4o plan offers unlimited queries and document uploads.
Chapters
Hey everybody, this is Shelly and we are live on Shelly's Alley. It's Friday, December 13th. Uh, unfortunately today's guest had a bit of a family emergency and wasn't able to join us, so I decided to just jump on, um, and just say hi to everybody and also show you a a neat trick that I've been goofing around with using the version of ChatGPT that you can install on your computer. Now, many of you may know that I use a Mac, so if things look a little bit different than they do on your Windows system, that's because the Mac interface. I just wanted to kind of show you what I've been playing around with, and feel free to ask me any questions via the chat as well here as we, as we demonstrate a few things. First off, what I'm going to do is I'm going to share my screen. Actually, I think I can share just a particular application, and I should be able to use that instead of the entire screen here. So let's try that. Oh, doesn't look like I can do that. Let's just go ahead and share the entire screen. All right, so it's going to be a little weird because you'll see like a window-in-window situation here going on, but that's all right. So let me show you ChatGPT. And this is activated by just simply using the— in the Mac would be the Option+Spacebar command, and I believe on a Windows would be the Start+Spacebar or Control+Spacebar. But you can see it brings up a ChatGPT window right here, which is what I'm going to use. I'm going to expand this so that it looks a little bit bigger, see how far I can make it. No, that's about as far as it gets. That's okay. All right, let's take a quick look here and let me just show you so sort of how I start out. What I've been doing with this is I've been using the, um, the, the intelligence of ChatGPT and its archive to create lists of websites that I can then add to a Google custom search engine, or, uh, what they call now Google Programmable Search. Finding these lists of websites could be a little tedious, but it is one of the best ways to use Google Programmable Search. So right now I'm, I'm doing a search, conducting a search for people that are specialized in a particular area of material science called quantum transduction. So I'll start with just essentially making a list. I'm going to ask GPT to make a list of the top 20 websites I can search to identify quantum transduction experts in the materials science field. I'll just start with that. Now what it's going to do is it's going to make a list, which is specifically what I asked for, but what I actually want is a table. So I'm gonna let it do its thinking. You can also give it a command to think about this or, um, you know, take it a step deeper and things like that. But for now, I just want to start with the initial list. You can see it gave me 20. So now I'm going to say, great, um, put that list into a table containing only the domains. And I'm just— so I can include them in a Google programmable search. So we'll go that way. Now it should just generate them into an actual table. Actually, it doesn't have to be a table, it could just be a, a list, and you specify a list with only the domains. Right, so now I can grab this list like that and basically just copy it. Now, I haven't closely inspected this yet, but I will show you what that would look like if you were to do that. Let's move on to a custom search here. I'm sorry about that. There we go. And I'm going to go ahead and create a brand new one. Just call it Quantum Transduction. It doesn't really matter what you call it because it's just going to show up to you. Search specific sites. In this case, I'll just start with the very first one here, which is the quantum.northwestern.edu, and add that. That's pretty much it. It's already created the custom search engine. Now, the next step is I'm going to go in and customize it. And the first thing I'll do, in this case the only thing I'm going to do, is to add a bunch of more websites as a list. And it says enter a list of sites, one per line. Here's where I'm going to enter that list. Now keep in mind that you already pasted this one as the first one, so we remove that one because otherwise it's going to come back and say, hey, that one's already there. And here's where you could go down and go, all right, well maybe I don't want anything in other countries. Uh, maybe I also don't want to pull up, let's say for example, uh, LinkedIn. Maybe I don't want to get any profiles from LinkedIn at this point. I've already done that. So we've got that list. It's, it's essentially got about 17 now since we removed the, the, a couple of them. Once you hit save, you'll notice that they get added to the list of domains here. It might take a minute, but, oh, looks like it's already detected one that was in here. Let's see if it's added some of the other ones. It did. Okay, so it added about 15 of them. Let's see, so it's got Quantum at Northwestern, it's got iowa.edu. I could sit here and just kind of cross-check these. I'm not sure why it said that there was a duplicate. It's probably a duplicate in the list. Um, this is the first time I'm doing this particular search engine. So if you run into that, just check for duplicates in the, in the lists. I'm a little surprised that that happened, but now we're here. From here, well, you can quite simply just go to the actual custom search engine itself and start searching. We can issue any kind of basic search commands here, like for example, we'll just use something like research scientist. I'm going to put that in quotations. You can also think of other job titles like principal investigator and things like that, but you'll notice that the results are only coming from those websites. That we specified. So starting with the job title, the best search strings are going to include combination of a job title, some sort of, let's just say, company name or organization name, and some jargon, some kind of technical language. Now in this case, we already specified the organizations being the websites, but we could specify a particular organization, or we could just add some keywords like, for example, condensed matter. There we go, condensed matter, and see if that brings down our results. And there you go, now it's about 731. Still a little bit too many, but