Shally's Alley Live

Welcome, welcome, welcome! You are on Shally's Alley. Today is January 13th, Friday the 13th. Yep, I know, scary. The first Friday the 13th of the year. Actually, how many Friday the 13ths do we actually have in a year? That's a good question. I might, might need to ask Alexa that. Hey Alexa, how many Friday the 13ths are there in a year? Oh, according to Alexa, 1 to 3 times a year. Okay, well, anyway, this might be the only one this year. So here we are, we are live on Shelly's Alley. I don't actually have a guest today. But I do have a topic. I'm trying to lower the music here so it's not too loud. There we go. All right, hopefully that's not too loud. Please, uh, join me today in asking questions if you, if you would like to try some experiments. My topic is ChatGPT. In case you don't know what ChatGPT is, it is an open-source AI chatbot that can do some pretty interesting things. Now, I'm not a huge believer in the whole concept of artificial intelligence, especially because we're really not at a point where artificial intelligence is fulfilling the promise of really being able to synthesize unstructured information. But what ChatGPT OpenAI team have been able to do is to create some structure in an otherwise very— excuse me— they're very unstructured environment. So they've created all these models that are applicable to a robot that essentially can interact with humans in a little bit more of a natural way. There's still quite a quite a few things that it trips up with. But what they're doing is they're actually outsourcing— this is pretty smart— or rather not outsourcing, but crowdsourcing. They're crowdsourcing the error checking of this. So by opening it up to everybody, they are allowing everyday users like you and I to go in there and interact with it and share whether or not something actually worked. You can give a thumbs up or thumbs down if something had the answer that you expected. So where is this ChatGPT? Well, it's actually, although it's public, you do have to register through your, I think, your Google account. Let me publish a link to this here in the overlay or whatever that's called. I forgot the name of it. But you know, the text overlay, I think they call it a ticker. There it is. So here goes, chat.openai.com/chat. So once you go there, it'll ask you to register. Now this has been in the news and all kinds of places, so it's actually a little overwhelmed with, with participation, which is actually— this is a good thing. But what it means is that the response times are a little bit slower than normal. When I first started testing this, it was pretty fast. Now it takes a minute to kind of come up. And then the other disadvantage is that it doesn't remember previous conversations that you had with that, with the chat, other than the one that you're currently in. So if you log out and come back, it has erased the memory. But in the same session, it does remember, so you can continue to have a conversation with it. So it's still pretty cool to check it out, and there are actually some really useful things you can do with it. There's another one though, and that is the playground. So today I want to show you a couple of examples of how I've been playing around with ChatGPT from a recruiting point of view, and also what I can do with the playground that is the sort of test bed for ChatGPT. And it has even more exciting applications for recruiting and for sourcing, believe it or not. There's some pretty cool stuff. So I think it'd probably be a good idea for me to share my screen on, on this here and maybe just start from scratch on this rather than show you what I already did, because if I try to show you what I already did, you're probably not going to believe me. So here we go. I'm going to go share screen and switch over. So this is the basic interface when you land on it. I'm going to zoom in a little bit here to make it easier on your eyes. Hopefully this is not terribly confusing. So basically, the new chat is here. And all you do to start a chat is just type something in here. Now you can ask questions, you can ask pretty straightforward questions, like the same kind of questions you could ask one of your virtual assistants. But you can also ask deeper questions, like something, for example, one of the examples I give you is explain quantum computing in simple terms. So you can say, explain software development in simple terms or something like that. So let's try that. Let's see what it has to say. Now, here's where it's slow. Normally, it's a little bit faster, but it's actually not that bad. You can see it starts creating an answer that's pretty reliable. There are some gaps and some weaknesses that I've run into, but generally speaking, they're more advanced uses that you might want to think about. We're going to go ahead and delete this one here so that it forgets it. Now I want to show you the examples that I wanted to share with you today. The first one of those is to simply rewrite an email. So I'm going to look at my templates real quick here. So, you know, I'm not, you know, pre-configuring this or faking it in any way here. I've got a couple of templates for outreach. So here's one example. Let's see, I can use this one here. There we go. Actually, no, I'm not gonna use that one. I'm gonna use something a little bit more, a little bit lengthier. There we go, this one here. Alright, so I'm just gonna copy an old template and I'm going to type in the command rewrite this email colon and then paste the email. Now, um, you might wanna remove the placeholders for, for merge fields. So just, you know, hi Bob. And I think that's pretty much the only one. So there it is. So rewrite this email, hit enter. It's pretty cool. What it does is essentially, um, uses its, its database or its models to make the email a little bit easier to read, sort of synthesizes it and summarizes it. And the outcome is actually kind of impressive. So it's starting to rewrite it, and we'll just give it a minute to finish. It made the first paragraph