← The Artificial Intelligence Show
The Artificial Intelligence Show
#130: Worrying Future of Jobs Report, Altman: “Confident We Know How to Build AGI,” Why OpenAI’s Agents Are Delayed & Trump $20B Data Center Investment
01:19:52
Loading summary
Transcript59 lines
- [00:00]
Paul Raitzel
You're going to have graduating college kids who do not want to go work for companies that don't allow them to use ChatGPT. The idea of going to work for a company where you're not allowed to use AI, that's like a non starter for a lot of this talent. And that may actually be what we need to trigger faster adoption within enterprises when they realize like where do you get the top talent? It is certainly not by shutting off access to Generative AI Tools welcome to the Artificial Intelligence show, the podcast that helps your business grow smarter by making AI approachable and actionable. My name is Paul Raitzel, I'm the Founder and CEO of Marketing AI Institute and I'm your host. Each week I'm joined by my co host and Marketing AI Institute Chief Content Officer Mike Kaput as we break down all the AI news that matters and give you insights and perspectives that you can use to advance your company and your career. Join us as we accelerate AI literacy for all. Welcome to episode 130 of the Artificial Intelligence Show. I'm your host Paul Raitzer along with my co host Mike Foot. We have World Economic Forum Jobs Report, more interviews with Sam Altwin, a great article from Ethan Malik and a whole bunch of other updates. And it is Monday, January 13th as we're recording this and it has already been a busy morning for like major news items from the White House Nvidia, all kinds of stuff happening we're not even going to get to in this episode, but usually when the weeks start off like this. Mike, my what I've learned now having covered the space for over a decade, if Monday morning starts busy, that's usually a prelude to a whole bunch of stuff happening. And I would say there is a lot happening on a Monday morning, way more than normal. So I would expect we are in for a crazy episode next week when we have to cover everything that's about to happen this week. So episode 130 is brought to us by the AI Mastery Membership Program we've been talking about on the show. If you're looking to drive AI transformation for yourself or your company in 2025 should definitely check out our AI Mastery Membership Program. This is a 12 month membership that gives you access to education, insights, experiences and everything you need to really drive transformation in your career and your company. Mike and I do exclusive content for this Mastery Membership Program. There's a Generative AI Mastery series where Mike does demonstrations of Gen AI tools and apps. We have an Ask Me Anything session once a quarter which is a live ama. And then we also have quarterly briefings where we do sort of the top 10 trends. And there's a bunch of other stuff. There is. I'm, I've mentioned I'm working on plans for this Master Membership. They are very close. I'm not going to be able to announce them probably in the next week or so, but we are very close. We actually think we're going to probably roll out phase one of this potentially later this month. So stay tuned on some updates to the Master Membership program. And then there's going to be some other bigger announcements as we go throughout Q1 and into Q2. So it's a great time to get into this program. You can go to SmarterX AI/AI dash mastery, or you can just go to SmartRx AI and click on Education. And the Mastery Membership Program is under there. Pod150. We've had a lot of listeners taking advantage of this promo code recently, so check out Pod150. It's Pod150. That'll get you $150 off the annual membership. And there are enterprise plans available as well. So reach out to us if you're interested in Master memberships for your team. This episode is also brought to us by AI for Writers Summit. Writers who embrace AI will lead the future of storytelling AI writing tools are revolutionizing how we create, refine and scale content. I know Mike is doing it all the time as the Chief Content Officer at Marketing Institute, unlocking new possibilities for writers, editors and content leaders. Question is no longer if you should use AI, but how to use it effectively to stay ahead. At our virtual event, the AI for Writers Summit, you'll learn how to harness the latest AI tools to responsibly transform storytelling with speed and precision, enhance productivity without sacrificing creativity, and build strategies to future proof your career or content team. The summit is taking place virtually from noon to 5pm Eastern time on Thursday, March 6th. This is our third annual summit. Last year we had over 4,500 attendees. Crazy. The first year we did it, we had over 4,000. So it's just, it's growing every year. But so this is the third one. There is a free registration option. This is a key. It's a sponsor supported event. So there's a free registration option and then there's also a paid ticket option with on demand access. So go to aiwritersummit.com again, that's aiwritersummit.com and you can learn more about that program. All right. Interesting jobs report, Mike. Certainly A job for Notebook lm. I don't know if you'd used it, but the first thing I did was took that beast of a report and threw it into notebook LM. Was it 270 pages? I think it came in at.
- [05:09]
Mike Kaput
Yeah, it's actually 290 pages. It is a mammoth report from the World Economic Forum and it is called their Future of Jobs report. This is a highly anticipated report they release regularly and this one brings together perspectives from over a thousand leading global employers. And collectively these employers represent more than 14 million workers across 22 industry clusters and 55 economies from around the world. Now there are a ton of stats in here worth paying attention to. Like we just mentioned. Go drop this into NotebookLM and query it for yourself. But there are some really big ones that struck us that are related to AI. So this is kind of the money quote here. According to the report, quote, half of employers plan to reorient their business in response to AI. Two thirds plan to hire talent with specific AI skills and 40% anticipate reducing their workforce where AI can automate tasks. Now they are also looking to upskill existing talent where possible. In response to expected AI disruption, these respondents said that reskilling and upskilling of the existing workforce order to have them work more effectively alongside AI. This was the most anticipated workforce strategy by 2030, 77% of the surveyed employers plan to implement some type of strategy like that to make sure their workers are effectively trained to work alongside AI. So Paul, as we dive into some of the stats here, like that quote really puts it into perspective I think, like especially that part about the 40% of leading global employers anticipating reducing their workforce due to AI. Like, does that sound pretty serious to you?
