Visage.jobs

Talent Acquisition Blog

recruitment | technology | innovation

The Implications Of GPT-3 On The World Of Talent

Talent Sourcing Webinar Series

Craig Fisher and Rory O’Doherty

Transcripts

Craig Fisher 0:00
Hey, it’s Craig Fisher and I’m here with Rory O’Doherty from olas.io. Hope I said that right. And I know Rory from his previous incarnation, where he was in a startup incubator, and Rory, why don’t you just tell us about you and what you do. And then we’re going to talk about GPT three today.

Rory O’Doherty 0:23
Cool. Thanks, Craig. And you got all us and my name, right? That was pretty impressive. Yeah, so my background, I spent the last three years at Tom tech labs, which I’m sure some of your viewers are familiar with. And contact labs started off as an incubator around time acquisition technologies, and then kind of pivoted into a research and advisory firm for corporate TA and staffing firms. So I lead the advisory services, they’re doing a lot of reviewing of technologies, really diving into how they work across corporate, large ta small organizations and staffing firms and. And then about three months ago, just as the pandemic was hitting, I decided that it was time to try and kind of bring something to market myself. So I started creating what’s called odesk.io. olis is the Irish for knowledge through experience. And the whole concept is empower users empower candidates with knowledge about digital careers, the skills they need, what the jobs will actually entail, what are they the real requirements and one of the kind of bonus requirements and educate them. So through short form video, educate them in those skills, make sure they’re hitting hard objectives and getting tangible, quantifiable experience in those roles in those skills, and then connect them up to employers for full time employment. And it’s an area that both ta leaders and candidates have been kind of talking about. about for the last number of years, and it’s something that no, I think needs to be in the world. So we’re going to bring it to the fore. And, and today, I think we’re talking about GPT 3am. I right.

Craig Fisher 2:12
Yes, you are absolutely. Right. So I love what you’re doing with your new platform. I think that’s great. And we’ve talked about this a little bit before. It’s sort of like social talent, but for job seekers. And I think there is a need for that. And so I think that’s wonderful. And I hope to be able to contribute.

Rory O’Doherty 2:31
Yeah, awesome. I appreciate that. And if I could ever get to that the quality and the level of the social talent team that would be, you know, a dream come true.

Craig Fisher 2:41
Absolutely. Well, it’s pretty good stuff. I know because I’m on there.

Yeah, so when, when I was with the Aegis and you were with talent tech labs, we looked at a lot of companies that were starting to incorporate natural language text into their bots. Right? And trying to do that outreach. GPT three is absolutely, you know, the next iteration, and it’s going to be a hot hot mess and hot topic.

Rory O’Doherty 3:14
Yeah, yeah. I mean, it’s, it’s already starting to get some media attention it, I think you’d be economists did an article about it a couple of days ago it is, you know, the very front of that, that hype train or the, you know, the hype cycle. So, we’re definitely gonna be hearing about a lot over the next couple of months. And, and, you know, hopefully we can dive into where it might be actually useful for ta data’s for recruiters, and where, you know, it’s where might some need some improvement.

Craig Fisher 3:47
So for those people who might not know at all what we’re talking about, once you explain briefly, what is GPT three, what are the applications and you know, what’s coming next year?

Rory O’Doherty 3:59
Yeah, sure. So GPT three stands for generative, pre trained transformer three, which is a mouthful in itself. And I don’t know, the data scientists don’t shouldn’t be, you know, in charge of naming things. But essentially, it’s a third iteration of this model that open AI research lab out of San Francisco has developed. It’s a text in text out models. So you, you input text, and it spits out other texts. And what’s really impressive about it is, is how natural language machine learning works, right? So there’s data that comes in, there’s a number of parameters and they match up and they’re weighted, and then they spit out a an outcome. That’s, that’s an approximation. Right? And up until now, the amount of parameters has been kind of limited to around just under 20 billion GPT three GPT three has 170 5 billion parameters. So it’s just getting extremely narrow in that approximation, obviously, for not obviously, but for good machine learning outcomes, you want high parameters, high variables, well waited, well understood. And then an absolute ton of data to get, you know, the big data pay on your sides. And maybe put that in context, Greg, for a second. You know, I studied economics and finance in, in university. And we used to model things, you know, I modeled em, breaking the entry in San Francisco, I believe, if I remember correctly, in order to try and help the police department there, figure out where they should be putting their their officers. I had five parameters. Anything above that, I couldn’t understand it, the model broke. And so these guys are doing you know, something a lot more impressive than that.

