← The Artificial Intelligence Show
The Artificial Intelligence Show
#151: Anthropic CEO: AI Will Destroy 50% of Entry-Level Jobs, Veo 3’s Scary Lifelike Videos, Meta Aims to Fully Automate Ads & Perplexity’s Burning Cash
01:13:06
Loading summary
Transcript69 lines
- [00:00]
Paul Raitzer
I think the class of 2025 is going to face some different challenges in terms of these entry level jobs. I think by the time the class of 2026 emerges going to be a full blown, I don't want to say crisis but like is going to be top of mind and cord economic discussions. Welcome to the Artificial Intelligence show, the podcast that helps your business grow smarter by making AI approachable and actionable. My name is Paul Raitzer. I'm the founder and CEO of SmartRx and marketing a 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 151 of the Artificial Intelligence Show. I'm your host Paul Raetzer along with my co host Mike Kaput, who is back with us after getting, getting an episode off.
- [01:06]
Mike Kaput
You're right.
- [01:07]
Paul Raitzer
So just a reminder, we do the weekly drops every Tuesday, but last week we introduced the AI answers series, which is Kathy McPhillips and I, where we host and answer questions from our virtual events and summits. And so if you missed episode 151, we did have a Thursday episode last week. So you can go back and check that out. We went through, I think it was 19 or 20 questions from our audience that you can go listen to. And now Mike and I are back with our weekly edition. So this week, not, I don't, it wasn't a crazy week of like model announcements and things like that. But man, the AI job stuff just sort of took on a life of its own last week, I feel like. So we're going to, we're going to dig more into the impact of AI on jobs because it seems like it's really crossing over now into mainstream conversation. And so we're going to, you know, get into a lot of that and some really good recent articles that came out. So this episode is brought to us by a new webinar that we talked about actually, I don't know, a week or two ago, shared that we were going to do an AI deep dive where I'm going to go through Google Gemini deep research for beginners. So this webinar is now live to register for it's coming up June 25th. That is going to be a free session through SmartRx. You'll see a live demo of Google's deep research capabilities in Action show you how you can supercharge your work and walk away with real insights. This is ideal for beginners. So if you're, if you're really familiar with Deep research from Google and or OpenAI, you know, you might learn a few things, might just be interested in the topic that I'm going to kind of walk through the deep research project that I ran. But if you haven't used deep research yet, which, you know, having asked this question many times at conferences, there are very few executives and professionals who I've seen raise their hand, say they're doing this. So that was the whole idea here is let's show how to do this. Because Mike and I are both convinced this is an incredible value multiplier for professionals who really know how to use these tools. So we're going to go through that again. It's AI Deep Dive, Google Gemini Deep Research for beginners. You can go to SmarterX AI and click on Education and then just click on the Google Deep Research webinar. And we will also add the link to that in show notes. So again, that is coming up June 25, I think at noon is when it would be live. That's usually when we do our webinars.
- [03:32]
Mike Kaput
Yeah.
- [03:32]
Paul Raitzer
All right. And then the second big thing happening this week is the AI for B2B Marketer Summit presented by Intercept. So this is one of Marketing AI Institute. It's virtual events. We do three virtual summits as of right now. Portfolio kind of keeps expanding. But this is coming up Thursday, June 5th from 12 to 5pm Eastern Time. It is packed with incredible sessions from top B2B marketing experts. You'll learn real world strategies to use AI to grow better, create smarter content, build stronger customer relationships and much more. And thanks to Intercept, there is a free ticket option. You can see the full lineup and Register now at B2B Summit AI. Again, that is B, the number 2B Summit AI. You can also under Marketing Institute, if you're on that site, just go under events and it's listed there as well. Okay, so let's dive into the topic that everyone seemed to be talking about, including when I was at like parties with friends. It was crazy. Like this is. This actually seemed to just hit the mainstream. I don't know. So yeah, let's dig into Dario Amade.
- [04:42]
Mike Kaput
All right, Paul. So yeah, the talk of the town is AI's impact on jobs. This is really kind of breaking containment here and a big reason for that is that we just had a report this week that anthropic CEO Dario Amade went and told Axios that half of all entry level white collar roles could disappear within five years thanks to AI. And not just disappear, but basically, he thinks vanish at some point overnight. He says CEOs will quietly stop hiring and start replacing humans with AI agents the moment it makes business sense. And in his estimate, unemployment could spike to 10 to 20% as a result. And yet, he says few leaders are warning the public. Congress is largely uninformed or unwilling to talk about this. And the White House is very quiet, not about AI, but about the possibility of job disruption. Not to mention many, many workers who are kind of outside the AI bubble seem to be pretty unaware that this can be happening. So Amade, who is of course helped building this technology, says companies have a duty to be honest about what's coming, which is I guess why he decided to talk to Axios about this in a very frank way. However, even so, that contradiction is a bit jarring. In just a couple weeks ago we got Claude 4, which exhibits very, very powerful capabilities in white collar work like writing code or doing summarization of legal documents. So Paul, maybe walk us through kind of what's going on here. Amade is building this stuff, but also saying that it could have this monumental negative impact on entry level work. It's not the only warning we're getting right now. There's two leaders at LinkedIn, their head of Economics in the Americas and their Chief Economic Opportunity Officer. Both of them published Recent articles about AI's threat to entry level jobs specifically, what do we need to be paying attention to here?