- [07:09]
Paul Raitzel
Yeah. So I, I mean this is like we said, I would definitely put this into Notebook lm. It is the best way to interact with it and, and kind of like get, you know, run, do your audio overview and listen to like the 10 minute version of this thing. But the thing I always look at in these reports is this is obviously a vast study. 14 representing 14 million workers is not insignificant. But at the end they do break it down by country. Like you can drill in and go like, I think page 214 to 215 has the US data, so it takes its overall data, then it breaks it down. Now the thing you always have to keep in mind with these kinds of reports is they are relying on people who may not fully understand AI to answer questions about the impact of AI. So it's just like you have to always keep that in the back of your mind. Obviously a lot of the people they interviewed may have deeper knowledge of generative AI and different AI capabilities, but there's a reasonable chance that people they're interviewing don't actually fully comprehend the state we're in right now. So. And we'll kind of come back around to that theme throughout today's episode. But the one thing I noted is just I went to the United States pages like I mentioned, pages 214 to 215 and the question about share of organizations surveyed that identify the technology trend is likely to drive business transformation. So AI and information processing technologies. Mike, as you alluded to, 94 in the United States. So it's almost unanimous that everybody sees this as like the driver robots and autonomous systems. Now keep in mind it's a diverse collection of industries. So if you're in, I don't know, financial services, robots might not impact you. Your definition of autonomous systems may or may not, you know, decide that you would say yes or no to this question. So you always have to keep that in context. There was quite a bit about generative AI. And again this is where notebook becomes very helpful. Just you can drop it in, let it do its processing and say okay, talk to me about any generative AI highlights from the report and it'll pull those out and it'll highlight them and cite them and you can click through and go look specifically, specifically at those sections. So that's what I did. I did not read the full 290 page report, but I'll call out this excerpt real quick because it shows you the uncertainty that is present within these projections as we look out the next five years. So the report says while the full extent of long term productivity gains from the technology remains uncertain, workplace studies have identified various initial ways for generative AI to enhance human skills and performance. Performance. Looking further ahead, some observers argue generative AI could empower less specialized employees to perform a greater range of expert tasks, expanding the possible functions of roles such as accounting clerks, nurses and teaching assistants. Then it goes on to say, however, without appropriate decision making frameworks, economic incentive structures and possibly government regulations, there remains a risk that technological development will be focused on replacing human work, which could increase inequality and unemployment. It says gen AI currently remains limited in performing tasks that require physical execution, certainly nuanced judgment or hands on application skills rooted in human interaction, including empathy and active listening and sensory processing abilities and manual dexterity, endurance and precision currently show no substitution potential due to their physical and Deeply human components. So it kind of goes through and it's trying to assess like what are the skills that are core now, what are going to be the skills of the future, which jobs are going to like, increase, which jobs are going to decline. They did call out a study. They looked at data from Coursera and the report looked at data. I think that most of this research was conducted from May to August 2024. So it's relatively fresh data from research reporting standpoints. So they, in the Coursera data, they say Coursera data generated for the Future of Jobs report reveals significant growth and demand for generative AI training among both individual learners and enterprises. Demand for AI skills has accelerated globally. However, the drivers of demand differ. In the United States, demand is primarily driven by individual users, whereas in India, corporate sponsorship plays a significant role in boosting gen AI training globally. Individual learners on Coursera have focused on foundational gen AI skills and conceptual topics such as prompt engineering, trustworthy AI practices and strategic decision making. So this is interesting because as I've mentioned, I've been doing a lot of thinking around our own AI Academy and our, you know, courses and certifications and things like that. And I was actually having this exact conversation with our team last week when I was sort of sharing the internal vision for where we're going with this stuff, what we see and we've, we have, I mean we've, our intro day classes had over 25,000 people register for it. We have hundreds of members, I think over 500 mastery members at different times. Our piloting AI courses, over 1500 learners scaling eyes over 300, I think. So we have a decent sample size of our own data to look at. And what I'll tell you is it is a lot of individuals and corporations that are taking the initiative to learn this stuff on their own, sometimes on their own credit card and on their own personal email account. They are paying for these things. Sometimes they are getting supported by internal programs funded by those programs. But it is, it is rare for us to have the conversations and these are people often working at very large enterprises that are taking these initiatives on their own. It's rare that those conversations are coming in and saying we want to upskill our whole workforce. That, that hasn't been what we've seen at least in the United States yet. I think that that will shift in 2025. I believe that we are entering a phase where enterprises are going to realize the urgency to rescale and upskill their workforces. So I found it interesting when, when the report looked at, well, what are the core skills today? And it's interesting, they, they had like analytical thinking, resilience, flexibility, agility, leadership and social infants. Creative thinking are sort of the top four. But then when they say what are going to be the fastest growing skills in the next five years? Number one, AI and big data. Number two, networks and cybersecurity. Number three, technological literacy. Four is actually staying steady at creative thinking. So then it goes into, okay, which employers are planning to upskill their workforce. And it said 85%. So, okay, people get this. They then got into the human and machine factor. So now combined it says proportion of tasks completed predominantly by technologies versus predominantly by people. Today the combined is 52%. By 2030, they see that rising to 67%. And then it gets into some of the fastest growing job titles. Now this is kind of where I'll wind up my thoughts on this. The top fastest growing jobs. Big tech, big data specialists, number one. Number two, Fintech engineers, number three, AI and machine learning specialists. Now this is why I say you have to be cognizant of who's answering the questions. So we have business leaders here who may not be in the know of what all the AI labs think the future looks like and they're the ones building it. And so I think that there was an interesting balance here because on the Pod Joe Rogan podcast last week, Mark Zuckerberg said, Mark Zuckerberg said a lot of things on the Rogan podcast we're not going to bother getting into. But and what, what he said that's relevant here is, quote, probably in 2025, we at Meta, as well as the other companies that are basically working on this, are going to have an AI that can effectively be a sort of mid level engineer that you have at your company that can write code. Zuckerberg said Meta will reach the point where all of the code in its apps and the AI it generates will also be done by AI. So we have business leaders who like, yeah, you know, like 40% are impacted or like, you know, they think they've got time, they got to figure this out. And then you have the people in the AI research labs like, oh no, we won't need as many AI machine learning specialists because the AI is going to do that this year, not 2030. So there's always this balance. And then they did have, how will your business respond to AI developments? The number one answer was 77% reskilling and upskilling existing workforce to better work alongside AI. And then 69% hiring new people with skills to design AI tools and enhance appropriate for the organization specific skills. And then the third one was 62% hiring new people with skills to better work alongside AI. So quick side note, I mentioned a bunch has already happened on Monday morning. OpenAI actually released their economic blueprint this morning, which I have not had time to get into, but it says OpenAI is releasing a new economic blueprint that lays out our policy proposals for extending America's global leadership in AI innovation, ensuring equitable access to AI and driving economic growth across communities as AI becomes more advanced. We believe America needs to act now to maximize the technology's possibilities while minimizing its harms. So all that said, AI and the economy is off to a high profile start for the year and I expect this to be a highly trending topic as we move forward throughout the year.
- [16:21]
Mike Kaput
Yeah, I found it interesting too that the report looked at on page 63, like barriers to AI adoption, which is also something we asked about in our annual State of Marketing AI report. And the most common response at 50% of people was lack of skills to support adoption, which is not remotely surprising to us. Paul, that's the top barrier of four years in a row now. Conservative, it seems very conservative. And it's followed closely at 43% by lack of vision among managers and leaders. So basically validates exactly what we've seen in research. I also thought it was kind of interesting that that list of core skills and fastest growing skills, obviously there is technical literacy, AI and big data, but a lot of these really sound to me like powerful soft skills. I'm not sure people are getting degrees in some of these.
- [17:11]
Paul Raitzel
Yeah, that's. And that's the hard part. When you hear interviews with like Altman and people like that where they try and pin them down on this question. Yeah, like what should people study? What should my kids go to school for? It's a lot of like very soft answers. Soft skills and soft answers. It's like, okay, well what do you study to get those and how am I going to monetize those? Like, what are the jobs I'll be using those soft skills in?
- [17:35]
Mike Kaput
All right, our second big topic this week. So last week, if you recall, we touched on a recent essay written by Sam Altman. It's titled Reflections. In it, he made some really striking predictions about AI. So this week we wanted to actually expand on our discussion of that essay since we only touched on it briefly last week. And then we also wanted to mix in some other comments Sam made to Bloomberg this past week. In a new interview. So first in reflection, Sam revealed that OpenAI is now, quote, confident they know how to build AGI. He also said that because of this, the company is now looking beyond traditional AGI towards what he calls, quote, super intelligence in the true sense of the word. He basically believes super intelligent AI could dramatically accelerate scientific discovery and innovation beyond current human capabilities. He also predicts in this essay that we'll start to see the seeds of all of this kind of come together starting this year. He has repeated a couple of times now, both with Bloomberg and in this essay, that we're going to maybe see the first AI agents join the workforce and materially change how companies operate starting this year. In reflections, he acknowledged that these kind of predictions about super intelligence sound like science fiction to a lot of people. And he writes, quote, that's all right. We've been there before and we're okay with being there again. We're pretty confident that in the next few years, everyone will see what we see. So in a new interview with Bloomberg, he echoed a lot of these talking points. He talked up how the company's O3 model achieved a major breakthrough by passing the Arc AGI challenge. He repeated the prediction that 2025 is when we might see true AI agents join the workforce. And not to mention, he talked a bit about his feud with Elon Musk and his controversial million dollar donation to President Elect Donald Trump's inauguration fund. So, Paul, we're going to take a couple pieces of this one by one. And first up, like, I'll be honest, like, it is crazy to me that we have gone so quickly from even speculating whether or not AGI is even possible to Altman himself saying, basically, yeah, we know how to build it. He even went so far as to say in this Bloomberg interview, quote, it's impossible to overstate how non mainstream AGI was way back in 2014, right before they started OpenAI. He said people were afraid to talk to me because I was saying I wanted to start an AGI effort. It was like cancelable. It could ruin your career. So what changed?