look what we're starting to get. There's meet the researchers at Chicago Quantum Exchange. We're getting some jobs, which we can eliminate, and we're getting a few other authors from different places. So you wanted to eliminate the jobs, that's a pretty easy one. You can simply just eliminate it from the actual programmable search itself by removing the word jobs from the URL. That's one of my favorite ways to do it rather than trying to play games with, you know, using the minus or not symbol. So you can actually add websites to exclude and you can use jobs by simply adding an asterisk dot. In this case would be org forward slash and then jobs. This will remove the word jobs from any.org domain. You also probably want to do that with some of the other domains like.edu and maybe.net And so we've got.org,.edu, maybe.gov, and maybe we can also add.com, although I don't remember seeing a whole bunch of.coms there. So doing this will eliminate the word jobs from any of those domains. If we were going to try the search again, we should see that those are now gone. So we'll try once again, research scientist, and condensed matter. I should have probably just copied this before, but I didn't. And now you can see 700 results. It's still a little bit too many, but you'll notice that none of those job postings come up anymore. So this is— I don't like playing games with, with trying to eliminate and do a lot of not commands. So easiest way to do that is just simply eliminate it. Right from the root, which are the, um, the domains. Now keep in mind, we've only eliminated jobs. There may still be careers or a job or something like that. So it's up to you and your little mix of websites to identify that. But you can see how easily this peels back a ton of drilling around with a Google search, and it makes it super easy to create a programmable search for a specific discipline. To take this a little bit further, if you want to continue using the programmable search for these websites but add some, let's just call it refinements, what you can do is add keywords that would expand your search by including other similar terms, almost like creating your own custom dictionary, if you will. To do that, you can add either refinements, which we'll talk in just a second about, and/or also you can add query enhancement, which as you can see here, it's using synonyms. This is where we might go back to our ChatGPT and say, make a list of job titles that people who are experts in this field may have. We should probably specify, make a list of up to 10 job titles. Let's just specify. Otherwise, whenever you say make a list, you're sort of at the whim of whatever the interpretation is, which frequently is 10. But if you say 20, for example, that might be a little bit better. So now we see a lot of quantum repeated. You can easily remove that by saying repeat the list but remove the word quantum. And now you'll get a modified version of that with the word quantum removed. We can easily grab this and put it into our query enhancement here by— let's just add quantum. And we can add each one of these one at a time. For example, I just pasted it, so we'll start with material scientist, add that one. Then we'll do information researcher. Not really a big fan of that one, so let's go ahead and delete it. I really like the other one here that says research, research scientist, which we already knew was a good one. And then principal investigator is another one. Condensed matter physicist. We'll try that one and also do principal investigator, which is the other one that I liked a lot. You can keep going. You can have up to 10 of these synonyms that will be evoked when you utilize the word quantum. So let's go back to our search now and and utilize our new edit, which is pretty much just the word quantum. I'm going to use just quantum by itself to show you what it's actually doing. You'll notice that it's searching for the word quantum, but it's also adding some of those other keywords. So what we can do here is also add condensed matter. And we start to get much more accurate results. We'll add our research scientist, any other jargon that you can think of, any other keywords, maybe an organization perhaps. Um, a pretty good one would be like national laboratory to identify some of the national labs. And now that's essentially a company name without having to add each individual company name. So this is a way for you to really quickly generate a custom search engine using ChatGPT. You can even ask ChatGPT to actually help you. Uh, so provide instructions on how to create a Google programmable search engine. If you are not quite familiar with it or uncomfortable with it, haven't done a lot of it, it will literally just walk you through to, you know, go here, then get started, then add some sites, then add some synonyms, and so on. You can see it sort of gives you like bullet points. So this is just using basic ChatGPT. You notice I do have GPT-4o. I pay the $20 a month to actually get the advanced stuff. The biggest difference is that doesn't run out, or it doesn't like tell me that I'm out of queries for day, which I was running into a lot. That was like the main reason I decided to pay the $20 is I used— I got most of what I needed, but I used it so much that in a typical day it would come back with the, oh, we're going to limit your results now because you've used it too much. So that's really the main reason. There are a lot of other things you can do with the paid version. For example, like you can upload documents and you can do the web search directly and things like that. But that's the primary stuff. So, um, I know it's kind of a short one today, but I, I hope you enjoyed my little escapade into utilizing the ChatGPT desktop application, if you will. Drop me a note and suggest any other topics you might want me to see, uh, or covered here in Shelly's Alley. So, we'll see you next Friday. We should have our guest, Ryan Eakin, who wasn't able to make it today. He'll be back in the next couple of weeks. But next week, the guest is going to be Ray Thomas on December 20th. So, come back and tune in for that. Until then, be kind to yourselves. Take care, everybody.
More from Shally

Scaling Talent at Hypergrowth: Leadership, AI, and Hiring Excellence with Cindy Nicola