a lot shorter and a lot simpler, but it didn't change the meaning, which is fine. We don't want that. When it gets to the end here, I want to show you a couple of things that you can do with this. So after it's done its original rewrite, you can say things like change the tone, make it shorter, make it more exciting, things like that. So what I'm going to try here is once it finishes, I'm going to try making it shorter. You can just say— this is one of the things that I am really impressed with, is that you can say make it shorter, where with other chatbots and other robots, you would really have to say, make it 10% shorter or rewrite it with X amount of fewer words. You have to be very specific, but I can actually just say, make it shorter. It'll rewrite it in whatever it considers to be a shorter length, which is pretty cool. It starts to abbreviate the content. The next thing that I want to show you is I want to make it more exciting. Looks like it made it a little bit shorter, but not much shorter. It is a bit small— sorry, slow. Normally, when I was testing this out, it was quite a lot faster. This is just because of all the users that are in it right now. You can see it's considerably shorter. I didn't do a word count or anything like that, but it does look shorter. So now I'm going to say, make it more exciting. And now it starts to actually rewrite the message much more than before. Where before I was just summarizing it and synthesizing it, now it actually is starting to change the tone a little bit and making it a little bit more enticing. So you can see what it's beginning to do. Are you ready to join the team that's revolutionizing the industry, etc.? I'm going to stop it because I think I'm going to use one example that doesn't have as many bullet points in here. See if I can find another one. It doesn't have as many bullet points. That one's got a lot of bullet points in it too. That one doesn't. That one still has bullet points. Well, let me use this one though. It's a little bit better. Actually, this has the About Us section. Yeah, I don't like that too much. That's okay. Let's stick with this one. For a moment because I want to show you a couple of other pretty cool things here as well. Oh, right, sorry, I went to the wrong place here. This is not the same ChatGPT as before. I went to the other one. Apologize for that. So what I want to do now is I want to— hey, where did it go? I guess I have to start over. Remind that. Okay, here we go. So now I'm going to say rewrite it. Check this out. Rewrite it in the voice of Tony Stark. This is kind of funny. It's funny and cool, but you can, you know, rewrite it in the voice of or in the tone of and things like that. Going to change the game. See, pretty cool, huh? So that's one thing you can do is have it rewrite emails. However, it's not just about emails. You can also have it rewrite entire job descriptions. So I'm going to use that this job description right here that I, that I had saved for this, and I'm going to make it a little bit more exciting. So here's a job description. I'm gonna just type in here, I'm gonna delete this and start over. I'm gonna say rewrite this job posting actually and make it more exciting. I'm gonna paste the job posting. Right here. Zoom out a little bit so you can see more of what's going on. So this is the original job posting. Join our mobile engineering team, we're looking for talented developers, etc. Now here is the rewrite. Pretty cool, huh? Are you a talented iOS developer looking for an exciting new change? Join our team, play a vital role in shaping the future of our platform. It's pretty exciting. I think it's fun stuff. It even summarizes the bullet points so that they're not as bullet pointy. This actually can turn into a pretty cool email, although a little bit lengthy. Although this is long, it is a lot better than the original job posting, which was much longer, and now it reads a little bit more like a message. I can take this one step further and say rewrite it and make it shorter, or make it shorter, or make it more exciting, and so on. I can give it all kinds of other commands to tell it to make it more into an email. So let's do make it much shorter. See what it does. It's crashing. Okay. Reload here. Make it shorter. See what happens. Oh, right, I forgot. I forgot the other one. So I've got to reload this. There we go. Make it shorter. Now, this is the one that it just rewrote. And let's see how much shorter it actually makes it. So it's 1, 2, 3, 4, 5 paragraphs. Kind of lengthy for an email. Actually pretty lengthy for an email. Let's see what it does. Wow, look at that, one paragraph. Pretty cool, huh? So there's another thing that we can do with this aside from, um, you know, rewriting it and making it shorter. You can even try saying like maybe write a pitch for this job or write a recruiting email for this job. So I'm going to now say write a recruiting email for this job. And take the job description again. And I'm gonna, I'm gonna start over, so I'm gonna delete this one. That's how you start over. And I'm just going to paste in— it says write a recruiting email for this job, and then here's the job. Let's see what it does when I tell it to do the job from scratch. Dear candidate, we're excited to offer the opportunity for you to join our mobile engineering team. Our app, used by millions of users monthly, is acknowledged for its innovation and influence of marketing. Look at that, it's pretty cool, huh? It just took that job description and made it into essentially a pitch email, which I think is, um, pretty smart. I mean, we all can do this, so it's not replacing something that we couldn't do. But what's cool about it is it's, it's giving you a, a head start. It makes it a lot faster to rewrite content, and you can tell it commands like, you know, make it more exciting, make it more colorful. You can add, you know, additional information to it. So there's plenty of room for the human to still have impact on changing, you know, the way that it looks. It's