Episode Transcript
[00:00:00] Good morning everyone. Today I want to spend a couple minutes recapping the last week's activities at the Unleashed conference in Europe, which is one of the largest collections of senior HR leaders in the world. And some of the big announcements from OpenAI and a few other tech things. On the AI front, it's becoming clear to me that Research Labs, as they're called, OpenAI, Anthropic, Deep Seq, Google, maybe you might include Microsoft in that, are beginning to have problems figuring out how to monetize their products because they essentially are investing over a trillion dollars in infrastructure, multiple trillions. If you look at OpenAI itself, there's one discussion going on that Sam Altman wants to invest 7 trillion in his infrastructure and he's already well along along those lines. And much of their revenue is coming from licensing the platform technology to other software vendors, who in turn license it to us or our companies. But if they can't monetize it to the actual end users, that's a little bit of a house of cards. So what happened in the last week is something that I just want to point out to you. In the OpenAI case, an article came out from the Information which discovered that the head of product at OpenAI is the woman who used to be the head of product for Instagram, and almost 20% of the employees at OpenAI came from Meta. Now Meta is the example of the big platform company in the last technology wave, that is social networking, who basically decided that the way to make money was to sell advertising. And the Meta people multiple times have made it clear they think the advertising business is by far the best business in the world because you really don't have to serve customers, you don't have to have an enterprise level salesforce. And it's infinitely scalable as long as you get enough eyeballs. And the way the advertising business works is the more eyeballs you get, the more ads you sell. By the way, if you've noticed, I don't put ads on our podcast because our business model is not driven by advertising in any way. It's actually the opposite. We're more of an authority and we promote and sell and value our based on our expertise, based on our knowledge, based on our data, et cetera. So anyway, but in the case of OpenAI, it looks like they're going the other direction. The product Sora that was announced recently, the video generation tool, is more or less TikTok, except using AI. And you're going to see ads in there, you're going to see ads within the OpenAI experience because it's going to recommend and answer questions around who's advertising with them. It's to going. And even though they've now got an open platform to promote third party apps, which I talked about last week, I think it's called DevKit. The third party apps that come in through OpenAI are probably also going to have to pay a fee, sort of a renting fee, just like the way Apple charges 30% on their stuff when you buy something through Apple. That is a weird negative trend from my perspective because what it does is it disincents OpenAI from figuring out how to create the right answers to the questions we ask. And I have specifically noticed we stopped using OpenAI as the primary search or L L M and Galileo, you can still use it if you want, but it starts with anthropic. Now is that OpenAI makes a lot of mistakes. I find in my detailed research stuff it is not accurate, it's not attributing answers to the right sources and sometimes it apologizes but doesn't really do what it's supposed to do, that's kind of this kiss of death for them. Because since Microsoft uses that technology, I wouldn't be surprised if Microsoft's not too long into the future going to replace OpenAI with something else. Although maybe they can't. And you know, those of you that use it can make your own judgments, but it's an interesting problem. I don't know if there's enough revenue from consumers directly and even businesses to justify this multitrillion dollar investment that's been made. Just to give you a sense how how big these numbers are. The capital investment in AI infrastructure in the United States alone, which is over a trillion, and I think it's going to be close to probably 1.5 trillion by the time the year ends is 3 or 4 or 2 and a half to 3% of the GDP of the United States. Would you spend 2 and a half to 3% of all of your money on your AI search engine? I kind of doubt it. They're already way over their skis. So, you know, this may pay off in other forms of monetization, but I'm a little bit worried that more and more of these AI vendors are going to go in that direction. You know, that aside, we don't do it that way. We have built Galileo in the opposite form that it is completely focused on value and quality data. We did two workshops at Unleashed that went extremely well. Those of you that came, I'm sure You feel the same way I do. Galileo, as an example of a quality, business centric, focused AI, is starting to generate even more software success in the market. We're going to publish a bunch of examples of what other people are doing with it, and I think that's where this stuff needs to go. If I look at the work we're doing with SAP, the work we're doing with Microsoft directly, the work we're doing with many of you, the quality of the answers and the richness and depth of the answers are worth paying for. And that takes effort on our part to find the right data and tune and train the model so that it is useful to you guys and. And to build the right user experience so that you can easily build applications or use it. That's a completely different problem from Instagram. Completely different problem. And when 20% of the employees met at OpenAI come from Meta. I gotta wonder. Okay, I'll get off my little soapbox on that. I did a podcast last week on the fears of AI. I can't believe how many people asked me about that in Europe. I think a third of the questions I got from all of the hundreds of people I talked to had something to do with fears.