Craig Fisher 5:59
Yeah, it’s wild. I think Elon Musk is getting really interested in this. He, you know, he talks about our, the digital extension of ourselves, and, you know, that will be embedded in our heads at some point soon. And he wants to be, you know, along for the ride. And he feels like that that’s how we get on board with this, you know, the AI machine that is going to kind of take over according to him. And so we’re going to we’re going to strap ourselves into that and be part of the solution, not part of the problem.

Rory O’Doherty 6:35
So I could see that happening. I also think he’s gonna go mine gold on the moon or something as well. So you know, he’s got a few plans going on. Yeah. And, but if you think about AI, right, today’s the kind of three brackets of it, there’s narrow, strong, and then super, and narrow is where we’re at, right? It’s it’s augmented intelligence. It’s, it’s UiPath. It’s, you know, workflow. That are automated. And GPT. Three is at the far end of that, then like you got strong, which is more your iRobot more you’re kind of robots who are approximation of humans, or you can make that cognitive leap between one topic and another. And a super AI, which is what he was talking about what you’re talking about is, you know, when AI is much, much better than us at everything, right?

Craig Fisher 7:31
Well, and so that kind of brings me to my question. In a situation that where you’re going to use GPT, three, and you’ve got that many parameters, doesn’t someone still have to program those parameters?

Rory O’Doherty 7:44
Yeah, to an extent, like all of these modules, all these models need to be trained on some sort of training data. And the programming as such isn’t isn’t as complex as it has been in the past. And it’s only going you know, more and more that way. As, like web development has become more and more low code or no code. So with machine learning become a lot more accessible to your, your, your everyday person. And, but there definitely is a degree of programming. So you do need to feed it with examples of what you want.

Craig Fisher 8:19
Yeah, and I think that there have been statement writers for the simple AI that we have right now, doing this long enough, and cataloguing all of those examples, right, that I think they’re available, you know, for sale, right on the open market, you know, you can just kind of kind of buy those examples and feed it into your data, right?

Rory O’Doherty 8:38
Yeah, very much. So it will, it will require

you know, this is the start of the hype. This is the start of GPT. Three, it’s in beta for researchers right now. There’s gonna be some tweaks made as soon as it comes out live. I’m sure there’s gonna be people jumping all over it saying, you know, let us be your service provider. That is the straight For you let us do all the backend stuff for you. And it’s, I don’t want to go too technical here. But on the back end, it’s a simple enough API, HTTPS, and most back end developers should be able to, you know, make it work. Right.

Craig Fisher 9:15
Yeah. And so I think about it for, you know, recruiting tech, and employers. And so we could take our sponsor today, visage as an example, right? They’ve got this system where you put in a job and a group of 4000 Global sources bid on submitting candidates for that job and you get to crowdsource it sourcing, but on the on the user side on the employer side where you put the job and you get back 12 candidates for your first sprint or whatever it is, and a button to easily through that API or through their death dashboard right there to contact that person. Now, right now we’re writing the contact information or not the contact form Permission but the message hasn’t Yeah, yeah. But you know, it’s a smart CRM. So inputs first name, and it makes it a little personalized. I can imagine with GPT. Three, that can all be automated and fairly simply and really quickly, right?

Rory O’Doherty 10:15
Yeah. So that that’s the beauty of it, you know,

most recruiters or most talent acquisition professionals spend a decent amount of their time crafting these handcrafted a emails or LinkedIn messages or through massage their messages. And, you know, the hit rate is pretty good if you spend your time crafting each one manually. And it’s pretty bad if you just use a template and you just stick it in someone’s name, because there’s no you know, context there. And that’s where some recruiters, you know, get a bit of a bad name the industry, right, and the idea between behind GPD three, and how I could see it being applied to this kind of context is if you train it on it Couple of your previous messages, then it starts to learn your voice starts to learn your tone and your be able to speak with your authenticity, then you just start typing a little bit of whatever email you want to send. And it would be able to create a message you want to send, and you should be able to create a very good approximation of what that will be. And so the the inputs that you have to give it, just a couple of words. It’s the same as the way I kind of conceptualize it is, when you’re typing in Google, Google spits out five or six different, you know, auto filled answers to your search. This is literally billions of times more accurate than that. And that blew everyone’s mind. So think about what we can do going down the line with this automated messaging.