- [06:49]
Paul Raitzer
So as I was saying kind of at the open, I do feel like this, for whatever reason, him saying this just sort of caught the attention of everyone in, mainly in the mainstream media. I put this on LinkedIn. So we're recording this on Monday, June 2nd. This will drop on Tuesday, June 3rd. So I put this on LinkedIn five days ago, so it'd have been like Wednesday or something. I think when this came out, a normal LinkedIn post for me will get 5 to 10,000 impressions. As of this morning this post is at 151,000 questions. So and like the comments within the thread are, are super productive. Like I was actually super, like really impressed that people were being very level headed and having some disagreements, but really good dialogue. So there's I don't know how many comments there were. 2. 276 comments and 71 reposts. And so this is far beyond the usual like AI bubble of like my close network of people that always talk about AI. This superseded that. This was a lot of people that I don't usually see in the comment threads posting thoughts and questions and concerns. So it, it just feels that for some reason him saying this just sort of move the dialogue forward, which I see as a very positive thing. Obviously on this show. For a long time now we've been kind of pushing this as something that needed to be talked about way more. When I, when I saw it, I'll come back to this research. But I like immediately went and like ran a deep research project to try and see what was going on. You know, if, if these things he was saying were actually starting to happen. Because my personal theory, my hypothesis is that we will start to see data emerge through the summer and into the fall that shows AI is having a clear impact on jobs, that the quiet AI layoffs we've been talking about are going to start to compound and become more obvious what's actually happening. And I think that this will get accelerated because next generation models are coming. We will get GPT5 at some point, likely this summer, I'm assuming, or early into the fall. We'll get Gemini 3 at some point this year. Elon Musk is hyping up the next version of Grok, which is, I assume is going to be Grok 4. So these new models are going to emerge. They're going to start to accelerate the impact across different industries. As we discussed in episode 149, you can look at the total addressable market by salary of professions and you can actually start to get a gauge of where the investments are going to go. And then my hypothesis continues into next year. And I think that that's when we start to really see disruption. So I think the class of 2025 is going to face some different challenges in terms of these entry level jobs. I think by the time the class of 2026 emerges in May of 2026, it's going to be a full blown, I don't want to say crisis, but like is going to be top of mind and cord economic discussions. What we have to keep in mind is May of 2026 is right in the midst of the start of midterm elections in the United States. So for any international listeners we may have who aren't familiar with how the US Government works, US Congress is the legislative branch of the government composed of the two chambers, the House of Representatives and the Senate. Congressional elections determine represents from representatives from the states in the federal government and then which political party will hold majority in each chamber. For the next two years, Congressional elections happen every two years. At that time, one third of the Senate and every seat in the House is up for reelection. And so the midterm elections will take place November of 2026. So I think that this becomes a major topic for the midterms. And as Mike, you and I talked about for this presidential election cycle in 2024 in the United States, it was not discussed at all. Like we just kept saying, where's, where's the conversation about AI? And then as soon as the new administration came in, day one, it was now a topic. So I think that it's still largely being ignored, but I don't think that that's going to sustain for long. And I can just like qualitatively again. In the last week I've been at multiple functions with family and friends where I have parents of high school and college age kids asking me unprompted, like, what should we be doing? What my, my son's majoring in computer science. Is that viable? You know, they're thinking about going to business. What should they like? You're now getting these way more educated questions about the impact this stuff is going to have. And so I think that that started to cross over. And then if the economy keeps struggling, if tariffs and inflation continue to happen, and then you layer in an unanticipated unemployment number or the bigger concern I actually honestly have right now, and I haven't had a chance to dig into this data deeply, is underemploy. So if you're not familiar with the concept of underemployment, that's when someone is working at a job that doesn't fully utilize their skills, education or experience. So maybe they're working part time and they actually want full time employment or they're taking a job at a retail store when they came out with a computer science degree. So it doesn't show up in the unemployment numbers because they are employed, but they can't get the job that they want or they can't get the full employment that they want. And so my guess is that's going to start to be a much bigger issue going into next year as well. And so I don't know, I just feel like we may have hit that tipping point where this really starts to become a concern for people, they start taking a far greater interest in it. And if constituents across, you know, the United States start taking greater interest than their Congress and Senate have to start taking a greater interest. And if, if jobs truly start being impacted, unemployment and underemployment going into the 2026 election cycle. It, it has to then become a major political issue. And I don't know what that means yet. Like, I, you know, I think, and I, I do think I mentioned this a couple episodes ago. I think the church is going to get involved. I, I fully anticipate so. I actually saw something this morning about a sect of the church that sent a letter. I think it might have been to the Pope. No, actually it was to the Trump administration, basically vocalizing that they want to slow AI down. So it's just, it's gonna, it's gonna cross over into a true societal issue very soon. And I, and that could go a couple different ways.
- [13:44]
Mike Kaput
So we've talked about before that we need to have the conversation around AI's impact on jobs. Part of Amade's thing, whether you agree with it or not, was him kind of saying, we need to be a lot more frank about this. Axios published a great follow up piece about leadership in the age of AI and what they're doing as a company and how they are having more frank conversations with their own employees about what AI could do and how they're going to move forward. But a lot of this is just people still saying, like, let's talk more about it, which is great. But I guess the question that follows is what should we actually do about it?
- [14:24]
Paul Raitzer
Yeah, I, you know, I think individually we have to start looking at our own industries, looking at our own companies, being more proactive, thinking about, you know, think about our kids, what their schools are teaching, and if you need to take on more responsibility, prepare your kids. Which again is like, I'm lucky. My wife and I are lucky. Our kids go to an incredible school, but even they're not teaching AI, they're not not allowing it, but they're not teaching it. And so I'm very much looking at this like, how do I be proactive and actually preparing, you know, our kids? I was at a friend's house last night having a conversation. You know, his nephew is in, I think he's in his freshman year, maybe sophomore year. And they're just like straight up having this conversation. Okay, like, what is, what's the career path look like? What are the majors you're interested in? How would you layer AI over those things? So I think it's a mix of being proactive individually and then like as a community starting to, to focus on like more of what we can do.
- [15:34]
Mike Kaput
Okay, so we're kind of talking here about, you know, Amade's comments and some of the kind of things, you know, people at LinkedIn are saying about all this stuff. But our second big topic this week is kind of closely related to this first one, because while we're seeing all these warnings come from AI leaders and others who are kind of employment experts, there's plenty of weight to those arguments. We're seeing it in our own industry. But I guess the bigger question here is how seriously should we be taking the argument that all this is happening right now, like really urgently? Like, I think on a long enough timeline. We agree there's going to be major changes to how jobs work in the age of AI, but is this really worth worrying about as urgently as some of these leaders would have us believe? That's kind of the next question that follows from this. And we wanted to look at a few different proof points that we're seeing at the moment in order to maybe stress test or kick the tires on the assumptions behind some of these job loss warnings. Because, you know, it gets headlines when Amade says half the entry level jobs are going away. But why do we believe that? So, Paul, as this we get into the second topic, I wanted to ask you to give us maybe a sense of what we're seeing here that's supporting this idea.