- [20:17]
Paul Raitzel
Yeah, you know, we've obviously been covering this topic along the way. I think the models have gotten smarter. That's changed. There's been a lot more reports put out, evaluations done on the AGI spectrum, trying to assess kind of where we are. And so I think it's just become more commonplace and I think the labs are just talking about it more. And I thought it was really funny how this interview started. The OpenAI PR team obviously wants Sam out in front telling the story for some reason because the reporter literally says in the article. Your team suggested this would be a good moment to review the past two years, having spent a time in the PR world that is a reporter or that is a PR person reaching out to a reporter and saying, hey, Sam is available to talk. We think now is a good time to regroup this. And the reporter saying, okay, yeah, I'll tell the story you want told, but I'm also going to ask him all this other stuff while we're doing it. But I've never actually seen a reporter actually write that up front and say, that's why this interview is happening.
- [21:15]
Mike Kaput
And they're probably in the room or.
- [21:17]
Paul Raitzel
On the phone, oh, there's three PR people in the room. Like, trust me, you and I have done this before. So I don't know. Like, I thought it was really interesting to go back to like the AGI not being sort of like taboo to talk about. He kicks off the interview talking about how they recruited so much talent, like a density of like the best AI researchers in the world in those early days. And he said the pitch was just come build AGI and the reason it worked. And he the quote, I cannot overstate how heretical, heretical it was at the time to say, we're going to build AGI. So you filter out 99% of the world and you only get the really talented original thinkers and that's really powerful. So then you know his reflections post again. We talked about it a little bit last week, but we didn't go into this Bloomberg article much and there is just a lot of really interesting things in there. He retells the story of like his early dinners with Ilya and some of the formation of OpenAI. So if you haven't like heard the story of the early days of OpenAI, this is a reasonable representation of it. Like I would read Genius Makers by cademet's too. That tells more of the story. But so dinners with Ilya Sutskova, who is one of the co founders and now has safe superintelligence. He tells about taking over as CEO in 2019. He talks about the launch of ChatGPT, says, quote, I thought it was going to do pretty well. The rest of the company was like, why are you making us launch this? It's a bad decision, it's not ready. And then he said, I don't make a lot of we're going to do this thing decisions, but this was one of them. I actually, I think that's an interesting moment just from a leadership perspective. As the CEO and founder of multiple companies through the years, there is this fine balance sometimes as a leader where you do get committee thinking, like you involve people in it. And then sometimes as a leader you have, you have to trust your instinct and your vision and you have to just do something. And so I thought this was an interesting moment because we will have books in business classes for years that will look back on the origins of OpenAI and how it was all done and the decisions that were made. And so it was kind of cool to hear that little bit of insight. He talks about the company structure, his firing. He does get into the thoughts on AGI. I find this kind of frustrating, honestly, like the way he talks about AGI, it kind of drives me nuts because they're the ones that have been pushing this concept, literally their mission statement since 2015. And what is AGI? Well, when you ask Sam, this is the answer we get, quote, I think AGI has become a very sloppy term. If you look at our levels, our five levels, you find people that would call each of those AGI. And the hope of the levels is to have some more specific grounding on where we are and kind of the progress is where it's going rather than is it AGI or is it not AGI? So then the reporter says, well, what's the threshold where you're going to say, okay, we've got AGI now again, I would think OpenAI could answer this question. They've been asked a thousand times. So he said, the very rough way I try to think about it is when an AI system can do what very skilled humans in important jobs can do. I'd call that AGI. Then there's a bunch of follow on questions like is it a full time job or part time? Can it start as a computer program or decide? And it just kind of goes into like. And then, and then it says, now we're going to move the goalpost always. Which is why this is hard. I'll stick with that as an answer. It's like, stick with what is an answer that I. So I. So I guess if I go back to this, he is currently defining AI as a system that can or AGI is a system that can do what very skilled humans in important jobs can do. Which is not a great definition, but that is the definition that we are going with. There was also insight into the pricing of chat GPT, which I thought was fascinating. He said, we tested two prices, 20 and 42, 42 being the hitchhiker Guide to the Galaxy. Fairly confident they picked 42 for that reason. But he said people thought 42 was a little too much. They were happy to pay 20, so we picked 20.
- [25:12]
Mike Kaput
Okay.
- [25:13]
Paul Raitzel
He said it was not a rigorous. Hire someone and do a pricing study thing.
- [25:17]
Mike Kaput
Yeah. For all you data driven startups out there, sometimes people are just making it up.
- [25:21]
Paul Raitzel
This is how decisions are made, it's how brand names are picked, it's how pricing models are picked. He gets into AI safety Roadblocks to progress, which I think their economic blueprint they published today probably is tied to this. But he does say a lot about like, their work in chip building and developing their own chips and they're going to have more to say on that this year. Fusion, he talks about being a huge thing, nuclear fusion, and the need to reduce regulations around that because that'll help accelerate energy in the United States. In particular, the Trump thing, he made a personal donation and the reporter called him out on that. Like, you don't share a lot of common beliefs with Trump, like why would you make a personal donation? And he said, I don't support everything Trump does or says. I don't support everything that Biden does or says, but I do support the United States of America and I'll work to agree, I'm able to with any president for the good of the country. I actually thought that was a really good answer. I thought it's like, okay, I could respect that answer. Um, I think AGI will probably get developed during this president's term and getting that right seems really important. Supporting the inauguration, I think that's a relatively small thing. So basically he's saying like, the, the greater reason here is like, we, we just have to do this together. Like it's a very important time gets into Elon Musk, gets into US Infrastructure and the need for that to be the focus of this current administration. So the, you know, I would definitely read Sam's personal Reflections post because it came as a result of this interview. The, the guy said, like, you, you've, you know, written before, why haven't you written in a while kind of thing. And so he wrote the Reflections post as a result of this interview. And then the interview itself does cover, you know, quite a bit of topics. It's a, it's a good read in Bloomberg.
- [27:00]
Mike Kaput
His waffling on AGI seems really strange to me since literally in Reflections he says, I think we know, we're confident, we know how to build it, like wouldn't that. But we don't know what it is, assume a definition.
- [27:13]
Paul Raitzel
It drives me nuts the whole way.
- [27:17]
Mike Kaput
One last note here, like, just very quickly, like, how confident are you in his. You know, he keeps talking up these agent predictions where I'm talking a few topics about some delays on the agentic front at OpenAI, like, how confident are you that they're going to be joining the workforce and kind of changing how companies operate starting this year?
- [27:34]
Paul Raitzel
I, I think they will within AI research lab companies and frontier companies. I think that they're going to be heavy human in the loop. They're not going to be like the autonomous AI agents that everyone envisions. But we talked about that on a recent episode with the state of AI agents. It's humans train them, humans figure out the data, humans do the integrations. They set the work of the AI to kind of create what the workflow is going to be. The AI executes the workflow, the human oversees it to make sure it doesn't screw up or get access to something it's not supposed to get access to. Human helps improve it. Like, I think there's going to just be a lot of human in the loop, but you're going to be building a ton of these agents, formerly known as bots, but now calling them agents that carry out a string of tasks. And so I do think that those are going to start to make an impact in some of these businesses. I just, I don't think they're going to be as autonomous or as reliable as maybe they're perceived to be.
- [28:31]
Mike Kaput
All right, so our third big topic this week, kind of piggybacking on the aspects of this last conversation about Altman. Ethan Mollick just published an essay where he writes, quote, recently something shifted in the AI industry. Researchers began speaking urgently about the arrival of super smart AI systems, a flood of intelligence, not in some distant future, but imminently. He kind of calls these signals collectively quote, Prophecies of the Flood, which is the essay's title. Now, what makes all this interesting, he says, is that these rumors and signals are not coming from the usual tech hype cycle, but often from researchers inside major AI labs who appear genuinely convinced we're approaching a fundamental breakthrough, which we talked about a bunch last week in our super Intelligence segment. So he talks about some of the signals that we also discussed. Things like OpenAI's O3 model achieving 87% accuracy on a test where even PhDs with Internet access only scored 34% outside their specialty. It's solving incredibly difficult math problems. It is performing well on the ARC AGI test, and it's demonstrating fluid intelligence that can match or exceed human baselines. He also mentions we've started to see the emergence of practical AI agents. Malik especially talks about Google's Gemini with Deep Research as a prime example. We've talked about deep research quite a bit. For instance, he asked Deep Research to research startup funding methods. It analyzed 173 websites and produced a comprehensive 17 page report in minutes. So all of this, Malik believes, are these kind of prophecies of this flood of super intelligence coming. But despite all that, he says we should kind of temper excitement with realism. So even if the labs are right about reaching AGI, he argues the pace of actual adoption and integration into society will likely be much slower than the technology's theoretical capabilities. He also says that while AI researchers are focused on all the technical challenges around alignment and safety, there is way less attention being paid to how society should adapt to and shape the deployment of these technologies. So Paul, like one of the themes that kind of keeps coming up in my opinion in these types of essays and the commentary from the labs seems to kind of fall into this bucket of like very few people really understand what's coming. Like in your opinion, what is coming? Why do so few people seem to really get this?
