Episode Transcript
Speaker 1 00:00:06 Hi you guys. Today I want to talk about chat, G P t. We have to talk about this cuz it's all over the internet. And even though it's very, very early, I wanna give you some perspective so you know what's going on. So what is this? First of all, A G P T is a language model. It stands for generative pre-trained transformer. Not a very meaningful phrase, but basically what it is, it's a bunch of algorithms engineered by many, many software engineers and mathematicians that looks at the frequency count and the adjacency and the usage of words and doesn't know what a word is by the way. It doesn't understand a word, but it knows a word as a string of characters and it, it knows that a word has relationships to other words and it knows how they go together. And basically it's like a language machine.
Speaker 1 00:00:57 It has no knowledge of its own, but it's very good at understanding language and answering questions based on a corpus or database of knowledge. And so what's been going on at Open ai? Open AI is a company, it it, it is supposed to be a non-profit company, but I'm not sure if that's true, is they have been working on this for many years, I don't know, five or six years. The source code came from Google originally. A lot of the source code is public domain and they have been training it, training, it means indexing and looking at data with large amounts of data to the point that it seems to be pretty intelligent. And I have used it and, and asked it a few questions like what are the best practices for recruiting? What are the best practices for training? Who is Josh Buren, et cetera.
Speaker 1 00:01:46 And it's answers are, you know, pretty good. They're fairly elementary, but if you didn't know anything about the topic, you could consider it to be an education and training system. So you can imagine that companies like Microsoft who owns a piece of open ai, Google, Salesforce, ServiceNow, I mean, anybody who's in the technology business is trying to figure out how to use this. And based on what I understand of the roots of this, there's a lot of this code out there. And there are a lot of AI engineers that know how to build something like this. So the engine itself, although open AI is getting 90% of the press, is actually kind of a commodity. I know that sounds odd. It's kind of like saying any piece of software you buy is a commodity because somebody else could engineer it. And that's actually true. The problem is of course, when you go from Workday to Oracle to SAP to ADP to Paycom, they have different engineers, different designers, different target customers, different use cases they're studying.
Speaker 1 00:02:46 So there will be a lot of these G P T types of systems and most likely they will have different strengths and weaknesses. Now, the rumors of Google doing this are true. I know some people there, there's a, a pretty interesting article about a product called Sparrow, which is one of the Google projects that does this. And it's advanced beyond what G P T does. Apparently it does some things that G P T doesn't do, for example, that it has ethical boundaries. The engineers created a manifesto of ethical rules. And so the Google one, which is not available in the outside, outside world, will not do things like it, will not give you financial advice, it will not give you medical advice, it will not discuss race or gender or bias issues. And there's about 10 or 11 of these ethical rules and they've figured out how to code that into the algorithm.
Speaker 1 00:03:37 I'm sure it's not perfect. I'm sure they're gonna come up with all sorts of problems, but that's one thing that's different. A second thing that's different is the more advanced AI systems. And this is by the way, true for talent as well. If you look at eightfold, gloat, retrain, tech wolf, workday, et cetera, the advanced second generation AI learns. It doesn't just index more and more data, but it learns from the users. So in a chat interface, there is chat engineering going on so that the second question you ask makes the first questions answer smarter because the system knows that it didn't quite get it right or the user is asked, did this answer, how well did this answer work? What was wrong with it? Was there an error with it? Et cetera. And that can go back into the engine and improve it. So these are self-improving, self-healing systems if they're engineered correctly.
Speaker 1 00:04:28 They do take a lot of compute power and one of the things that's holding up some of the startups is they need a lot of data from a reliable source and then they need a lot of compute power. So the VCs are funding dozens of startups working on this. And I, my experience as an analyst over the years is there's no clarity as to who's gonna lead this at all. And I have no expectation in the open AI is gonna be the leader. Clearly there's rumors of Microsoft getting behind it and throwing it into Azure, which would be great, but I'm not sure if they're gonna make that bet yet because they don't own all of open ai, they own a piece of it, they don't own all of it. Google is very significantly working on this. My personal experience with Google Photos has been pretty spectacular.
Speaker 1 00:05:14 If you've ever used Google Photos, you can see how the AI gets better and better and better to the point that you can ask a a question like, where was Julia in January of 2018? And it'll find a photo of Julia, my daughter. And that's pretty good at it. It wasn't like that when I first started using it, but it is now. Now what does this all mean for us in HR and HR tech? It means a lot. There are many applications of this, and let me just mention a few of them that I've already discovered and many more to come. First of all, the chat interface is obviously very useful for candidates during the recruiting process. Self-service tools for HR service delivery, q and a, databases for problem solving, training, education and so forth. Um, let me tell you about the training part of it, which I think is gonna be one of the biggest.