- [16:53]
Paul Raitzer
Yeah. So there was a couple of things that articles that sort of surfaced right around this time that caught our attention. So one was from the chief economic opportunity officer at LinkedIn, Anish Raman. And I'm just, I'll call it a couple excerpts from this and we'll put the link into the article. But it was a New York Times op ed piece on says I'm a LinkedIn executive. I see the bottom rung of the career ladder breaking. So in this one it says breaking first is the bottom rung of the clear, clear career ladder. In tech, advanced coding tools are creeping into the tasks of writing simple code and debugging, the way junior developers gain experience in law firms. Junior paralegals and first year associates who once cut their teeth on document review are hand work overt AI tools to complete in a matter of hours. And across retailers, AI chatbots and automated customer service tools are taking on duties once assigned to young associates. Continued, these changes coincide with a shift appearing in the latest employment numbers. The unemployment rate for college grads has risen 30% since September 2022, compared with an 18% for all workers. And while LinkedIn's workforce confidence index, a measure of job and career confidence across nearly 500,000 professionals, is hitting new lows and general uncertainty. Members of Generation Z are more pessimistic about their futures than any other age group. Meanwhile, in the recent survey, over 3,000 executives on LinkedIn at the VP level or higher, 63% agreed that AI will eventually take on some of the mundane tasks currently allocated to entry level employees. Continued their research suggests that professionals with more advanced degrees are more likely to see their jobs disrupted and those without. While the technology sector is feeling the first wave of change reflecting AI's mass adoption in this field, the erosion of traditional entry level tasks is expected to play out in fields like finance, travel, food and professional services. So that was the first one, Mike, because, like, who has better data than LinkedIn when it comes to this stuff? And then Kevin Roos had a great article and I'll go through a couple excerpts of this one. So he wrote this month, millions of young people will graduate from college and look for work in industries that have little use for their skills, view them as expensive and expendable, and are rapidly phasing out their jobs in favor of artificial intelligence. That is the troubling conclusion of my conversations over the past several months with economists, corporate executives and young job seekers, many of whom pointed to an emerging crisis for entry level workers that appears to be fueled at least in part by rapid advances in AI capabilities. I'll go through a couple more excerpts here, but I would highly recommend reading this whole article. It's very good, he says. You can see hints of this in the economic data. Unemployment for recent college graduates has jumped to an unusually high 5.8% in recent months, and the Federal Reserve bank of New York recently warned that employment situation for these workers had deteriorated noticeably. He quotes Oxford Economics report that says there are signs that positions are being displaced by AI at higher rates. He goes on to say, but I'm convinced that what's showing up in the economic data is only the tip of the iceberg. In an interview after interview, I'm hearing that firms are making rapid progress toward automating entry level work and that AI companies are racing to build virtual workers that replace junior employees. One tech executive told him his company had stopped hiring anything below an L5 software engineer. So just a grading level. Mid level goes on to say, this is something. This is a quote from Molly kinder at Brookings Institution. This is something I'm hearing about left and right. Molly studies the impact of AI on workers. Employers are saying these tools are so good that I no longer need marketing analysts, financial analysts, and research assistance. Now Kevin did say, if there's A silver lining for recent grads, it's that it's, at least for some of them, the threat of AI replacement seems to be lighting a useful kind of fire. Some young workers that he spoke to are using their experience with AI to vault themselves ahead of their senior colleagues. And others that are steering more are clear of traditional ladder climbing. So that actually goes back to Mike. On a recent podcast we talked about how these like young workers are going to show up and be like, why are you, why are you doing it this way? Like why, why are like spending two weeks on that thing? And then I'll just share a quick anecdote. Like I put this on LinkedIn on Sunday morning, I just like woke up and like we had, we had meetings on Friday at the company and we were talking about like our ideal customer profiles and customer journeys and you know, developing customer journey maps and Personas and all these things. And it was like overwhelming amount of work that probably needed to happen. It's like, ah, who's got time to like do this? And so I literally just like pulled out my, I think I was using co CEO and my, my custom GPT I built and I'm just like, cup of coffee and I'm like, ah, can you help me do this? And it just because it already knows in its system instructions our revenue model, it knows our audience is like, it knows all those things. It's trained on it. It just started spitting out these ICPs and I was like, oh cop, these are really good. And so I just kept like. So I gave it the list of the eight that I thought of and then it would just build these things out and then it build out customer journeys. And again, it was just this representation of once you understand what these things are capable of, you just look at work differently. And I think I said in the LinkedIn post, you know, I owned a marketing agency for 16 years. People paid us do these exact things. And if a company had come to me and I was still running an agency six months ago, a year ago, that would have been $25,000 minimum. I probably would have priced it on a per ICP basis. I would have probably said 2500 to 3000 per ICP times 8. Like that's how we would do pricing. And so it was easily that and it was probably easily 50 plus hours of work. And if we were quoting it, it probably would have been like, you know, two, three, four weeks of work because some senior strategist would have to do the work. Some other person above them is going to probably have to review and approve it. So the whole thing just changed in under an hour. I built eight ICPs that are now going to be like the foundation of our team's conversation to take them and edit them and build on them. So I don't know, Mike, I think this goes like, yes, there's this disruption that's going to happen, I think, at the entry level for sure. I think, you know, this goes to how we talk about building companies. So anybody listening to the show have heard me say, like, AI Forward is sort of like how we define these companies. But Jeremiah Oyang was actually shared a post on LinkedIn last week where he was sort of. He had heard the AI Forward thing at an event in Silicon Valley. And so Jeremiah had tagged, I think you and I may be in that, and tagged the podcast. And so I was like, adding some context for him to where, like, the origin of that came from. And so I thought I would add this to this because I think this is an interesting part of the conversation. If you have an AI native company, which is what I consider like a startup, like, we're just going to build a smarter version of a company from the ground up, it's going to require more people. There's never been a better time to do that. It's like what we're doing at SmartRx. It's, you know, we can just be really smart with, like, the people we bring in, infusion of AI. We don't have to lay anybody off. We actually, like, can bring them in, we can pay them more than normal. We can do all these things, give them more vacation days, because we're able to do that from the ground up. We're able to just build the company. And my hope is we enter this sort of golden age of entrepreneurship and that offsets some of the job losses. So, you know, we've double our team, the size of our team in the last, like 30, 45 days, and we're still adding more people. So we're going to be a growth engine for the economy in a small way. Like, we're going to do our part to like, build a company and make a difference, but we're going to do it with a fraction of the total employees we would have needed, you know, years ago. But then what we're seeing happening, where these jobs start getting impacted is the AI emergent organization. So these will have you as, like, the traditional organizations that are trying to infuse AI now, they're going to require fewer people to grow. So if you've Got an existing company, say it's 100 people, a thousand people, 10,000 people, whatever. And you start infusing AI to build a smarter company. If the demand for their products and services remains flat or modest, they're going to cut the workforce. Like, if you're not growing even with AI, you don't need as many people. Like, we know that equation, right? If you grow, then you can grow without having to hire as many people as you would have. So we had a quote from Financial Times, Janet Trunk Hill, who's the global chief executive of ey, she recently said at a Milken Institute annual conference that her firm would not cut jobs in response to AI, but could do more with less. She said, quote, I like to think we can double in size with the workforce we have today. So there you go. I mean, they're just telling you, like, we're not going to hire as many people.
- [25:50]
Mike Kaput
Yeah.