[00:06:05] Fears of job loss, fears of AI taking over the world, fears of errors and so forth. You can listen to that podcast. But just to summarize again, AI is a tool. It's not a being, it's not a person, it's not an object, it's a tool. It's a statistical, probabilistic tool, at least the way it works today.
[00:06:24] And you're going to have a lot of important responsibilities in training it. We had a really fascinating presentation by IBM this week in one of our internal groups talking about their experience with AI and hr. You know, they're way, way down the learning curve, but they still have a lot more to do. And one of the things that the presenter talked about was the dozens of new jobs that were created to manage and monitor and utilize the Ask HR agent that IBM built. It didn't start it as an agent, but it became an agent. And there's a lot of things to do to run these systems. So there's going to be a lot of jobs created and important jobs created to manage these things. One of the people I talked to right before I left Paris was telling me that they had the Microsoft copilot and they were trying to use it to build a policy database. And he said it was a nightmare because as soon as he connected it to SharePoint it found every version of every policy they'd ever created, including the draft versions and the versions that were never published and so forth. You know, kind of a messy problem that if you don't have your policy database up to date, clear and governed by the right owners and monitored correctly, it doesn't matter how good the AI model is, your employees are going to get the wrong information. And actually, the IBM folks talked a lot about that. Not only have they created policy owners for the 6,000 HR policies in IBM, but they've. They're starting to build an. A policy agent that looks through all the policies and then monitors the policies against regulatory changes in all the places they do business. Anyway, that's the big story here is how do you get quality data into these things? Not can you buy something off the shelf and cross your fingers and hope that it works? So anyway, that's a little bit on that. The second big area of discussion that's hot at the moment is the complex thinking skills and the skills of using AI. How do you train somebody to use these things? And what is the ultimate direction we're going to go? And let me go back to a conversation I had a little bit earlier in the podcast about recruiting. You know, a lot of the recruiters have told me that the testing and assessment that they've done in the past is starting to fail because the candidates are using AI to take tests and school teachers basically are finding the same thing. If you give somebody a writing assignment, kids will essentially cheat. Although I'm not sure it's called. I wouldn't call it cheating. I would just call it using the tools available and they'll write the answer or the paper automatically and they won't read it or learn anything. Given that that's going to be the standard part of our life, that there's going to be an AI button everywhere you look to write something for you. You could argue. And somebody asked me this, a couple people did that. We're going to become stupider.
[00:09:02] I don't know if that's a word, because we're not going to have to write. And writing is a thinking process. Writing is not really a writing exercise. It's a thinking exercise. And that is absolutely true. When you create or generate something with OpenAI or another LLM and you don't think about it, it's okay if you're just trying to consume information, but if you're trying to publish information and create things, you didn't learn anything from doing that. All it did was randomly jumble together tokens and spit out a bunch of words and you didn't learn anything from that at all. So what are the skills you need in AI in addition to learning about prompting and how the engine works and data management and all that? It's complex problem solving, it's didactic Socratic thinking, it's debating, it's data analysis. Early in my career, one of the most important developmental experiences I had as a young man was in high school when I was on the debate team. My English teacher in freshman year in high school decided I was going to join the debate team. I was very shy at the time, that was the last thing I wanted to do, but he forced me to do it. And it was really transformational in my life. Not only did it help me with the shyness, but really taught me how to think, how to do research and put together a logical argument. And if you've ever watched high school debate, you'll be amazed at how much intelligent, complex thinking goes on in high school. And I've basically used that skill almost my entire life. I did have a lot of jobs where I didn't need it. I had sales jobs and some tactical marketing jobs. What I do for a living and what I think a lot of you want me to do is I collect information and I debate and discuss what with the providers of that information, what it means, and then I share the findings with you. Now I don't, I would call that a lot more than complex problem solving. I think it's analysis, it's inquiry, it's curiosity, it's pattern matching, it's applying information against the real world and making sense of it and using the historic memory and wisdom that I have to try to draw conclusions through. There's a word for this, I don't really know what it is, but if you don't have these kinds of skills, the AI is going to lead you into the weeds. You're going to assume that it's giving you the right answers and you're going to look stupid in a meeting or you're going to make some bad decisions, or you're just going to add no value at all. You're going to be creating AI slop. So you know, these things are dangerously self confident. And this dangerous self confident nature of the AI platforms challenges your