Speaker 1 00:06:07 So if you look at our corpus in the Josh Burson Academy, for example, in our research library, we have thousands of research reports, case studies, articles, podcasts, videos, PowerPoint presentations and so forth in the corpus of Burson, all about HR training, learning, leadership, culture, tech, et cetera. I happen to know where it all is because I developed a lot of it and I just have a good indexing system in my mind and in my computer. But every time we build a search tool and we figure out how it's gonna work, it, it has strengths and weaknesses. So we've categorized our content into different types and so forth. Well, if we, by the way, we are gonna do this, if we took a tool like G P T and we pointed it at our corpus, it could be a very, very intelligent education tool and problem solving and answering question and answering tool for hr.
Speaker 1 00:07:02 Now if you just did that on the internet, it probably wouldn't work very well because the internet is filled with ads and marketing and all sorts of BS that people have put up over the years. But if you take a very narrow domain of information, that's very powerful. And let me give you another example that is very, very applicable to many of you. There was a company five years ago, a really smart guy that spun off from Facebook who tried to build one of these. He couldn't quite get it working, but he got doggone close. And what his system did is it read through a corpus of knowledge, created a chatbot, but more than that, it created learning tools. It created assessments so that you could test yourself whether you understood what was in there. And then it created summaries or learning outlines. And so his first client was a large bank, I won't mention the name of the bank and guess what they used it for compliance.
Speaker 1 00:07:54 They have a massive database of compliance rules and articles and use cases. And of course what happens when you join a bank is you gotta go through training, learn all that stuff and you never quite know all of it. Well, you know, he, they were convinced this bank was convinced this was gonna solve their problems and they were gonna allocate about a hundred million dollars to these guys to just, you know, use this. They couldn't quite get it to work and it didn't quite happen. And he lost patience and went off to do something else. But now that's possible. So you could 0.1 of these systems to your sales training, your product marketing, your manufacturing process, your IT process, your compliance, your safety rules. I mean, think about a big oil company or a big bank, a retailer, all of this knowledge and information that we've tried to design into courses by manually creating instructional design, which is a problematic domain.
Speaker 1 00:08:52 Anyway, these tools could do that. And I, I saw one this week that did it. I saw a vendor, I won't mention the name of the vendor cuz they're about to change their name that did this. This is a vendor that has deep mastery level courses in it, ai, machine learning, data science, things like that. And they took open source code spinoff from G P T stuff, pointed it towards their knowledge and created two things. They created a quiz system that automatically creates quizzes, which by the way is a actually pretty high value. And then a q and a that you could ask questions like a teaching assistant. I mean everybody would like a teaching assistant on something. And the way theirs works is at the end of the answer, when you get the answer back, you tell the answer how well it met your needs, and you give the system advice on how to make the answer better so the answer gets better.
Speaker 1 00:09:41 And these guys, this is a small company, they're not very big, they're smart as hell. So, so there's a lot of opportunity here as far as the market goes. There's gonna be a scramble of activity. People are gonna wanna partner with open ai, Google's gonna announce something, people are gonna wanna partner with them. A lot of this is gonna be platform as a service. Some of the big AI vendors in our domain in hr, eightfold, gloat, Workday, retrain, tech, Wolff guy, hi, they didn't build chat. They built skills, inference skills, adjacency, job architecture, intelligence and things like that. They're all looking at this, they're all gonna figure out if it's something they want to get into. I don't know if it's a domain they will get into, it's probably a different problem, a language model. But if you look at Paradox, for example, which is the leader in chatbot driven recruiting solutions, they have been spectacularly successful.
Speaker 1 00:10:39 Paradox is gonna turn into a really big company because of what they can do in recruiting. So just in all the domains of HR and recruiting in training, imagine if you had a leadership coach that has read every leadership book. Well not everyone because some of 'em are kind of stupid, but read a lot of highly curated leadership books and you could ask questions about leadership or scenarios as a manager to make you a better manager. I mean, this is not that far away. This stuff really could happen. So I am really bullish on it. It's one of these markets where in the early stage of the market, the apparent winners are oftentimes shot in the back, as they call it, by arrows from the fast followers often overtake the pioneers. That's a very common Silicon Valley kind of phrase, but AI is a pretty big company, so you're likely to be around a long time and they've obviously put a lot of their code out there for others.
Speaker 1 00:11:34 If Microsoft gets into it, which I'm sure they will, you know, they have a massive corpus of data because they can go through office, all of your office files. Now the problem with Microsoft is Microsoft has this product called Viva Topics that actually was trying to do this. Most companies have a lot of junk in their, their Microsoft libraries. I don't think you want to read every email I've ever sent. Some of them are great, some of them are not. It would be a pretty messy database. So we have to do some work deciding what domain we want to point this technology at, and then creating a quality process for selecting data and improving data without it getting messed up. Because most of the negative articles about chat G P T talk about things like information warfare. Imagine if the Russians used this and created a corpus of knowledge on conspiracy theories about the United States government or conspiracy theories about the vaccines.