- [31:13]
Paul Raitzel
Yeah, I actually saw a tweet this week from or last week from Vedant Misra who I've mentioned on the show before. He sold an AI company to HubSpot in 2017, which is when I got to know him. My agency at the time was doing a lot of work with HubSpot, spent two years on reasoning at OpenAI and now he's a reasoning researcher at DeepMind. So he tweeted there are maybe a few hundred people in the world who viscerally understand what's coming. Most are at DeepMind, OpenAI, Anthropic X, but some are on the outside. You have to be able to forecast the aggregate effect of rapid algorithmic improvement, aggressive investment in building reinforcement learning environments for iterative self improvement, and many tens of billions already committed to building data centers. Either we're all wrong or everything is about to change. So I think your question about like maybe why don't people get it? He calls it out like they don't understand how fast these things are improving. They don't understand the concept of self improvement, recursive self improvement. They don't understand the value of reinforcement learning to make these models really, really smart in specific domains and then more generally. And they don't comprehend how much money is Already allocated over the next five years to building out these data centers, which, as we talked about last week, are AI factories, they're intelligence factories. So when you take all that into consideration. Now, Vedant was actually a closing keynote at our Macon 2022 event. And I asked him in that session, why are you working on this? Like, because, again, AGI in 2022 wasn't common conversation. And so I knew Vedant's background. So not only did he built and sold the AI company, he. He was at Columbia University in theoretical cosmology, experimental astrophysics, and experimental particle physics. This is a really smart dude. So he could literally be doing anything in science and AI and business. Why are you working on reasoning and AI? And he said, just kind of summarize, because I believe a future of abundance is possible, and AGI is how we get there. And so he said, like, there's people like me who believe AGI is within reach, and we want to build it, we want to see it there. So I say that for context, because then Sam tweeted on January 10, which I think goes to what Ethan is saying. So this is the tweet from Sam. Prediction, the O3 arc. So O3 is this model that sort of set off the forthcoming model that sort of set off all this conversation around superintelligence in the last three weeks. The O3 arc will go something like this one. Oh, damn. It's smarter than me. This changes everything. Ah. 10 minutes pass. 2. So what's for dinner? Anyway? 10 minutes pass. Can you believe how bad O3 is? And slow. They need to hurry UP and ship O4. And I think that's what happens in, like, this world. Like, we have these. This insane alien intelligence, and, like, it's amazing. And then you just sort of go on with your life and you, like, forget how incredible it is. So I was. I was actually sitting this morning with my wife. I was trying to get ready and, like, drinking my coffee and getting ready for this podcast, and I was trying to, like, conceptualize how to explain to someone. So I bounced this off of my wife. I was like, hey, would this make sense if I was just, like, explaining this to you, like, what's happening? So bear with me for a second here, Mike. Tell me if this makes sense. So I think what's happening is many leaders still struggle to grasp the capabilities and potential impact of our current generative AI capabilities. So let's take one person on your team. We could do this with any business. Mike and I have done this. Like, give me any business Give me any team member, any person in an organization. We're going to assess what they do for their job. We're going to take like the 80%, like what are the three to five things you do for 80% of your work and then we're going to build three custom GPTs and train them how to use those GPTs to do what they do every day more efficiently. So let's assume we take a single employee, we break down their job into tasks and then we build three custom GPTs to help them do their job. You will unlock a minimum of 10% efficiency gains in their job within. I would. I'm trying to be conservative here. I'm going to say 90 days, but it probably within nine days. Like this could happen fast. So I'm being intentionally conservative to prove a point. I think in many jobs today that could easily be 20 to 30% gains if they're properly trained how to use these tools. So let's play this out for a second estimate. There's 22 work days in a month on average. 21.76 I think, to be exact. And you work eight hours a day. Very few people actually only work eight hours a day. But we'll use this for simplicity. So that's 176 hours a month. For a full time employee, we'll round that up to 180 for simplicity sake. A 10% efficiency gain would save 18 hours per month. Make sure I'm doing my math here, right Mike?
- [36:03]
Mike Kaput
Yeah.
- [36:03]
Paul Raitzel
Per employee. So One employee saves 18 hours. If you take a 10 person company with full adoption, we do the same process for 10 people, that's 180 hours a month. Or one full time employee a month has been saved by just doing this one exercise. In 100 person company, that's 1800 hours a month. Or 10 FTEs in a thousand person company, that's 18,000 hours a month or a hundred FTEs. In a 10,000 person company that's 180,000 hours a month or a thousand full time employees. This is with today's models. OpenAI hasn't improved GPTs in over a year. They launched these things. They added a couple of like tweaks to it. But if you give us OpenAI's existing GPT capability with their existing models, we could do this. Every single company in every single industry. I can't think of a knowledge worker we could drive 10% efficiency gains for. So now you tell me how AI won't completely disrupt the job market and the economy in the next two years. The Only way it doesn't happen is continued lack of understanding. We have an AI literacy problem. We talk about this all the time. People don't realize GPTs can do this, that it's that simple to build a GPT and assist your job. The other factor is human resistance to change. Yeah, these may slow things down in some industries, but the tech can transform. The probability of disruption is too high to do nothing because in this scenario we just need fewer people doing the same work. So the way we think about it is like everybody has this choice. You can keep doing what you're doing or you can accept that you can change things and you accelerate your AI literacy and capabilities and you become kind of AI forward. So if you want to go through this exercise for yourself, just this is why I created the Jobs GPT tool. So you can go to SmartRx AI and slash jobs GPT or just go there and click on tools. You can put your job title in there and it'll break down for you the different ways you can use generative AI and it gives you like an exposure key. I think this is critical. Stuff like this is. It is doable. Today we're doing it in our own company. I've said this before, we have eight employees now. We function like a team of probably 30 to 50 people in terms of what we're able to accomplish in a day or a week. And every company can do the same thing.
- [38:28]
Mike Kaput
I love how you laid this out. Not only the math, but also isolating what's going on here. Like either people don't understand the math or don't know this is possible, which is a lack of awareness. Or like you said, they do know it's possible, but there's resistance or there's not telling anybody. Right. It's striking to me that this is the absolute baseline scenario that is a layup today to do and it's already showing in the numbers. That's essentially a 10% reduction in the amount of people you need for at least the current stuff.
- [39:06]
Paul Raitzel
So if you don't produce more.
- [39:08]
Mike Kaput
Yeah, I would be. That would be eye opening alone to me, especially if I was someone. Understand, I understand the resistance, but unfortunately I think if you resist long enough, what's going to happen is someone else is going to make the decision for you of how your continued employment.
- [39:25]
Paul Raitzel
I would think it's such a competitive advantage for the companies that think about this and say, wow, like, okay, let's do that now. Again, our whole positioning is don't get rid of those 10 FTs or 100 FTES, like go into new markets, create new products, pursue campaigns you didn't have time to run. Like everybody's got more work than they know what to do with. So have a plan, be proactive in redistributing reskill, upskill like we talked about in the first topic, and prepare those people to, to create additional value for the organization. Don't just look at it and say cut. But I'm also a realist that publicly traded, venture backed and private equity owned companies will cut people. That is inevitable. We've, we're already starting to see headlines this year of plans to not hire any more engineers to reduce workforce from, and it's going to be from tech companies first. That's where we're always going to hear about it. And so you're starting to see that occurring already.