- [25:50]
Paul Raitzer
So. And then I'll, you know, my, my other thought here was I did before I read the Kevin Roos article, I actually wanted to see, like, is this happening? Like, can we identify some indicators that we are actually starting to already enter this job loss period? And so I gave a prompt and we'll drop this in the show notes, but the prompt was. Dari Amaday, CEO of Anthropic, recently stated that I could wipe out half of all entry level white collar jobs and spike unemployment to 10 to 20% the next one to five years. Can you create a research brief based on any 2025 data points that could be leading indicators? We are already entering the phase of job displacement and disruption, for example, difficulty of recent graduates struggling to land jobs. Only use 2025 data. So that was the prompt I gave. I think I used OpenAI's deep research for that one. I'm not going to go into all the details, but here's what the conclusion said. While definitive pronouncements about the full realization of Amade's predictions are premature, the 2025 evidence strongly suggests the initial phase of this predicted disruption is underway. This underscores the urgency for all stakeholders, including educational institutions, policymakers, businesses and individuals preparing the intent of the workforce to acknowledge these early signals, which Mike goes back to your question about, like, what do we do? Yep. Proactive strategies for workforce adaptation, comprehensive reskilling and upskilling initiatives focused on AI literacy and human AI collaboration, and a fundamental rethinking of career paths for new entrants into an AI powered economy are becoming increasingly critical. The 2025 data does not offer a complete map of the future, but it does provide clear evidence based call for proactive engagement with the transformative potential of AI on the world of work. So now Mike, this is like literally game time decision by Mike and I. So I've been working on this thing. I'm really frustrated, honestly with the lack of advancement being made to prepare for this. If we prepare for the future of work, the future of education, I am grateful for all the people specifically in higher education who reach out to me every week asking what else can be done. There are a lot of people who want to make a difference and want to think this thing through. So like about a year ago I started working on this idea to build these like AI Impact Summits where we would try and actually project out the impact. We would bring in economists and philosophers and business leaders and educational leaders and government leaders and kind of bring people together to start having like think tanks around this stuff. And so I've been sitting on this idea for a while and then I was kind of ready to just sort of punt it into 2026 because there's a lot going on. And then honestly, like the last two weeks of this podcast has like kind of pushed me to it to a point where it's like, okay, we have to do something. We can't just keep talking about this stuff on this podcast with no, no hope of like, well, okay, we want to help. And I feel that from our audience, like people like, okay, we get it, like job going to get messed up, like what do we do? So what we're going to do is this idea I have for these sort of the Smarter X Impact Summit on the future of work and education. We're going to put up a, an interest list. So I'm not going to make any promises that this event is coming, you know, this summer, this fall. I've got a million things we're working on, but I think this is incredibly important as part of our AI literacy project, as part of like our bigger mission to try and have AI have a positive impact on the economy and society. So we're going to put up an interest list and what I would ask here is it just provides a little bit of information about the event series. Like what we're thinking about is doing this series that would be geographically diverse. It would be an ongoing series of one day events. We bring people together to sort of explore, discuss and debate these pressing AI topics with a big focus on jobs and a big focus on education. And so there's going to be a simple form you can Just express interest of, you know, kind of organization, job title, geographically where you are, and then just your interest in attending it, sponsoring it, hosting, co hosting, speaking, receiving updates. It's a super simple form that I literally threw together on Google this morning. And this information will help me kind of figure out how to. To prioritize this. And if the interest is high, we will, we will accelerate that, that planning. I want to do this. I've wanted to do this for a while and I think it's really important. I haven't honestly, like formalized the model exactly in my head. I have a vision of what this needs to be, but we're just at the formative stage right now. So if it's something that's of interest to you, check out the show notes. We will drop a link to that form. And then, like I said, just simple. We're not going to market you. This is not like a marketing database thing. This is. If you're interested in this, that's all we're going to use this list for, is to communicate with you specifically about this. If you choose to opt into something else, that's on you. But this is purely just to gather some interest and gauge, you know, if, when, where. Those are all the things that I'm kind of thinking about. But I just feel like we have to do more. Mike, honestly, that's.
- [31:07]
Mike Kaput
That's awesome. I couldn't agree more. Especially, you know, I'm glad we're talking about it more. But yeah, it couldn't be more important to accelerate actually finding solutions because you and I have talked about. I just don't. Anytime someone's like, well, okay, think about UBI or this other thing or, you know, it's possible. Interesting stuff, but none. It just seems like way, way, way far in the future to me or illogical or implausible maybe. So real solutions are needed.
- [31:32]
Paul Raitzer
Yeah. And like the qualitative stuff I was referring to earlier, like, that's, you know, some. I'm with my high school buddies last night and we're just kind of catching up and explaining what's going on. They're just like a couple were like, I had no idea. Like, I didn't know it was this far along. And what's happening? Like, what are people going to do? How does, how does it get solved if people don't work anymore? I was like, ubi, like that. That's the lab's answer to everything. They don't tell you what that is or how it works. And yeah, it's just. I don't feel like anybody is presenting trideas, and I'm not saying we're going to be the one to figure it out, but if we have to be part of accelerating the conversation around it, then right then that's what we need to do. And I feel like right now that's the best role we can play in this, is just try and get people together and drive this conversation forward and not like behind these closed doors, like maybe these conversations are happening in government, we're just not being told about it. But yeah, we need to get this, like kind of shine a light on this.
- [32:26]
Mike Kaput
All right, for our third big topic this week, we are talking about synthetic content because we are almost certainly not adequately prepared for the flood of AI generated content that we're now seeing, thanks to things like Google's release of its incredible VO3 video generation model, which honestly shows us just how close we are to being unable to tell what's real and what's not online. So not a new problem, but is definitely been accelerated by the fact that VO3, which was released just a couple of weeks ago, already has users using the tool to flood social media channels with hyper realistic videos. They're incredible. We're going to actually show examples of them in a later segment. You should definitely check them out. But while plenty of companies, notably TikTok and YouTube, are rolling out some policies to tag content made with AI, the technology may be moving faster than efforts to flag it. Now, Paul, I know you posted the following on this topic. You said, I hope there is a plan for X, the social media platform as well as Facebook, Instagram, YouTube, et cetera, to clearly mark AI generated videos like those from Veo. I assume there is a way to know through Google DeepMind's Synth ID. It seems irresponsible at this point to not publicly tag them on social media. So Paul, how bad is this problem right now thanks to VO3?
- [33:57]
Paul Raitzer
I'm. I'm not sure how bad it is. I don't. So when I, when I started seeing the VO3 videos I was like, like there's, there's no way people are going to have any clue this is AI, right? Like, and, and so I live, I would say like the vast majority of my time on social media is spent on X, which is where I curate the vast majority of my AI news and on LinkedIn, which is where I spend most of my time, like in engaging and sharing. I do not spend much time at all on like Instagram, Facebook, TikTok, so I'm not sure what's happening there. I just know from a Facebook and Instagram perspective, my guess is the audience who is spending a lot of time specifically on Facebook is probably not the audience that, that largely comprehends what's happening in video generation right now.
- [34:52]
Mike Kaput
Right.