Craig Fisher 11:53
So I can I can envision

doing a a wonderlic type assessment And giving that information to the GPT three engine and saying learn my voice and learn my mannerisms and learn my motivations and how I would address someone from this assessment. And it can be as simple as that.

Rory O’Doherty 12:14
Yeah, I mean, it’s all about looking at what the inputs and outputs are going to be. So maybe it’s a matter of taking that person’s profile through massage or through any other, you know, online database, and understanding what what, how they should be contacted, how they should be engaged, added to that, how you like to engage people, and then you’ve got a recipe for success. There’s a tool platform called human intelligence that kind of does that and i i can imagine one is chomping at the bit here. Yeah, I’m sure he is. And there’s a couple of others you know, Chris knows is a good one that tries to get at that at that, you know, unique way of approaching In candidates, and I don’t think anyone’s corrective. Yes, I might be, you know, slapping their knuckles for saying that. But, you know, it’s still an area to explore.

Craig Fisher 13:10
Yeah, I used to use the crystal nose tool a lot. And then they started charging too much money for it. So I quit. But I think human intelligence takes it to the next level. And then once somebody gets their hands on GPT three, who knows how far could go That’s amazing.

Rory O’Doherty 13:25
Yeah, it’s really cool. And like, when you think about the so that’s the outbound messaging you think about, you know, where your expertise might be, and, and kind of recruitment marketing and inbound. You know, job descriptions are the bane of this industry. They’re, you know, pretty painful to use. Most organizations that I’ve dealt with, tend to have a bank of them on a shelf, take them off every now and then you know, dust them off, clean up few words, change the tea and on the bottom and update the salary maybe and take them out into the world. That’s not best practice. And it’s not good practice either. And but with gP gP t three, and once again, we’re going to name it something a lot cooler. And the idea is that you can take those old job descriptions as training information, and then just describe what you want, right? I want someone who’s a purple squirrel, who has all these different characteristics and a skills, create a job description that’s going to, you’re going to catch this person going to catch their eye. And then obviously, it’s up to the recruiter or to the recruitment marketer to get it out there. But it’s one piece of the puzzle.

Craig Fisher 14:43
So I’ve been thinking about this for a while. This has the capability to create a personalized job description on the fly based on somebody’s profile, and where they’re clicking in from.

Rory O’Doherty 14:57
Oh, that’s an interesting way of looking at it. Am I I don’t see why not. Right? The

I mean,

it’s all about getting the training data in there. So if you have enough info on that person, if they’re coming in through, if they’re coming in completely cold, right, and they don’t have any cookies, or they’re, you know, they’ve never touched that career side before. might be a bit challenging, but if you have the data on them, yeah. 100%

Craig Fisher 15:28
Yeah. So my thinking is that, you know, apply with LinkedIn kind of died. But if you’re still applying through LinkedIn and dragging your profile data over, then I don’t see why that couldn’t work. Or even if you uploaded your resume in PDF format first, but I don’t see that that you do that before clicking on the job link. So click on the job link from your LinkedIn profile. Then, in that method, it could work

Rory O’Doherty 15:57
or if you’re, you know, signed into indeed or if you’re You’re signed into Glassdoor any of the job boards, right? If you have a profile on any of them. And then also if you if you go one step further, and you think about this as a leaves on career sites and is powered by someone like phenol or someone, and why would they aggregate data across different, you know, suppliers, that no matter who you’re coming in from, you get a personalized job description. That’s, that’s pretty cool. That is arguably in the future a bit, but but really quite a neat way of thinking about it. Yeah. Awesome. Yeah,

Craig Fisher 16:33
I think I think we’ll see that in our lifetime, which, you know, let’s be honest, we don’t really see that many very revolutionary things in recruitment technology.

Unknown Speaker 16:45
Ah, yeah, that’s, that’s probably fair.