- [40:18]
Mike Kaput
All right, let's dive into this week's Rapid fire. So first up, Paul, you, back in March 2023, wrote this really impactful essay titled the Law of Uneven AI Distribution. In this essay, you suggested that the benefits of AI will not be equally shared across society. Instead, the value that people and companies get from AI will depend on three big factors. First, their understanding of the technology. Second, their access to it, and third, their willingness to accept it. Like for instance, some enterprises still block employees from accessing things like ChatGPT, which limits the employee's ability to use AI in their job. Some educational systems are struggling to adapt curriculum to account for AI capabilities. That in turn limits how much value students get out of AI. So these kind of institutional decisions create really dramatic differences in how people experience and benefit from AI technology, not to mention your own personal comfort can unevenly distribute benefits. If you're willing to share more information and data about yourself or your company with AI, you'll get more value out of it. So in light of kind of our main topics this week and the rest of what we're talking about in the rest of the Rapid Fire, this essay has been really top of mind. So we wanted to quickly revisit it. So Paul, as I was rereading this essay this morning, like, I have to say this definitely really holds up for me. Like every line in this sounds as true or more true than when it was published. I guess my question is like, why is it so important to revisit this at this moment in AI?
- [42:01]
Paul Raitzel
Yeah, I just think it's becoming more obvious the companies that aren't moving forward. And I think we're now getting to the point because I wrote this, what, two weeks before GPT4 came out, it was three months after ChatGPT emerged. I think we're now starting to see it play out. The companies that aren't moving forward, that aren't developing and understanding three literacy programs, that aren't giving access to the technology, aren't accepting some new risks that come with using this, they're starting to now fall behind in their peer groups, with their competition in their industries. And so I think it's now just starting to play out what we sort of theorized was going to happen. And so that post again we'll put the link in the show notes, you can go read it. But the post ends with three actions you can take and these still seem pretty relevant today too. The first is be curious and explore AI. So however you learn best, books, podcasts, courses, videos, whatever it is like go seek that stuff out and start taking the next steps to learn this stuff to figure out what data you, your family, your company. So in your personal lives, you and your family and business side, your company, what you're willing to give up in exchange for the convenience personalization productivity that comes with these AI tools. So if your company isn't willing to take on more risk then you're and give up some level of security, then you're kind of going to Keep not having ChatGPT turned on in your company, things like that. And then the third was consider your company's policies and principles around that AI usage and do what you can to ensure a balance between innovation and responsible application. So yeah, I think it's just good to like it puts a framework to why aren't all companies doing these things and those three components of the law, you know, acceptance, access and understanding are still as relevant today and now we're just seeing them play out.
- [43:45]
Mike Kaput
And definitely something to also perhaps consider in your career. If you're working for one of these companies that is restricting all these tools, you are probably going to be at a longer term disadvantage along with your company.
- [43:56]
Paul Raitzel
Yeah, it's hard to make the decision to change jobs but if, if you look out over the next three years and you do not see that your company embracing generative AI, I would seriously consider a career change. Like it's, it's going to get to the point over the next few years where if you have sat still it's going to become a problem. It's not yet you can go learn this stuff. You can, you can push within your existing company to get that AI council build, get some courses, you know, integrated in get generated policy like you can. If you feel like you can make change there, do it. But if you can't, and no matter what you do, you're just going to sit there. It's going to be a problem in the next one to three years for your career.
- [44:35]
Mike Kaput
Our next topic Box CEO Aaron Levy shared some really interesting observations in a thread on X about how AI is transforming enterprise technology. And this is based on a bunch of recent meetings he had with dozens of leaders at those companies. So, really briefly, here's kind of what he says he's seeing. Number one, the AI first enterprise is emerging. He says companies are starting to think through how entire functions get reimagined in a world of AI. Number two, enterprises want choice in their AI stack. They increasingly want flexibility in using different models that are good at different things. Number three, we need more interoperability in AI, especially with AI agents. Once those really come online, we'll need AI systems that are much better at talking to each other. Number four, your AI stack will define who you can hire. Kind of to the point we just talked about. He says AI will actually define employee choices in the future, with more AI native employees simply expecting that their company will enable them with AI to be as productive as possible. Number five, the role of IT will change tremendously. Agents will make it increasingly responsible for actually getting work itself done versus just deploying and maintaining software that enables the workforce. And last, but certainly not least, number six, it is still in his mind, quote, insanely early. In his perspective, it is inevitable that every enterprise will be transformed by AI. But this is going to take time. So Paul, he ends by kind of saying, overall, this is the most energized he's seen enterprises in nearly two decades of being an enterprise software. It's very exciting. We're on the cusp of major changes to how business and work happens in the future. Now, all of this is awesome and definitely we're seeing this kind of motivation among enterprises, but we've also seen like, how slow this can kind of be and how long this can take. Like, this will probably take a little time, no?
- [46:34]
Paul Raitzel
Yeah. So I think it's a really good post. I. I think the energized thing is interesting. I don't know who, like, it depends on who you're talking to. I don't know that energize would be the first word I would use to describe most enterprises we talk to about this stuff. Overwhelmed, uncertain would probably be like 1 and 2, right? Energize might fit in there for a couple of people. Within the teams, but it's usually not the whole team or the whole company. I hope that shifts like I hope the people he's talking to is what we start hearing more of. Couple of quick thoughts. Number four, I think this is going to be a major issue for companies. I think you're going to have the next crop starting this year with graduates that probably started a little bit last year, but I think more so this year you're going to have graduating college kids who do not want to go work for companies that don't allow them to use ChatGPT. Like it's just you, you get so used to it in your school work. The idea of going to work for a company where you're not allowed to use AI, that's like a non starter for a lot of this talent. And that may actually be what we need to trigger faster adoption within enterprises when they realize, like Sam said in the beginning, where do you get the top talent? It is certainly not by shutting off access to generative AI tools. So that's going to be a trend that HR department is going to have to deal with very quickly. Number five, I worry about honestly this idea that it is going to like build or hire all these AI agents and they're going to start controlling how the workforce is done and we're going to be less reliant on HR people and more relying on it. That's terrifying to me. I don't think it is the right group that should be dictating AI agents and what the org chart looks like. So I think that's a very unclear role how that's going to play out in different companies. But I do see it probably wanting more control when the AI agents are part of the org chart and I think that's going to create a lot of friction in companies. And then number six, that we're insanely early. I 100% agree with that. We say this all the time. People look around and they think that they're behind their competitors and that everybody else has this figured out. They don't like. There are so few companies that actually have this all figured out. So I think we are absolutely still early and I think all six of his points are worthy of note.
- [48:48]
Mike Kaput
Our next topic is about OpenAI and AI agents. So OpenAI is taking a notably cautious approach to launching its AI agents according to some new reporting from the information. Now, this has led to delays in the releases of OpenAI agents and agentic capabilities. However, the information says the company may finally release its computer using agent software. As early as this month. So what's been behind the delay so far? One primary concern, according to the information, is prompt injection attacks. AI agents are designed to automate complex tasks by taking control of your device. So things like going to make an online purchase or editing a spreadsheet for you. But there is a big security concern related to these activities. So just for example, imagine an AI agent searching for an outfit for you for your holiday party, but they accidentally land on a malicious website that tricks it into forgetting its original instructions and instead stealing your credit card information. This is an attack known as a prompt injection. The information says it's been particularly worrying for OpenAI. Now, what's really fascinating here is that competitors like Anthropic and Google have already started to launch versions of their own computer usage or computer using agents, where OpenAI has been the most hesitant one, focusing on developing all these additional safeguards. That seems really striking to me, Paul. Like, out of these big three, OpenAI, Google, Anthropic, OpenAI is the one I would probably classify as most likely to, like, move fast and break things. But they're the ones holding off the most on AI agents like Google and Anthropic. Why are they the first ones putting this technology out into the world? I mean, they talk so much about safety and alignment.