- [34:53]
Paul Raitzer
And that these models are capable of doing seeing with vo. So I, I, I was just curious, I just like started bouncing around, like, what are the different platforms doing? So like Tik Tok, I went to and we'll put the links and you can go research this yourself. They have auto labeling. So it says Tik Tok may automatically apply AI generated label. Because I was also seeing people saying things like, I think it was on X where it's like, yeah, I posted some on Tik Tok and they took it down and I can't even challenge it. Like, like the, if TikTok decides it's AI generated, you're just cooked. Like, you can't go back and change it. So they will automatically label something if it meets their criteria. So this may happen when a creator uses TikTok. AI effects or uploads AI generated content that has content credentials attached to it. A technology from the Coalition for Content providence and authenticity, C2PA, which I think Mike, we talked about in 2024 when that came out. So that is one thing. Except when you look into that organization which was created by Microsoft and Adobe in 2021, doesn't appear to be very many companies participating in it. None of the AI labs are listed. It's just Adobe, ARM, Intel, Microsoft and TruePik. I don't know what TruePick is. So it was supposed to be something to like detect deep fakes and alert if it's synthetic media. But best I can tell, it's not like widely adopted or it doesn't have a big impact here because again, the ones building these things are like, you know, Runway and Open AI and Google and like they're not, they're not listed here. So TikTok seems to have a, you know, you're supposed to tag it yourself, but if you don't tag it, they might, and if they're the ones that tag it, you're in trouble. YouTube, which I can't understand this one because they don't say anything about that I could find about using. So again, YouTube's owned by Google. You would think they would have the Sith ID technology from DeepMind baked right into YouTube. Maybe they do, but it's not obvious on their support pages. So again, YouTube sort of relies on creators to tag the content that goes through and explains to you how to do this and the trouble you'll get in if you don't disclose it. They. They say content of viewer could easily mistake. So the things you're supposed to tag as content of viewer could easily mistake for real person, place, scene, or event that is made with altered or synthetic media, including generative AI. We're not requiring creators to disclose content that is clearly unrealistic, animated, or has these special effects. The one that again, I'm most familiar with would be like X, because that's where all these VO2 or VO3 videos are appearing. And it all there says is like, they have an inauthentic content thing. You may not share an inauthentic content on X, which is funny because Elon Musk is probably like the biggest sharer of this stuff. So then they call like synthetic and manipulated media and it kind of goes in again, like, you got to tag this stuff and then meta. I'll put the link in there as well. They rely on people to tag their stuff. They don't tag images or require you to tag images. They basically say examples would be like, a video appears realistic of a group of people walking around an outdoor market. So, like, you're not. They're all kind of saying the same things. And then the Google DeepMind Sith ID Mike, which we talked about on a recent episode, because they released updated version of this. So Google has the ability to insert these watermarks in things created by DeepMind technology. But to have that widespread, you would need like a partnership with X to. To kind of pass that information through. Right now it looks like the solution on Syth ID is if you come across a video on X that you're not sure if it's real, you would have to like, take that video, give it to Sith id, and then it could then tell you whether it thinks it's real or not.
- [38:36]
Mike Kaput
Right.
- [38:36]
Paul Raitzer
So it just seems like we're just not there. Like, like the organizations are mentioning it in their, you know, their terms of use and, and their guidance for creators. But I, I don't know when like, I'm gonna be on X and see a video. And like, I mean, I, like, I literally double check everything out, even from like verified sources because it's like, well, they might be feeling fooled. So I just assume everything's fake now. Like, even we had the, the drone, you know, the Ukraine, Russia drone attack this weekend. First time I saw that on X, I was like, that's fake. But I just assumed it was right. And So I went and found, like, a legitimate media source. I'm like, oh, man, that I actually did that. Like, that was. So, yeah, I've kind of arrived at that point where I just doubt everything until I verify it's real.
- [39:22]
Mike Kaput
And especially too, and we've talked about this for several years now, that your average person also is very ill equipped to even understand what's possible. And that's never been more important, important or more true than today, because V3 is just jaw dropping.
- [39:40]
Paul Raitzer
It's wild.
- [39:42]
Mike Kaput
So on that point, as we dive into rapid fire, our first item is we actually wanted to show off a couple viral videos that are created using the VO3 video generation tool. And they're kind of going viral because they also have a quirky little AI focused story behind them, which is that these are very, very lifelike videos. They're being shared in the form of some clips on X, and then there's some longer videos on YouTube and they're all under this theme that they call it in the titles of the Prompt Theory. And what this is, is these videos are created through VO3 prompts, and they tell these stories of AI characters who basically are like, refusing to believe that their AI generated. So there's kind of a, kind of blow your mind aspect to this as well, going down the rabbit hole of, like, what it means to be AI generated. And it's really, really eye opening to see this put together in this way. So, Paul, before I get your thoughts on this, we're going to kind of take a real quick look at one of these clips from the Prompt Theory. A girl told me we're made of prompts. Like, seriously, dude? You're saying the only thing standing between me and a billion dollars is some random text?
- [41:00]
Paul Raitzer
Honestly, the biggest red flag is when the guy believes in the Prompt Theory. Like, really? We came from prompts? Wake up, man. You want to convince me that this perfect creation behind me is the result of ones and zeros, a binary code, and nothing more? It makes no sense. We're not prompts. We're not prompts. Yeah, the. I mean, this stuff is so crazy. When I first saw a short clip of this Twitter X, it was one of those, like, dystopian moments where you just. I watched it like three times and the clips get, like, progressively more. I don't even know the right word here. Disturbing, I guess. Like. Yeah. So if you're listening to the podcast, it's hard to, like, understand the true impact of these things if you don't, like, see the video. So either go click on the link or flip over to our YouTube channel and watch it. They are indistinguishable from reality. So it truly looks like it's a comedian or it's a husband, you know, talking to his dying wife. Like it's. All those things look completely real. And then just this sort of bizarre, them being aware they're just prompts and like asking the prompter to give, I don't know, like it was. It just affected me in a weird way when I saw it and I was trying to like, process it because it just. It's on multiple levels. You're trying to deal with this, that the video looks so real, the audio sounds so real, and then that it's creating this whole prompt theory with a prompter that's almost like a godlike figure that they're like asking for the. The prompter to prompt them something in their life. Like, it is. It's so messed up, honestly, it gets really unnerving.
- [42:50]
Mike Kaput
Yeah, that's a good way to put it. It' of praying. They're basically saying, hey, can you prompt this for me to addressing kind of off screen, you know, the viewer almost. Right.
- [43:00]
Paul Raitzer
Yeah. And. And when I mentioned earlier, like, the church is going to start to get much more involved. This is the kind of thing where if you're like the church, these. The kinds of things that are red flags, where it's like, whoa, hold on a second. Like people are going to start believing like crazy things. And I don't know, that's why I said this is like a deep one, that it goes multiple layers and you just kind of keep peeling it back and thinking about how messed up it is. But yeah, if you haven't watched them, we'll drop the nine minute YouTube or the. Yeah, the link to the nine minute video. It is. It messes with you. Yeah, yeah.
- [43:37]
Mike Kaput
All right. Our next big topic in rapid fire, Paul, is that Meta is shaking up its AI division to keep pace with rivals like OpenAI, Google and China's Deep Sea. So this past week, Meta split its generative AI efforts into two new teams. One is focused on AI products like Meta Assistant and Instagram features, the other on foundational research, including llama models and reasoning systems. So from some reporting by Axios, they said, quote, in an internal memo sent Tuesday and seen by Axios, Chief Product Officer Chris Cox laid out the new structure which will see efforts divided into two AI teams. An AI products team headed by Connor Hayes and an AI found AGI Foundations unit co led by Ahmed Abdal And Amir Frankel, now Meta's AI research unit. Fair apparently remains separate from this new structure. Axios also said no executives are leaving as part of the changes, nor are any jobs being cut at the moment. This all comes after months of internal issues at Meta related to AI. There are reports of burnout, infighting, lack of focus. They've been stumbling. Llama 4's release was delayed. There was a leaderboard controversy we talked about when they basically were trying to game the results of popular chatbot scoreboards, and they've had some AI talent also begin to leave. So, Paul, what does this actually mean for Meta? And I'm curious how, if at all, does this impact or reflect on Yann Lecun, who's one of the godfathers of AI who works at Meta? We talked about a ton on the podcast.