thinking. And if you're not somebody who's used to thinking deeply, maybe you didn't go to college, you didn't study English or math or history or get involved in some of these non linear parts of the world, you're going to have to get there from here because these things do not know what's right and what's wrong. They don't care. They're just looking for statistical relationships. They that's why the social networking aspects of AI are so dangerous because you know, you go to Twitter or one of these other biased social networks, everything is generated by AI based on patterns from other people. So it's self reinforcing errors, mistakes, inflammatory comments or exaggerations and but it does it in a very confident way. So you have to be logical and question these things. And every time you use AI to generate something, unfortunately you have to read it and see if it's even close to what you wanted because your name is on it. So I think there's a lot of complex skills that are going to be needed to manage these things as an end user, not as a data manager to make sure they're doing what you want them to do and eventually you'll become more confident in them and you won't worry so much. In the case of Galileo, I don't question it that much because I know the corpus that's in it and it doesn't really make mistakes because there's nothing in it that could encourage it to make mistakes. If you click a button and open it to the Internet, which you can, then you're more likely to get some weird stuff in it. But it's so deeply trained on the workforce, HR management, leadership, business data that it has, it's unlikely to do that, but it could if you opened it up to the Internet. But on its own it won't. But any system that you create inside of a company, which you're training yourself as an organization, you're going to have to understand how this data management stuff or how these answers are created and whether they're correct. And I don't want to just see any more posts on LinkedIn about the phrase complex problem solving. That's a vague, kind of meaningless concept. It's much more than that. I look at it as skepticism, analysis, challenging the reality, challenging the answer. It's really the scientific method of, of hypothesis testing, hypothesis testing, seeking truth. That's really the skill set you need. And I think my father and my background as a scientist has a lot to do with the way I think, not that everybody's going to be that way and not everybody's going to want to be that way, but we're going to have to be more that way because these things are just like I said, they're dangerously self confident and they're good writers, so they speak English or whatever language you speak very well and sound like a pretty good propaganda. They sound as confident as JD Vance, if you know what I mean. And we need to be careful that we don't always pay attention to them. Now, that leads me to maybe the last point I want to make from unleash the use cases of AI. I would say based on all the companies that I talked to and the various presentations I heard of and some of the ones I went to, there are a lot of great things going on out there. There are a lot of good use cases. But what you find in the companies that are doing great things is they're not just buying something off the shelf and seeing a result. They're putting a team into managing it, curating and training the system correctly, arranging and organizing the right data, watching very carefully what users do to make sure they're moving around the corners to the directions of questions people are asking. They are working on the architecture of how these systems connect together, which is starting to work. The MCP protocol is starting to work. It's starting to come to market. They're making sure that their data is, has good governance and they're working with the vendors that they trust. What I mean by trust is not that you trust them as humans, but the vendors that have experience with the scale and scope of the problems that they're trying to solve. Not every AI system is good at every part of hr, because HR is basically about people and management. So there's hundreds and hundreds of use cases.
[00:15:43] So I think we're reaching a point where the market's becoming a little more mature and you can be more discriminating on the tool sets you buy and the vendors you work with based on those tool sets. Companies like BetterUp that focus on coaching perceptics, who's focusing on employee listening, Sana and us in the area of learning. Galileo in the areas of HR advisory and HR consulting and benchmarking. The more, maybe narrow vendors are the ones that are probably going to add more value. And even at SAP, when I was talking to SAP about our integration with Joule, they told me they have more than 40 LLMs that they use in different parts of the system for different applications because the functionality of the AI needed is different and the underlying engine has to be different. That's kind of where this is going in terms of your daily life. I wouldn't worry so much about these things. I really think the biggest risk we have is you making a mistake, believing something that it says and making a decision that's incorrect or perhaps dangerous because you didn't question the system and what it told you to do. And we have to do that on behalf of our employees as HR people. But as an end user, you need to do it also. I would say the final thing is there is a lot of uncertainty and trepidation about AI out there in the HR world, and I think that's actually healthy, because this stuff is new. It's not as mature as you think. And even though it's widely used by the consumer population, the business users are going to be more discriminating and more demanding. And that's a good thing. Next year, things will mature a lot. Okay, that's it for now. Talk to you guys again soon. Bye.