- [50:42]
Paul Raitzel
Yeah. So a couple thoughts here. So one, you know, it's so interesting we start off with OpenAI has taken a notable cautious approach to launching its AI agents. In this context, we are talking about the original meaning of AI agents, which was computer use, meaning it sees and understands what's happening on your screen, whatever device that is, and it can take action, as you explained Mike, in that example. Right. That was what we used to mean by AI. Then we like turned it into whatever it's become today that any AI system that can take any action, basically. So it sounds like they're specifically talking about the computer use form of an AI agent. Here is why I give that differentiation. I think Anthropic came to market first and we talked about it at the time they did it, which I think was in the fall, if I remember correctly. It was like October, November, we talked about on the podcast. And I said at the time I thought it was very odd that Anthropic, the one who's supposed to be most safety conscious and aligned, was the first to market. My simplified explanation of why is they were trying to raise billions of dollars and they needed to demonstrate the technology. Now, their technology is not connected to the Internet. It can't really do Anything like it was just basically a public demonstration. Google's is a little bit different I think. I haven't personally tried it yet, but it's like attached to the Chrome browser and it can do some stuff. But like neither of them have rolled out full blown computer use. AI agents, we are not there yet. And so I don't know, like it may just be that OpenAI was trying to get all their 12 days of shipments ready and they were trying to do Sora and they were trying to do the O1 model and they knew this stuff doesn't actually work and they did. They prioritize computer like the compute power to the things that they thought worked and that they needed to get to market that people could actually use. Like I may be wrong, but that's my best guess. Here is OpenAI knows this stuff doesn't work yet. It's way too risky. And they wanted to put their limited capability or capacity for compute power to work on Sora and O1 and probably training O3 and maybe Orion, like they just. And Maybe Orion, maybe GPT5 has computer use baked into it. Like maybe, maybe that's why that's taking so long is like they're actually going to come to market with a full blown computer use that actually works. I don't know.
- [53:12]
Mike Kaput
So next up this year CES Consumer Electronics show just wrapped up at the end of last week and it should come as no surprise AI dominated the show, but actually not really in the way you might expect because Forbes noted that AI was quote, both everywhere and nowhere. It was everywhere because all these companies talked up how AI enabled their products worked. But it was nowhere because there weren't nearly as many like visual demonstrations on AI, according to Forbes. Like something you might see for more flashy technology like VR or ar. To be sure, AI was powering many of the products seen at Caesar. But behind the scenes it just wasn't stealing the spotlight with flashy demos and experiences. However, an AI related player did steal the spotlight. Nvidia CEO Jensen Huang gave this big 90 minute like stirring keynote that covered a huge amount of ground in AI. He unveiled Nvidia's next generation Blackwell GPU architecture, introduced a new series of graphic cards which deliver twice the performance of their predecessor. And then he detailed Nvidia's industrial scale AI initiatives, particularly the Blackwell AI supercomputer system. This is a system that is in production across 45 factories worldwide. It combines 72 Blackwell GPUs to create what Nvidia calls an AI factory. Huang also announced Cosmos, their first found world foundation model. This is for physical AI and robotics. It's trained on 20 million hours of video and designed to understand real world physics and environments. And it's an open source platform that aims to do for robotics and industrial AI what Meta's Llama has done for enterprise AI. As if that wasn't enough, Huang introduced Project Digits, a compact AI supercomputer powered by their GB110 chip. And this is a desktop size system scheduled for release around May that brings supercomputer class AI capabilities to individual users and developers now. Paul, there is an overabundance of hype in AI generally tons of hype at ces. So we kind of wanted to focus just on Jensen's keynote here. Nvidia is one of, if not the most important company in AI, so his vision for where we're going is definitely a strong indicator of the future since him and Nvidia are building it. You're a longtime watcher of Nvidia and Jensen. Like did anything jump out at you here about CES in general and about his announcements specifically?
- [55:59]
Paul Raitzel
I expected way more buzz on Twitter slash X than I saw because I, I wasn't like looped in every minute to what was going on at ces and usually within, you know, my notifications and the lists I have, the stuff that matters would bubble up and I saw almost nothing on CES last week. It was weird.
- [56:16]
Mike Kaput
Like, you know what's interesting there is I, in one of the articles we have linked in the show notes here, someone mentioned like halfway through a bunch of people had to leave because of the LA wildfires.
- [56:27]
Paul Raitzel
Oh really?
- [56:27]
Mike Kaput
Going back home. So that's possible that that was overshadowing it. But I agree I didn't see nearly enough. Even when doing research I was like going through post on X and stuff. It was, it was weirdly muted and quick.
- [56:40]
Paul Raitzel
Side note on that, my goodness, the you know, thoughts go out to everyone affected by that. We have one of our employees is out in la, not, not directly impacted but right on the outskirts and we're watching, you know, paying close attention like everybody else. It's ongoing situation just so tragic. It's like so hard to comprehend what happened there. And like yeah, just look at these pictures and I just can't comprehend and now like everything that's going to into like the rebuild for the years ahead. So just yeah, anybody out there has been affected. You personally, your family, your company, certainly thinking of you and you know, hopefully figure some stuff out we can get rebuilt quickly out there. So you know, back in the ces, yeah, it just wasn't much at all. And I think with Jensen, like, you know, a lot of the stuff you just went through, Mike, it's, it's pretty technical, geeky stuff. Like the average business person's kind of like what, like what did any of that mean? Jensen talks to developers and he talks to people who buy billion dollar data centers. So like when Jensen's doing talks, he's, he's, his audiences are people who are going to buy the next hundred thousand GPUs, but he is the guy and his company is the company that everyone building the AI depends on. And so why it matters to, to anyone listening if you don't follow Nvidia or Jensen closely, trust me, everyone who builds AI does. And so anything he says matters to those people. Anything he introduces may seem abstract to us this moment, but two years from now you look back and say, oh, okay, so their data set that trains like physical real world models like that actually enabled this whole crazy frontier we're now entering. That's how this all goes. Jensen made a bet on GPUs powering AI two decades ago and he was right. And so anytime he's sort of sharing his visions for the future, you got to pay attention to what he has to say.
- [58:42]
Mike Kaput
Our next topic is that AI startup Anthropic is poised for a new massive funding round that would make it one of the most valuable private companies in the US So the company has reportedly in advanced talks to raise $2 billion in a deal that would value it at $60 billion. This is more than triple its previous valuation of $18 billion last year. This funding round is led by Lightspeed Veg, Lightspeed Venture Partners and would make Anthropic the fifth most valuable US startup behind only SpaceX, OpenAI Stripe and Databricks. What makes this really interesting is Anthropic's revenue figures. The company is now generating 875 million about in annualized revenue, primarily from business customers of Claude. Now that's impressive, no doubt, but it is also a lot less than OpenAI, which projected $3.7 billion in revenue for 2024. So Paul, this seems like a pretty big funding round, but can Anthropic keep up with the enormous amounts of revenue and capital that OpenAI has Google already has at their disposal?
- [59:55]
Paul Raitzel
Yeah, I feel like we've been talking about this funding realm for like three months. I don't like this hasn't happened yet.
- [59:59]
Mike Kaput
Like officially, I mean to get it across the finish Line.