- [45:21]
Paul Raitzer
Yeah. If you want to know the story of the founding of Fair, the Facebook AI research group, and Yann Lecun, you know, going to work there, read Genius Makers. It's like the best inside story I've seen of how it all kind of transpired. Zuckerberg tried to buy DeepMind before Google acquired them. He tells the story of how Hasabas and Leg and Solomon, the three co founders of DeepMind, turned down more money from Zuckerberg to go to Google because they didn't understand Zuckerberg's vision for AI didn't align with the growth obsessed culture, kind of the move fast break things approach of Facebook, and were convinced that Zuckerberg did not share their ethical concerns over the rise of AI. So they just saw it as like a pure, you know, capitalistic play by Zuckerberg and where they felt like at Google they could be treated more as a research lab, which ironically, all these years later, generative AI emerges and they kind of get thrown into the commercialization side of it. But at the end of the day, I mean, I think they, it was probably a better fit for what DeepMind wanted to do, especially back at that time. So it is a fascinating story. Lacune is a, a very, very important figure in the history of AI, specifically the last, you know, 20 years. He's been a key part of major breakthroughs. He's controversial in his stance sort of against large language models, which I think sometimes he, you know, maybe just doesn't vocalize exactly like his point. But he's, he's kind of been dismissive of large language models that they're not going to get us to AGI and he doesn't even like that term, you know, new breakthroughs. Are needed. And I don't know, he's been right a bunch in his career and so we'll, we'll see how Paul plays out. But I do wonder, like, I think my understanding is he doesn't have direct reports there. Like, I feel like I saw an interview recently where they were talking about his, his work there. So it's always been an interesting relationship, an interesting working environment. It doesn't sound great. I mean, it sounds like there's quite a bit of chaos internally and, I don't know, reorgs are always interesting to see how they play out.
- [47:36]
Mike Kaput
So. In some more Meta news, the company wants to reinvent advertising by automating the entire thing with AI. Meta now plans to let brands generate complete ads from scratch using AI, ideally by the end of next year, according to a new report in the Wall Street Journal. So that means businesses could upload a product image, set a budget, and Meta's AI would handle the rest, writing the copy, producing other images or video, targeting the right users and optimizing spend. In fact, on the company's annual shareholder meeting last week, Mark Zuckerberg outlined the vision, saying, quote, in the not too distant future, we want to get to a world where any business will be able to just tell us what objective they're trying to achieve, like selling something or getting a new customer, how much they're willing to pay for each result and connect their bank account and we just do the rest for them. So, Paul, this seems like a pretty big deal if Meta can pull it off. I mean, you know, since way back to the beginning of Marketing AI Institute, we've seen vendors trying to go after this kind of golden goose. Right. Which is fully autonomous AI advertiser.
- [48:48]
Paul Raitzer
Yeah, I mean, one of the ones you remember, Mike, that we. Because, you know, back before generative AI, before ChatGPT, we spent a lot of our time trying to educate people on use cases. I actually built a tool that you could go through and rate the value of different use cases. It was largely using like machine learning and natural language processing and some other stuff, but we weren't at an age where you could just generate something with a prompt.
- [49:10]
Mike Kaput
Right.
- [49:10]
Paul Raitzer
And one of the companies that we spent some time with in those early days was trying to do something like this, but that the human would create the different headlines, images, offers, and then you would upload all these variants and then the AI would run, you know, sometimes millions of variants, mixing and matching, but it wasn't creating anything on its own. So, yes, this has been around for a long Time, it's been theorized. I mean, I would imagine Facebook just, you know, like Google and others, these massive ad networks, they're going to have the ability to do this now at a different scale. It's just, it's always funny when you're like when you're relying on the company that you're spending the money with.
- [49:54]
Mike Kaput
Right, right.
- [49:55]
Paul Raitzer
To dictate what's going to work and the efficiency of that spend and you know, they may have ulterior motives sometimes of, you know, how much your money to spend. But yeah, I mean, I, I again, I would say like if your career right now or your company makes its money creating ads on Facebook, I would maybe start thinking about what that looks like in six to 12 months because it's not going to look like it does today.
- [50:24]
Mike Kaput
It's a fair amount of agencies that I believe do things like that.
- [50:29]
Paul Raitzer
Yeah, yeah.
- [50:31]
Mike Kaput
So in other news, just two years ago, schools in Miami Dade county had actually banned AI chatbots because they were afraid of you students using them to cheat. But now they're actually doing an about face and kind of going all in because we just got news that Miami Dade county, which is the third largest school district in the US is introducing Google's Gemini chatbot to over 100,000 high schoolers. And this is the biggest rollout of classroom AI by any school district so far. So teachers are using AI to do everything from simulating historical figures, help grading essays, and even co designing lesson plans. There's a ton of anecdotes in a New York Times article reporting on this about students then kind of co working with AI to learn better and to refine their essays and assignments with guardrails in the AI. So it won't actually help you do the assignment, but will actually prompt you to work better on the assignment. And this shift kind of comes amid a national push that we've talked about a little bit. There's, you know, the Trump administration has actually called for AI literacy in K to 12. And Miami's approach is really interesting, not only for the scale but because is emphasizing both training their staff and putting guardrails in place for students to do effective work with AI. Now Paul, this kind of start ongoing AI and education conversation. There's a lot to like here. And what jumped out to me there was a line in the New York Times article that they, the, the school district actually rolled out AI tools in tandem with AI training workshops for its 17,000 teachers. They actually built something they called the AI Institute, which offers a bunch of like Courses and training for teachers, which I thought was exactly kind of what you've always been saying on the podcast is the way to success.
- [52:26]
Paul Raitzer
Yeah, teach the teachers is what we always say. Yeah, this is phenomenal. Like, I would love to talk with whoever organized this. Like, I mean, it seems extremely well thought out. I'm not sure who was actually behind the whole program and devised it. Obviously we have some connections at Google we could maybe talk with there. I'm just like scanning these notes, trying to see if we know what it is. But yeah, kudos. I mean, I think this is, you know, potentially a great blueprint for educational institutions, but this is exactly what needs to happen in enterprises. Like, it's the same, it carries over for sure. So we'll definitely follow along with this story and I could see, you know, revisiting this on future episodes. I would love to hear how this goes and especially leading into next school year. But yeah, I mean, this is a big leg up for these students. I mean, if you prepare kids like this and they go through this kind of training, they're going to be so far ahead of their peers when they get into college. So this is great.
- [53:28]
Mike Kaput
And while it is a huge school district and I realize maybe not every school district can mimic it perfectly, it is really interesting to see them threading that needle of. They're not just saying, throwing up their hands, okay, AI's here. It's gonna change everything. It's like they have guardrails in place with Google Gemini to be able to help the students actually learn and not just click a button and get an answer.
- [53:49]
Paul Raitzer
Yeah, I mean I would love to see this at my kids school. Like a custom version of Gemini that, you know, functions as an advisor, a mentor, a tutor, not as a answer engine.