not literally totally automated, but it does help us arrive at something that's usable much faster. Here's another thing that I thought was really interesting. I'm gonna delete this one again here and start over. This one is essentially a mock interview, which I thought was pretty cool. So I'm gonna basically tell it to— what, what I'm saying here is the paste here so you can read it on this screen. I want you to act as an interviewer for the position of Talent Sourcer. I will act as a candidate and you will ask me interview questions for the Talent Sourcer position. I know it sounds a little weird, but you have to kind of learn to talk to the robot. I want you to only reply as the interviewer. Do not write the conversation all at once. I want you to only do the interview with me. Ask me the questions and wait for my answers. Do not write explanations. Ask me the questions one by one like an interviewer does and wait for my answers. My first sentence is hi, and then I'm going to paste the position description. I was reading that to you so that you wouldn't get lost with the position. So I'm still pasting the position description, and now I'm going to hit enter. So I'm telling it, interview me for this job, basically, and I pasted the job. Let's see what it does. Hi, can you tell me about your background and experience in talent sourcing? Okay, that's a pretty good question. Not the best question, but it's a pretty good one. Uh, I'll say I have, uh, 25 years of experience in talent sourcing for some of the world's largest employers. Okay, see what it comes back with. The silence is deafening. Great. Can you walk me through a specific instance where you had to find difficult to find talent. Pretty cool, right? Like, it's just coming up with these questions based on the requirements. Okay, sure. I once found power lines— power line workers by looking up their testimonials. I think it's testimony. Yeah. In court cases, period. See, once typo there. Hope I don't get docked for the interview. Can you tell me why you have to use persuasive omnichannel communication techniques to engage qualified? See, it's, I mean, it's not like making things up that aren't already in the job description, but it is pulling out the questions that, you know, make sense. So I think that's pretty cool. Anyways, enough about that. Let me go back and see if we have any questions on the chat there. Let's see. It helps. Okay, so we got it helps. I don't know if this whole thing— yeah, it helps with editing and creating high-level information. Sometimes I say carry out a sentiment analysis and remove ambiguity from the write-up. That's a nice one. I like that one. This one was about the line character. Actually, it did not. That was all the text. It pasted all the texts. I don't know about the line character. Let's see, you have to type continue. Ah, right. No, it did not limit it in this case. That actually was the end of where it says per process qualification. I just didn't have a period in the job description. That was the end of the text. But yeah, if you run into that, you would just type in continue. Cool. Now let me show you this other thing. This is still ChatGPT, but it is not the public beta. It is the playground, which is still public. I don't want to imply that I have any special access that other people don't have. It's the same thing, but it is a testbed environment where a lot of limits are removed. The reason I find it useful is, first of all, it's better for doing things with code like I'm about to show you, but it also has no lag in response because it's not getting hit so hard. Let me push that up here and edit. Let's see. You go to this one here. It's a little bit different. It's beta.openai.com/playground. So beta.openai.com/playground. And when you go there, it looks different, might maybe even look a little bit intimidating, but don't be intimidated by it. It's really not, it's really not that intimidating. So here, You can control things like the different models. I've just been using the DaVinci model. There's a bunch of others, Curie, Babbage, there's Codexes. I've just been using the standard one. The temperature essentially shows you fewer or higher results and more or less complete results. Maximum length is the number of tokens that it generates. I think the maximum is 4,000, but then it'll shut down after a few examples because you're using a lot of tokens. I just left it with 256. Haven't had a problem with it. I could probably increase the length on some of the examples that I'm going to show you. The rest of it I haven't really played around with all that much other than just those, those first 3 settings. But what I do want to show you is how easy it is to scrape information. So the first one I'm going to show you is a website that is dev.to, which is a developer community. I've got it open over here, and I had the inspector open for a reason, but I'll close it. So this is a public site that did not— you can see I'm not logged in, did not require me to create an account, and it is a community website for developers, dev.to. Which has different, I guess, channels or groups for various specializations. In this case, this is the Vue.js channel. You can see here the, the official tag for the Vue.js JavaScript framework, and it has a few posts from individuals. So that's my starting point, and what I am doing is I'm saying scrape this website of that Vue dev.to/view and then scroll down 50 times. This is really not necessary, by the way. It's just because it— the reason is that this is an infinite scroll. So if you don't tell it to scroll, I think it only loads like a very small handful of results. I've already been on this website, so unfortunately it's already scrolled a bunch of them. But normally when you first land on here, it shows you only a limited amount. So that's why I had to scroll in there. Then I'm just saying scrape all the names and print it. Print it means print on screen, not like print to a paper. It just actually prints it on screen. If you just say scrape all