- [60:02]
Paul Raitzel
Geez. Oh, man. I don't know. They've been oddly quiet lately. You know, we've heard that they had issues with their training run on Opus, their biggest model that, you know, didn't work right. And they to go back and fix some stuff that just haven't been. I mean, we've. We've been surfacing them as much on the podcast lately. I would expect that will change. You know, I think that they're probably going to have some big stuff coming out here soon. I don't know. I feel like at some point they're going to have to pick. I say I use the word niche lightly here. Niche might be something very large. I don't know that they're gonna be able to like truly compete with OpenAI, but I don't know. I'm not. I will admit I am not the biggest Anthropic user. I have an account. I'll test it from time to time, but I don't really use Claude that often. I know people love it, I just don't know how it's differentiated. Honestly, I think it's more the personality behind it. And supposedly it was their AI safety alignment, but as we've talked about the podcast before for that was not actually why they created Anthropic. Like, they, they started messaging around being more like AI safety centric. And I think they maybe do some more research than most there, but they created it for capitalistic purposes. Like they saw an opportunity to leave OpenAI their founders and build something at the stage we were at in frontier model development. And that's what they did. So I don't know. It'd be interesting. I still don't. It would not shock me at all if Anthropic got acquired. I think that they're. They're ripe for acquisition and not just an acqui hire. Like some of these AI companies that have basically just ran out of money and didn't have revenue model Anthropic is a real company with real revenue and a ton of talent. And I think they could get acquired for a massive valuation on like, like an inflection AI where they just basically had to like, you know, hand over the company and go over to Microsoft. Yeah.
- [61:58]
Mike Kaput
And a note there. Obviously they would be an interesting acquisition target for any of the major players, but they have a very close, very richly funded relationship with Amazon.
- [62:08]
Paul Raitzel
Yeah, Google's got some money in there. Everybody's got money in there. But yes, Amazon is their, their big guy, the big supporter.
- [62:16]
Mike Kaput
All right, our next topic is that A major foreign investment in US Data infrastructure infrastructure has just been announced so President elect Donald Trump revealed a $20 billion commitment from Dubai based developer Hussain Sejwani and his company Demec Properties. This investment will fund new data centers across eight states in its first phase, including Texas, Arizona, Oklahoma, Louisiana, Ohio, Illinois, Michigan and Indiana. According to Trump, this investment could potentially double beyond the initial 20 billion commitment. Sajwani is an Emirati billionaire, longtime Trump associate. He indicated that November's election results influenced this decision, saying his company had been waiting for four years to increase our investment in the US to very large amounts of money. This follows a pattern of foreign business leaders pledging major US Investments following Trump's victory. One big one was that in December, SoftBank CEO Masayoshi Son announced plans for a hundred billion dollars investment aimed at creating a hundred thousand jobs during Trump term. So Paul, this kind of ties back to one of our main topics, Sam said in his interview with Bloomberg. One of the most helpful things the Trump administration could do for AI in 2025 is US built infrastructure and lots of it. It sounds like this is along those lines, like how does this fit into what we should expect from the Trump administration On AI?
- [63:47]
Paul Raitzel
It's all about infrastructure moving forward. We've been talking about this a lot on the show. We're going to talk about it a lot more as part of our Road to AGI series that's coming up. Chips, data centers, energy, like that's and then there's this whole issue we don't really talk much about is the labor to produce all of these things. We don't have enough electricians is a major issue. We had Microsoft an episode or two ago. We talked about them committing 80 billion this year for this kind of stuff for infrastructure build out. This is it is the story of the U.S. economy over the next decade is how fast can we build out the infrastructure to, to enable the super intelligence that these AI people envision building. And like you're, if you're ordering these things now, like you're, you're looking out three to five years like you're trying to get in line to buy enough chips to buy enough data centers to build enough energy supply to do these things, like they're not planning for 20, 25 data centers. That's all probably been acquired and bought. You got the land we're now building for like 27 to 30 when they think AGI will be here and we're on the path to super intelligence. That's what we're talking about now is like how do we position ourselves to own that globally?
- [65:03]
Mike Kaput
That's a decently high confidence bet too, given how intense these projects are. It's not just like pumping a bunch of money into a stock that you, that's liquid. Like, this is so much in advance, you have to believe. Land somebody. Yeah, somebody feels like they know what they're doing.
- [65:17]
Paul Raitzel
Yep. Yeah, it's. And then you're making some bets on, you know, as Sam right. Is nuclear fusion going to be a thing? And where do you make the bets on. On fusion? And, and then all this is, you know, under the issues of security like cybersecurity and the different risks related to this. And it's just such a massive area of interest. And that's why I said, like, we're going to have a whole separate series of podcasts that really dives into these because we need to go much deeper on these topics and bring in outside experts to talk about them. And that's the vision for the Road to AGI podcast series that will, you know, be launching soon.
- [65:53]
Mike Kaput
Next up, a few quick announcements from Grok, Slash X XAI this week. So first, X announced a major overhaul of its Premium subscription Service. There are three tiers now. Basic, Premium and Premium plus. Basic starts at 3 bucks a month. Monthly Premium is $8. Monthly Premium plus starts at 22 bucks a month. Each tier comes with a bunch more features. Premium plus offers almost entirely no ads in its experience. And importantly, Grok, which is X's AI chatbot, is available to Premium and Premium plus subscribers. All these changes took effect for new people joining the platform in December 2024. Existing subscribers will transition to the new pricing structure in their first billing cycle after January 20, 2025. Second, Xai just launched a Grok iPhone app in the US Grok as a tool is now free for all users, not just X Premium subscribers. And this app can now pull real time information, answer questions, and generate images. While Xai is working on a dedicated Grok.com website, there's no word yet on an Android app. Last but not least, a notable Tesla watcher online, his name is Sawyer Merritt. He routinely breaks Tesla news on X. He posted this past week that Grok is launching soon in Tesla vehicles. He posted a link to a live stream where Musk mentioned quote, grok in Teslas is coming soon. So you'll be able to talk to your Tesla and ask for anything. So Paul, I'm especially interested in your thoughts on Grok inside Tesla. Like that seems to kind of validate what you've talked about. A bunch of times on past episodes. That AI is really what links all these companies together.
- [67:42]
Paul Raitzel
Yeah. So the Gro team's working really hard to get people to switch from Chat GPT to Grok. Like, if you follow any other team on XAI team on Twitter, that's all they're tweeting is like, what do we need to do to get you to switch from chat GPT? So, like, they're. Elon is now not only suing OpenAI, he's like, going full force right at their user base or trying to. Yeah. So the Grok and Tesla things. I did talk about that in the fall. I was saying, like, I realized that that was like the distribution channel that and the Optimus robots is kind of where they're going with this. So for anybody who hasn't been in a Tesla or used their current voice system, it sucks. It's like the surrey of cars, basically. Like, you can ask it to open the glove box. You can ask it, apparently, to change the heat. I didn't even know you could do that. You can talk to it and ask it to, like, you know, navigate to X kind of thing, no pun intended, navigate to this location. And then I saw somebody tweet like two weeks ago that there's like 40 other prompts you can apparently use. I had no idea. So I've only ever asked it for, like, three things. You can't close the glove box, though, which drives me crazy. I can do a light show with my car, but I can't close my glove box anyway, so the thing I'm anxious to see, so I can envision very easily, just like I talked to my advanced voice mode in chatgpt through my car. So I'm driving. I just open that up and I'm talking through Bluetooth, basically, to. To Chat GPT. I. I could see very simply how they're just going to make it. So, like, you just hit the Grok button and now you can talk to Grok the same way I would talk to Chat GPT. The interesting thing becomes if Grok is integrated into the car systems. So if I can actually, when I'm going to point a, say, hey, Grok, like, actually turn here, because this is. I want to go through the valley and it'll turn and, like, take me to that. You. You cannot do that right now. You cannot interact with it or say, hey, Grok, I see a cop sitting up ahead. You're speeding. Slow down. Like, that's what I'm trying to. To understand. Like, are they going to actually allow you to interact with the car and have the car do things. Because if you envision this robo taxi world, they're seeing where there's no steering wheel and no pedals. How am I going to tell the car what to do if it's doing something wrong? This is almost essential that Grok becomes a key component of, of the full self driving, that I can actually talk to Grok and tell it what's happening or tell it what to do if I don't have a steering wheel. So I'll be fascinated to see how quickly they push it in. But I would imagine phase one is going to be I just have a chatbot in the car that I can talk to, not a system integration.