- [54:05]
Mike Kaput
Well, that's a good segue because speaking of answer engines, Perplexity, the AI powered search startup, is growing fast but burning cash even faster. So according to some new reporting from the information, Perplexity made $34 million last year, but burned about 65 million in cash as it spent heav heavily on cloud servers, AI models from Anthropic and OpenAI. And those power many of the search engines answers. And this is all according to financial documents that were seen by the information. Yet at the same time, it sounds like investors are still buying into Perplexity. They're reportedly raising 500 million at a $14 billion valuation. Behind the scenes, it's trying to scale quickly. They've got 200 employees, new product lines are being teased like a comet browser A budding ad business and experiments in E commerce. It scooped up teams from other AI startups like Sidekick and Rhymes AI. But they're still not really making as much of a tent in search as the other giants in the space. Google still handles 900 times more searches daily. According to the Information, ChatGPT handles at least 25 times more per day than Perplexity. Paul, how do you rate Perplexity's prospects in the market right now? I mean it's no doubt useful, but it seems like the value proposition is kind of being cannibalized by ChatGPT, which increasingly delivers accurate web results. And Google which is increasingly delivering AI powered search results, feels like they're getting squeezed on all sides.
- [55:45]
Paul Raitzer
Sides? I do not understand that. Valuation. 34 million last year at a 14 billion dollar valuation. That's crazy.
- [55:57]
Mike Kaput
Some AI math right there.
- [55:59]
Paul Raitzer
Yeah, I mean even if you assume a hundred million dollar run rate 2025, it's still just bonkers. The only thing I can come up with is I did see articles in the last couple days that they're in talks with Samsung and Apple to integrate to those devices. So it sounds like the Samsung deal is pretty close. Which it says two companies are talked to preload Perplexity's app and assistant on upcoming Samsung devices and integrate the startup search features into Samsung web browser. They've also discussed weaving Perplexity's technology to Samsung's Bixby virtual assistant. So I don't know, maybe they have some multi billion dollar deals sitting here with Samsung and Apple that aren't publicly known yet, which then would make it a little bit more reasonable. But yeah, I mean that multiple's nuts. But what is open AI? Is it what are they thinking? Like a $15 billion maybe this year? 15, 20 billion and they're valued at 300 billion.
- [56:59]
Mike Kaput
Yep.
- [57:00]
Paul Raitzer
I can't do that math in my head roll.
- [57:02]
Mike Kaput
No, yeah, yeah.
- [57:04]
Paul Raitzer
It just doesn't add up to me.
- [57:06]
Mike Kaput
Like but scale of Perplexity compared to that. Right.
- [57:10]
Paul Raitzer
Yeah. And I just again I, I, I've, I perplex, I, I love perplexity. Early on I was like I used it all the time. I know Mike, you were a big, you were used before. I was kind of convinced me to give it a go. I, I do just struggle to see it how it differentiates.
- [57:24]
Mike Kaput
Right.
- [57:24]
Paul Raitzer
Moving forward since it's just built on everybody else's models and those models are all going to do a better job of the things that Perplexity is trying to do and I feel like they're just scattering to all these different niches to try and like find somewhere to lock in and, and maybe it works out, or maybe they just get Aqua hired at some point, which I think is probably more realistic.
- [57:45]
Mike Kaput
Our next Rapid Fire Topic Box has released a 2025 survey of over 1300 IT leaders that were pulled between April and May of this year and they found some really interesting data around this. So 94% of the organizations where the people work that they survey are already using AI, but there's a pretty big gap between the dabblers and the doers. The highest proportion of respondents, 47%, say they're in the early stages of AI adoption, which includes pilot projects and kind of limited deployments. But those who are further along are already seeing some pretty serious benefits. So the survey shows that companies that consider themselves on the quote, leading edge of AI adoption are seeing 37% productivity gains on average from AI. Now, the vast majority also measure the success of AI initiatives in things related to productivity. When they were asked like how are you measuring the success of AI projects? Time savings was the highest proportion with 64% answering that. That was followed by employee productivity metrics at 51% and cost reductions at 43%. Interestingly, the report also finds the most common use cases for AI at the company surveyed are things like writing emails and communications, doing document analysis and getting insights from those docs and general purpose AI chat and research. Now what I found kind of interesting and made me definitely double take is the report also asks how mature is your adoption of AI agents for your IT system? Them and 87% say they're at least piloting basic AI agents in some way, including 41% piloting fully autonomous operations in select domains. Last but not least, respondents were also asked about how they're addressing the AI skills gap in their companies, with 58% the highest proportion saying they're relying on upskilling the existing workforce. Now, Paul, this is it focused but definitely interesting to see what some of these answers are. I was actually struck by some of the parallels to the 2025 State of Marketing AI report we put out last month because we also found companies are often measuring success in terms of time savings. We found that the majority of companies are still in that kind of piloting AI or experimenting with AI phase. So not apples to apples, but I did think it was interesting how some of this rhymed here. What did you think of some of the data here?
- [60:19]
Paul Raitzer
I think it's a good just perspective. We always talk about the research. You have to understand the audience that was polled, that was surveyed in, in the research and what their roles are, what the industries they're in, the size of the companies they're at. So there's no like right or wrong. I think the whole point we try and push on this show is be, be open to all the different perspectives with all the different data and then you have to like use that to of draw your own conclusions at kind of where we are. So I love these kinds of reports that are done in a thorough way and provide some perspectives on what that audience is looking at. So yeah, I thought it was good. I wanted to make sure we address this on the show and whenever we see research worth sharing, we try and share it. It's great use for Notebook om, like if you want to dig into this, if you're interested in this topic, like go grab the report, throw it in notebook from Google and have a conversation with it, build a study guide, do those kinds of things. That's, I love doing that with research reports, especially more dense ones.
- [61:22]
Mike Kaput
Yeah. And, and Box's CEO Aaron Levy is definitely worth following on because I think as well he's got pretty good commentary comes from, from a certain perspective as what I think to be useful commentary on.
- [61:34]
Paul Raitzer
Yeah, we cite him, we've cited him quite a bit. He is, yeah, he's one that I, I do get alerts from him. He is very optimistic about the future of work. Like every once in a while wanted to like leave a comment and I'm like, ah, it's not even worth it. He is very much in the camp of it's all going to work out and more jobs are going to be created. And I'm not sure where the optimism comes from. Honestly, like I've tried to read and like fully understand his perspective. I want to share that optimism. I'm just not quite there right now.
- [62:07]
Mike Kaput
I feel that, you know, honestly too, I feel like his survey respondents were really optimistic because I found this like 87% saying they're piloting basic AI agents to be a crazy high number. I don't, I don't work in it, so maybe that's it.
- [62:21]
Paul Raitzer
But there's just a very generous definition of that's what agent is right now, I would say.
- [62:29]
Mike Kaput
All right, next up, grief, love therapy. They're all going to digital thanks to AI and unfortunately the consequences may be getting ready to pile up as well. According to a new report in Psychology Today, now they look at how AI companies are now offering simulated versions of platonic companions, just like friendships, romantic companions, like boyfriends and girlfriends. And AI companies are also offering simulated versions of the deceased, promising comfort through avatars that mimic a loved one's voice and mannerisms after they've passed away. And even if that sounds a little strange to you, users appear to be relying on AI companions more than ever, according to this report. So the market for AI companionship was valued at 2.8 billion in 2024 and is projected at 9.5 billion by 2028. Now, the issue here, which Psychology Today points out, and that's kind of the focus of this article, is that while AI girlfriends, grief bots, therapy apps, they're booming, there's not that much research yet on what happens when we outsource emotional processing to AI chatbots. So some get a lot of affirmation from these AI companions. But while they can soothe you, they may also have bad consequences, like isolating you, interrupting, you know, very natural, though painful mourning processes, and even straining or replacing real relationships. So, Paul, I feel like we've been dancing around this one for a while. Like, it seems like we have to accept there's some type of real demand, whether we agree with it or not, from people for AI companionship. Now it's up for debate, like how far down the rabbit hole they're going with those companions. But on the other, this just seems like an obviously really slippery slope, especially when it comes to the stuff with the deceased.