the names, it commits them to memory, but it doesn't do anything with it. I'm just going to hit Enter here, scrape this website, scroll down a few times, scrape all the names, and print it, and look what it does. That's amazing. So I want you to think about something. These are not the, the username. It's actually identifying— it looks like it did scroll only so far, and I could, I could tell it to continue. But keep in mind two things. It's not, it's not ordered by the same order that you see them here. So you don't see Zoltan, and then Aaron. It resorts them, and I think it resorts them— I'm not exactly sure how, but it looks to be somehow alphabetic, maybe. I'm not sure. It's not the same order. However, the other thing I want you to realize is that it's not extracting the usernames. It's actually finding the name for the person by looking at the text. So it's, oh, it's looking at the, um, source code, and that's why I had the inspector open. So I'll show you what it looks like when I right-click and do inspect, which you can do with Chrome browser. You can see that inside of the source code, there's a bunch of elements like the author preview, which shows you the author name, and then it has this envelope with UI elements that are the, excuse me, the bullet points that are in there. I think I had that. Is this it or is this it? It was like, oh, here we go. There it is. The crayons and the UI element that shows the name and some of the other stuff. It's actually really digging into the code and it's super easy to just get a list of names which then we can go plug into, well, anywhere else basically. I did that with dev.to and then I also thought, I wonder if that's going to actually work with GitHub as well. Here's a list of people that are members of this— I guess they're contributors. I don't know what the proper way to describe it is, but they're contributors to this repo here, this vue.js. I was at this vue.js and then I went to the people, which gives me this list of members. The same thing here. Now, what I did differently with this one— sorry, trying to get back to my other page. There we go. What I did differently with this one is, uh, I gave it a couple of extra instructions, and I'll show you here why. So this is scrape this website, the GitHub one. Then I said scrape the links to all the profiles because this one has hyperlinks to these profiles. And then I said follow the links to all the profiles and extract the text attributes and print it. So a little bit different because I'm actually giving it like a second set of instructions. Don't just scrape and find the names. I'm saying scrape, find the links, go to the links and extract that, and then print it. And then it starts with links to all profiles. So I now have that, which that's not that hard to do. I mean, you could have just done that with, uh, with, um, sorry, the, uh, Instant Data Scraper. But then it starts to pull the text attributes. Look at that, the bios essentially, right? So it's really powerful and it kind of does some of the, the, the, the programming— using air quotes, not really programming— but does some of that, um, extra, you know, HTML-based or JavaScript-based stuff that, that is required to do. Now you can actually tell it to use particular programs as well. So for example, you could say scrape this website using Python, or with, with Python, then extract or find all the names, print it. Oops, print it. You can tell it to use particular code or applications. There's one called BeautifulSoup, which is a tagging parsing application. Let's see if this— I haven't done this before, see if it actually works. Nah, it stopped probably because of the— it doesn't know what I mean by with these. I probably need to add another command, but you can actually invoke other software as well that's out there. It's pretty cool, pretty powerful. So you start with the basic, you know, scrape this website and find all the elements, find all the names, things like that. You can tell it to essentially do manipulate things like a developer would, but without having to know any code, which I think is pretty amazing. You know me, I'm lazy, so the less work I have to do myself, the better. Anyway, I'm back over here now to see what we have in chat, see if we have any other questions. Oh, the shushing around. Sorry, I forgot to turn the audio off. Sorry about that. I had some music there when we started too. Not have quiet in the background, but I turned it off. So let's see, anybody have any other questions or things that you want to, you want to see me try? I've done, you know, answer questions about technology or to define things, you know, same kind of things that I've talked about before when I use my virtual assistants, but the answers are a lot more succinct. I think the biggest difference is it, it gives you a very high confidence answer. Instead of giving you sort of like— you probably have experience with, with, with, uh, your Amazon and Siri robots before where you kind of get the wrong answer. You asked it to give you an answer and then it gives you something completely different. So I find that this ChatGPT answers questions a lot more accurately, higher confidence. And the other thing that I found is it requires less framing things correctly so you can be more generous, like scrape this website, right? That, that gives me the liberty to be less concerned about how do I ask the question and more experimenting with different questions that I can ask to see what I can get. So anyway, hope you enjoyed today's Shelly's Alley. We don't have a guest, but we will have a guest on our next episode. I didn't have anybody that was able to commit to this time, So if you know anyone you want to be on the show, or you yourself want to be on the show, you know how to get me, you know how to get in touch with me. Go to the, um, Shelly's Alley website and, uh, or contact me via LinkedIn or whatever. And, um, we'll see you next week. Take care, everybody. Have a nice long weekend for those of you that have Monday off.