- [70:03]
Mike Kaput
Couple final topics here as we wrap up. So one topic that we've been tracking is that Francois Chollet, who's a prominent AI researcher, former former Google engineer, is taking his work on AI benchmarking to the next level. He is co founding a new nonprofit called the ArcPrize Foundation. This organization aims to develop better ways to test whether AI systems have human level intelligence. This foundation will be led by former Salesforce Engineering director Greg Komrot, and it builds on Chollei's previous work, which we talked about a lot last week and a little this week on the ARC AGI test. This is a test he created in 2019 has a bunch of puzzle like problems that require AI to adapt to new situations they haven't encountered in their training. This may sound familiar because we've talked a couple times about OpenAI's O3 model performing better on this test than humans. So it effectively like passed the test. It's the first time an AI system has done that. In other words, better tests are now needed since AI is advancing so rapidly. Paul, if you went back in time and told us like midway through 2022 that we would need to invent new AGI tests because a model already outperformed one of the top ones. Honestly, like I would have expected it at some point, but I would have thought you were crazy with this timeline.
- [71:28]
Paul Raitzel
Yeah, I mean it was inevitable. Like it's just the whole thing. And again, I learned this a long time ago. Like when you, when Google DeepMind won at AlphaGo, like when they created AlphaGo and won at the game of Go, that was considered impossible in, you know, what was it? 2016, I think that happened. It was, it was considered impossible. And Yann Lecun was on the record like a week before it happened, saying it was going to be at Least a decade. Yann Lecun Meta, who currently is saying, we're not going to get to AGI anytime soon. So Jan's been wrong before. And so the DeepMind team did the thing that they didn't think was going to be possible. So this is what always happens. Like, benchmarks are established AI research labs build teams to, to achieve that benchmark and then we do it and then like, oh, that wasn't AGI, like onto the next one. And so it's this constant thing. And so what I would continually reinforce for people is do not get caught up in definitions of AGI, whether or not it's been achieved or not, what the next benchmark is like. We pay attention to stuff because it matters and the labs are following it. But at the end of the day, the only thing that matters to you is how much of your job can AI do right now and how much is it going to be able to do 12 months from now? If you're a leader in a company, you have to ask those same questions of your team. That is all that matters. Because whether it's AGI or not, if you take a writer or an editor or an accountant or a lawyer or whatever it is, a consultant, and you look at their job and AI can assist like 50% of the time in 50% of the tasks that matters. Right now, that is a significant thing. Go back to what I started at the beginning. Like, even if you just achieve 10% efficiency gains, so the only evaluation that matters to you is how much of your job, the tasks that make up your job can AI assist with. And then we have like the human to machine scale I created years ago looks at like, how automated is it? Is it level one, where it's still mostly the human, or is it level three where it's mostly the machine? Like this is a task where the machine's actually doing like 80% of what I used to do for this task. That is all that matters. Go to go to the Jobs GPT. Go to SmartRx AI. Go to Jobs GPT. Put your title in. It will do this for you. Like, you don't even have to spend a bunch of time thinking about this. Just go in and assess it. That's the only evaluation that matters to you.
- [73:47]
Mike Kaput
Our last topic this week. Despite the availability of free open source AI models, businesses apparently are increasingly choosing to pay premium prices for proprietary solutions from companies like OpenAI and Anthropic. This comes from some new reporting from the information. So both OpenAI and Anthropic saw their revenue grow by more than 500% last year even as free alternatives from Meta, Mistral AI and others became widely available. Now this was not what many industry experts actually expected. Companies like Databricks and Snowflake had bet heavily that their enterprise customers would prefer open source models which theoretically offer more flexibility for customization and better security controls. So what is driving this unexpected shift? I mean, in a word, it's simplicity. Businesses have made it clear that they want AI solutions that work right out of the box, even if it means paying more or accepting some limitations in security features. Naveen Rao, who leads Databricks generative AI business put it this way, saying customizing these models is quote, a little bit too hard for most companies. It requires a lot of data engineering and cleaning to achieve good results. So both Databricks and Snowflakes actually have largely abandoned their efforts to develop open source generative AI models for customers. They're instead turning to techniques like retrieval augmented generation rag to improve model performance without having to modify the underlying AI models themselves. So Paul, like based on what we were talking about, honestly, even like, I don't know, like six months ago, this feels a bit of a 180 when we were seeing companies, you know, touting the benefits of open source for the enterprise. Like how does this compare to your experience speaking to working with, consulting with enterprises?
- [75:42]
Paul Raitzel
Yeah, I was fascinated by this honestly. Like I'm, I'm sure Zuckerberg and Meta would have a counterpoint to this and I haven't gone and like read up and seen if anybody has, you know, staked their counterpoint to it. But I mean Meta's betting the future on being like the Linux of AI models. They think everybody's going to build on top of these things. Honestly like my feeling was I was uneducated in this because I didn't see it. Like if you go back to last summer into the fall, I would have, I would have bet everything on this being true, that people are not going to actually build on top of all the open source models. And I have friends who like live in this world of data and technology who would fight me on that issue. And so I was like, maybe I'll just, maybe it's not my area of expertise. Like I don't know. But I mean having met with a lot of these enterprises and knowing it's not going to be the IT department necessarily driving adoption, it's going to be marketing or sales or service. Like no, we're not waiting six months for you to spend $5 million with some consulting firm to build a model. We're going to go get chat GPT right now and we're going to start doing this and then like six months from now you're going to take this away from us. Like we're realizing massive gains. Like no, we're not using the model you just wasted all that money on. So I'm seeing the reality of this isn't necessarily going to be an IT driven decision all the time time. It's going to be departments taking the initiative themselves to go get these models that don't require it to be involved and spin them up and start getting value in three days versus three months or three years. So it actually seemed kind of obvious to me that people were going to lean on the proprietary models and not like go into the open source world. But again I, I just assumed I was wrong. Maybe not, I don't know. So I will be really intrigued to watch this play out. I do not pretend like I have some, you know, unique insight here. I think it's just, I assume more business leaders are going to make these decisions, not I T leaders. And I've talked to enough business leaders that they want that simplicity thing. They just want to go and they want to get value. And by the time you start proving value of one of these models, like why would you switch? I don't know.
- [77:42]
Mike Kaput
Yeah, same as you. I can't claim to be an open source expert by any means. It's not my area of competency. But just based on seeing how much your average enterprise employee struggles to use out of the box tools, even just things like prompting or like wait, AI is different than traditional software. Like wait, how do I apply it and work with it to achieve my goals? I cannot imagine a custom built thing on top of an open source model is going to be somehow easier or better to use in a lot of those cases.
- [78:12]
Paul Raitzel
Yeah, I don't know. This is definitely one of those ones. Like when we're out doing our talks and running workshops with enterprise stuff out, I'm definitely going to have this conversation with people. Just try and get a qualitatively like where are we at? What are people thinking and saying? Because I'm very intrigued to see how this plays out.
- [78:31]
Mike Kaput
All right Paul, that is it for this week. Just a couple quick reminders. If you have not checked out our newsletter, go to marketingaistitute.com forward/newsletter. We round up all the news this week, including topics we did not get to in this episode. And if you have not left us a review, we'd really appreciate it. We use the reviews to get much better and improve the podcast. So if your podcast platform of choice allows you to leave a review, we would very, very much like to see you do that if you have not already. Paul, thanks so much for breaking everything down this week.
- [79:03]
Paul Raitzel
I feel like we've been on this for an hour and 20 minutes. Like I'm, I'm like anxious to check Twitter and see like, oh gosh, what has already happened and how much Nvidia stock is fallen in the last hour and a half. All right, thanks Mike. Thanks everyone for being back with us. We will be back with our regularly scheduled episode next week. Thanks for listening to the AI show. Visit MarketingAI institute.com to continue your AI learning journey and join more than 60,000 professionals and business leaders who have subscribed to the weekly newsletter, downloaded the AI blueprints, attended virtual, virtual and in person events, taken our online AI courses, and engaged in the Slack community. Until next time, stay curious and explore AI.