- [64:31]
Paul Raitzer
So anybody who listens, listens regularly. Episode 149. This is the topic I mentioned we had to cut because I just wasn't mentally there to discuss this one. So this is the grieving one I alluded to. Almost didn't make the cut again this week. So this is a rapid fire item. I'm just going to address this kind of quickly and then we'll probably have to come back around to this one again down the road. So when I owned my marketing agency, one of our largest clients for like a decade was a funeral home. And so I spent a lot of time in that industry thinking about that industry. And when I created the Marketing Institute in 2016, it was one of the industries. Like, I would talk to them about it and say, hey, listen, here's what I think is going to happen in your industry. It's sort of an inevitability that we'll be able to sort of digitize the dead. You can train these models on video audio. This is before Geni, this is before Chat, gbt, and that just accelerated it. So imagine being able to conduct interviews, take all these videos, all this audio and be able to, to train in its simplest term, a custom GPT kind of thing in a more advanced firm, a digital avatar with like a Void engine behind it where it's like, you know, seems real and it's trained to like behave and talk like the deceased. And I just always kind of assumed that it was inevitable because we live in a capitalistic society and there is money to be made doing this, whether it should happen or not. That is, I think at the end of the day it's going to be for individuals to decide. I am convinced you will have that option at some point in the near future. The commercial potential of this is far too great for venture capital firms to not fund this. So this is something I assume is coming and there's probably already some forms of this that we're not. I haven't researched the market yet to like really understand deeply what's going on. But like just to frame this. Here's the opening to the article Mike mentioned. Imagine this. In the final months of her life, your mother, while in palliative care, paid an AI company to create a digital replica of herself. The pitch was simple. This AI avatar would ease your grief, allowing her to live on for you and your children. Now, months after her death, you speak to this simulation almost daily. The voice is 70% accurate, the video nearly lifelike, and the illusion brings comfort. Yet your dependence on this digital ghost has trapped you in a state of suspended mourning. So like I said, this is a rapid fire topic. I could spend five hours on this one. Like this is probably over the last 10 years, one of the applications of AI have spent more time than most thinking about. And so I, I just want like people to start preparing for the fact that this will be a part of society. And I think that it's important that we start considering the perspectives around this and start, you know, understanding. And I, if psychologists and people like that aren't already proactively working on this, I think we need to, we need research in this area, if it doesn't exist already, about the impact it has when people can't go through the normal grieving process and the decisions people are going to have to make. Because I get it, like when you lose someone close, especially in a tragic way, potentially where you just want to, you know, key and, and this is, that's what I'm saying. There's no right or wrong here. There's just going to be. It is or is not. And I think it's, it is, it will, it will exist. The technology will exist. And we have, as a society, have to start preparing for that and what that means. And I'm not qualified to be the one to, you know, say that I can express personally how I feel and things I've been through. But I think at some point we'll probably have to pull in some experts and maybe have some, like, special spin off episodes where we just talk about this kind of stuff. Because I just think it's going to be really critical to society to. To understand that this technology is going to be there and it's going to get weird. Really weird.
- [68:44]
Mike Kaput
Yeah. I mean, reading that article alone, I think it's already getting weird.
- [68:47]
Paul Raitzer
Yeah. Yeah.
- [68:50]
Mike Kaput
All right. Our final topic this week is a project that researchers at Sakana AI are building called the Darwin Godel machine. And this is an AI system that rewrites its own code to improve itself. So this is an experimental agent they published about that evolves like a digital species. Each generation tweaks its own source code, then tests those changes on programming tasks. If the new version performs better, it survives and gets added to a growing archive of agents, over 80 iterations. They say this loop led to really big performance gains. So the system boosted its score on SWE Bench, which is a popular benchmark benchmark for coding, from 20% to 50%. And on a multi language benchmark called Polyglot, it went from 14% to over 30% accuracy, all without updating the foundation model that powered it. So in other words, it's AI that can learn indefinitely because it can update its own code. And that's a really interesting development. Developing self referential, open ended, improving AI. Now, Paul, the actual paper here related to this gets really technical, but the overall point seems to be that this could be a breakthrough that basically enables AI without updating that foundation model to get smarter. So what's important to be thinking about here and paying attention to?
- [70:19]
Paul Raitzer
We just wanted to make sure we address this one on the show. In the Road to AGI series, the first episode I talked about, you know, one of the things that could drive, like accelerate model development is self improvement, if these things can actually like improve themselves. And that's kind of the premise here. We talked about Sakana. I want to say it was like toward the end of last year. I know we mentioned them on the podcast because I think they're backed by some pretty significant players in the AI space, which is what caught our attention. I would say it may be a little while before you hear more about this in terms that would matter to you as like a business leader or practitioner. But this concept is one of the fundamental ways that the labs think they can rapidly advance these models to AGI and beyond. And so I just wanted to make sure people were kind of aware of this progress. This is probably a topic I'll go further into in the Rode AGI series as I start start, you know, building out those episodes. But yeah, I, I would not try and digest this research report. I. If you want to try it in Notebook, Lemon, say, give me this at like a fifth grade level. What does this mean?
- [71:32]
Mike Kaput
Right.
- [71:33]
Paul Raitzer
You might have some success there, but overall this is pretty dense stuff and it's probably, you know, six to 12 to 18 months away from finding its way into like major models. But progress is always being made on the research front and, and sometimes there's just research papers that surface and you're like, that one's going to be important. And I would put this in that category of like you just as soon as you see it, it's like, okay, flag. That one is like one to come back to.
- [72:01]
Mike Kaput
All right, Paul, that's a wrap on another busy week in AI. Appreciate you breaking everything down for us and demystifying what's going on out there.
- [72:09]
Paul Raitzer
Yeah, good stuff. And definitely check the the show notes. There's a bunch of newsletter only stuff too that we had to cut like Grammarly getting a billion dollars and going after an AI platform, interviews with, you know, Google, Elon Musk doing some stuff with Sam Altman and trying to block his deal. There's all kinds of fascinating reads. So yeah, as always, thanks everyone for joining us and thanks Mike for curating everything and pulling this together. Thanks for listening to the Artificial intelligence show. Visit SmarterX AI to continue on your AI learning journey and join more than 100,000 professionals and business business leaders who have subscribed to our weekly newsletters, downloaded AI blueprints, attended virtual and in person events, taken online AI courses and earned professional certificates from our AI Academy and engaged in the Marketing AI Institute Slack community. Until next time, stay curious and explore AI.