Rory O’Doherty 16:51
So I think we see a lot of innovative things coming from marketing tech and advertising tech. And yeah, you know, I’m I’m working with another patient right now that is got technology that’s from the armed forces and they’re trying to monetize it into our commercialize it into ta but yeah, just pure innovation for photon acquisition is not at the forefront. That’s a

Craig Fisher 17:19
well, I’m excited about it i think you know, I I get excited about a few things and you know, clearly design was one of them, which is why we’re doing this series. But I think that the the implications for GPT three are fascinating. And I’m studying the The Economist article that we read about it is just incredible. So for for our viewers and listeners if you haven’t read it is it is the economist, right?

Rory O’Doherty 17:50
Yeah, it is the economist. I’m I’m quite peeved because I wrote one about two weeks before that, and I just hadn’t hadn’t hit Publish. They got it out first. So yeah, read the economist one and then go read my one and see if we both talking at the same time talk. And, but I do think like, there’s a lot of as you’re saying, there’s not a huge amount of raw innovation in this space, but there is a lot of shiny object syndrome. And so I would be keeping an eye on this hoping to see it being incorporated into different products. Ideally, people are, you know, talent analytics people, ta leaders will started looking at it and seeing Okay, should this be in our product? Or should this be in our tech stack? And so the to go, there is one other use case great that I think is kind of fascinating, which is a generating search strings, Boolean strings, you know, I practice that, by all accounts shouldn’t really be a practice at this stage. That’s right. And so if you had Set up, and you just typed in natural language, who you’re looking for what kind of character, what kind of salary, etc. Then first step I see is, you know, it spits out all the parameters that you’d be able to use in a Boolean search. And then the second step is just just a skip spitting it out and just doing it right. Good doing it. Right. Yeah,

Craig Fisher 19:21
serving up. And I think you’re right, I think that here’s what I would say about it, because I’m a sorcerer at heart myself. And it’s okay to automate things and serve it up to you, but hunters want to hunt. It’s why people get into recruiting, right? And so, you know, you can’t take it all away, you have to leave them a little bit of something to play with and discover and and conquer, which I think, you know, we’ll find ways to do that.

Rory O’Doherty 19:47
Why 100% So, Hank, that M.

I mean, ultimately, you need to need to connect with these people. You need to craft the message correctly. You need to be able to understand their motivations and actions. convince them to engage, have a call, take the job, whatever it might be. There. So a lot of lot of art and science to this. Oh, yeah.

Craig Fisher 20:11
Yeah, absolutely. All right. So Rory, if people want to read your article, where can they find it? And how can they get in touch with you?

Rory O’Doherty 20:18
And I’m not 100% sure where the article is posted just yet. So I’m gonna have to talk to talk to my PR people. And I’m sure if you Google my name, and GBT GPT three, it will probably come up. And if you want to talk to me, I’m rory@olis.io. And we are bringing out the MVP of the product towards the end of this month, and Fingers crossed. It’s helpful.

Craig Fisher 20:43
Looking forward to that, and thank you for being on with us today. I appreciate it. And it’s great to see

Rory O’Doherty 20:49
cheers, Greg. Really appreciate it. Have a good day. All right.

About Visage

Visage is a sourcing technology that combines the expertise of 4,000 sourcers and AI to deliver high-quality candidate profiles within hours. Founded in 2015 by Joss Leufrancois, Visage simplifies candidate sourcing and outreach so you can focus on what really matters – people. Providing flexible sourcing solutions that can quickly scale up or down according to your unique hiring needs. Joss has spent over 15 years in the recruiting industry – after 10 years heading a top global recruiting firm, Joss decided it was time to revolutionize the way recruitment was done. Traditionally recruiters spend up to 60% of their time on robotic sourcing tasks: boolean queries, reviewing resumes one by one, finding contact details, and reaching out manually. Joss created Visage, to efficiently locate qualified and interested candidates by building passive candidate pipelines and automating multi-channel engagement (email, LinkedIn, text, and ads).

Share on facebook
Share on twitter
Share on linkedin

Weekly on-demand webinar series dedicated to sharing case studies on today’s most innovative practices for crowdsourcing top talent

Book a Demo