Speaker 1 00:12:29 And then put it out there as us q and a on government policy and put a big American flag on it. Well, I hate to tell you this, I'm sure they're working on this. They could make life pretty miserable for us. And you could do the same thing too, if you inadvertently 0.1 of these things towards a database that's incorrect or misleading or we're gonna have to train these things. Okay, the second thing I'm gonna talk briefly about is jobs. So of course there will be a flurry of people all freaked out about the fact that this is gonna do away with all these jobs and we're gonna have all this unemployment, blah, blah, blah. No, no, no. That's not gonna happen. I've told people this hundreds of times. I actually told everybody in the Netherlands this at the conference I was at. All evidence shows that when technologies like this are created, way better jobs are created than those that are eliminated.
Speaker 1 00:13:20 Software engineers. Satin Adela was quoted this week as saying that 80% of the code of software engineers can now be automated. That doesn't mean that 20 percent's not important. It's probably really important. It just means that software engineers doing something a little bit different. They're not writing routine code. Think about my job as an analyst. I do a lot of writing of things that I've said before and repeating over and over again and rephrasing and sometimes very frustrating to keep saying the same thing. I don't wanna do that. I wanna advance the state of thinking. I want to stitch together ideas that people haven't figured out yet. I wanna look at new data. So my job is gonna be affected, customer service jobs are gonna be affected, sales jobs are gonna be affected, recruiter jobs are gonna be affected, but they're all gonna be up-leveled.
Speaker 1 00:14:00 And let me give you the evidence of why this is true. Number one, we have an unemployment rate of 3.5%. None of this wonderful AI is eliminated any significant jobs. They've made jobs better, the self-driving car doesn't exist. And so forget about that. I don't know if it'll ever exist. I think Elon Musk totally is overselling that. And by the time they do get something working like that, the truck drivers will be doing something else and then we're still gonna need truck drivers. Second thing, as some of you have seen in the predictions presentations I've done, by the way, the predictions that's coming out next week, I'll do another podcast on that. If you go back to the 1960s, and I did this work and I have a chart to show it to you. And you look at the inflation rate of services in the economy and plot a curve of how much it's gone up and it's a steadily upward sloping curve.
Speaker 1 00:14:48 And then you do a similar curve over the same period of time of the average hourly wage of a worker in the United States. So you have goods and services inflation on one chart, wage inflation on the other chart, the curves diverge, the wage inflation one is going up faster. And over that period of time, wages have increased 31% faster than general inflation. What that means is that jobs are getting better. The average pay relative to inflation has gotten better over the last 35, 40 years. And I think that is because of this technology. I think this technology is making jobs better. Now, if you're stuck in your role and you really like cutting and pasting things from the internet and doing routine work, you're not gonna be happy. So it's not good for somebody who doesn't wanna do something different and learn how to add more value, but just get over that.
Speaker 1 00:15:47 That is our job in HR is to make sure people understand this. And we're working right now, gonna do some work with eightfold and some others, and we're gonna put together some materials on the jobs that are gonna be the most affected by this. And I'm sure you're gonna see this from lots of other people too. Okay, well that was 20 minutes I, I didn't want to go too long, but I wanted to give you the introduction to this. There are lots of dark clouds here in the public domain. There will be complaints of bias, there will be concerns about gender and other forms of racism and so forth in these systems. There are laws being passed, there's a new one in New York that is an anti AI discrimination law. And vendors are going to have to prove to the authorities that these systems are not biased.
Speaker 1 00:16:33 And that's hard to do when there's so many use cases they haven't studied yet. But I think that problem's gonna get solved. I just go back to my use of Google Photos and it just gets smarter and smarter and smarter. Sometimes it's irritating things it recommends to me, but in general it is getting better. And I think the big companies, the Googles, the Microsofts, the IBMs, they don't want these things to do anything wrong. And the number one risk they would have is if this AI created an error or created a bad experience for a human being. So everybody's going in the right direction. So that's my perspective on this. The next podcast I will put out in the next couple of days, we'll overview the 2023 predictions. The 2023 predictions are coming out next week. This is something I tend to work on for months.
Speaker 1 00:17:22 It was written right up to the beginning of this year. So it's very current. It is a planning document. It is not my pontification of the future. I try not to be a futurist. I wanna show you where things are really going and I hope you download it and read it. It's free. You don't have to have a membership to get it and pass it around your team and talk about it. It should help you plan your investments in management and HR and technology for the coming year. I am more than willing to talk to any of of you about this topic of the chatbot ai and we'll continue to keep you up to date on things that I'm learning in the process. Thank you.
