← The Artificial Intelligence Show
The Artificial Intelligence Show
#167: OpenAI-Microsoft Deal, Replit Agent 3, AI Avatars for Executives, OpenAI-Oracle Deal, FTC Targets AI Companions & Retail AI Case Studies
01:25:26
Loading summary
Transcript73 lines
- [00:00]
A
You can either be the hero or the villain here. This technology is going to disrupt a society that is inevitable. You have to get ahead of this. Your tech is going to disrupt people's jobs. It's going to do a lot of negative things, but it's also going to do a bunch of whole like incredible things. And you have to be leading in that way. You have to be proactive in that way to be viewed as someone doing good for humanity, while your technology might be doing the opposite sometimes. Welcome to the Artificial Intelligence show, the podcast that helps your business grow smarter by making a AI approachable and actionable. My name is Paul Raitzer. I'm the founder and CEO of SmartRx and 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 167 of the Artificial Intelligence Show. I'm your host Paul Raitzer along with my co host Mike Kaput. We are recording September 15th, 9:15am.
- [01:15]
B
We.
- [01:15]
A
Are like only like four weeks from Macon. I was thinking about this over the weekend, like oh my gosh, we gotta start building some, some stuff for Macon. So we are, we are coming up fast in our big conference. So I guess I'll lead with Macon since I'm talking about already. So October 14th through the 16th. If you don't know what I'm talking about, if you're new to the podcast and haven't heard us mention this before, our flagship in person event, Macon is happening October 14th to the 16th in Cleveland. We started this conference in 2019. This year is trending in an incredible direction. We are looking at probably 1500 plus attendees. The majority of the agenda is live at Macon AI. That's M A I con AI. We actually made some really exciting progress on some new keynotes last week. So I can't announce anything yet, but stay tuned. Hopefully by this time next week we might actually be able to announce some of the might be a final agenda. There's a couple of moving pieces still on that main stage, but we're getting real close so go check it out again. It's Macon AI m a I c o n AI dozens of sessions. You know, I think 40 plus speakers, just an incredible lineup. I can't wait. I just need to create my opening keynote. So I'm doing the move 37 moment for knowledge workers. Basically it's, it's probably the most excited I've ever been to create a keynote. I've every year. I love doing my make on keynote. It's always an original talk. This is probably the one I've, I'm most excited to create. It's something I've been working on for years and I just sort of forced myself to say, okay, I'm creating this talk for, for Macon this year. So it is completely original because it doesn't exist yet and I have four weeks to create. Mike is a session. We both got workshops. Just it's going to be awesome. So we would love to see our community there in person. Again. It's Macon AI. And then this episode is also brought to us by AI Academy by SmartRx. We've been talking a lot about that lately. It is a big focus of me, of my time, of Mike's time. We've built out our staff to, to build out AI Academy. So it's where a lot of our energy and resources are going. We've talked about some of the new series. So today I was just going to mention the Scaling AI course series. This is the third one that I created for the launch. It's a seven course series with professional certificate. It's built for, I don't know, I have to say like director level and above, but kind of like existing leaders or emerging leaders. People want to understand and take a leadership role in the adoption and scaling of AI within their organization. So it's got the AI forward. Organization is course one, AI Academy is course two, Course three is the AI Council, Course four is Generative AI policies, five is Responsible AI Principles, six is AI Impact Assessments and seven is the AI roadmap. So it really takes you on a journey of our five step framework for scaling AI within an organization of any size. You can learn more about that at Academy SmarterX AI and for both the for Macon and the AI Academy Mastery membership, you can use Pod100 as your promo code. That'll save you a hundred dollars off of either of those. So again, pod 100 for Academy and Macon. All right, let's get into it. We had, I guess we ended up with a new model ish last week, new agent 3 from Replit, which we're going to talk about and some progress on the OpenAI Microsoft partnership that might set the stage for some pretty wild stuff this fall. But. All right, Mike, let's get into it. The OpenAI Microsoft thing to start.
- [04:51]
B
All right, Paul, so OpenAI is a bit further along on its path to becoming a for profit company because this week OpenAI and Microsoft struck a deal that clears one of the bigger hurdles to this transition, which is Microsoft's approval. So after a summer of fraught negotiations, the two AI giants have agreed to extend their partnership. According to a statement from OpenAI. Quote, OpenAI and Microsoft have signed a non binding memorandum of understanding for the next phase of our partnership. We are actively working to finalize contractual terms in a definitive agreement. Together, we remain focused on delivering the best AI tools for everyone, grounded in our shared commitment to safety. So if this kind of goes through, Microsoft's tentative blessing seems to give OpenAI the green light. It needs to start presenting its for profit restructuring plan to state regulators. Now, that plan would shift OpenAI from a non profit controlled subsidiary into a for profit entity, one where Microsoft and the nonprofit itself would each hold roughly 30%. The rest would go to employees and investors. Now, this overall plan to transition to this for profit company is facing some fierce pushback. California and Delaware attorneys general are investigating whether the shift violates nonprofit law. Critics including Elon Musk and meta claim OpenAI is abandoning its mission and enriching insiders. And some have even started citing these tragic incidents. We're hearing more about involving ChatGPT and relationships with people to question the company's priorities. Now, the pressure has been so intense that the reporting showed that at one point, OpenAI's execs even reportedly discussed leaving California altogether. So, Paul, there's a lot going on here and a lot of implications, even of a short memorandum of understanding. What does this actually mean for OpenAI.
- [06:52]
A
Going forward at a really high level? If you're new to all of this? Again, if you've been a long time listener to the podcast, this is a recurring topic every few episodes. It seems there's something else related to this. The OpenAI Microsoft relationship and OpenAI's efforts to evolve the company structure. But again, if you're new to all of this, the basic premise here is Microsoft controls a lot of interest in OpenAI. They've invested over $13 billion into them. There is a contract that was created originally where Microsoft got access the most advanced OpenAI technology, but that if OpenAI determined they reached AGI, then Microsoft would no longer get access to this. It was a big sticking point, but Microsoft has a bunch of leverage as well. So the challenge with this relationship is OpenAI needs to raise insane amounts of money unparalleled in human history, amounts of money they think Trillions of dollars, never before done. But the only way they're going to do that is to go public. They're going to eventually not be able to do this as a private company, especially under the control of the nonprofit as it was previously structured. So all of this is all about moving OpenAI to a place where they can raise the amount of money needed to pursue their vision for omnipresent intelligence throughout society, basically. So that's like synopsis of what's going on. So there's all kinds of legal give and take that needs to happen behind the scenes. This is not easy. And then Mike, as you called out, there's not even like a given that they're going to get approval from California and Delaware or that some lawsuit like from Elon Musk isn't going to muddy all this up and just make this go on for a while. So I think everyone kind of assumes this will just work out, that like this will all get solved somehow. The lawyers will do what they do and we will find a way to like move on and OpenAI will eventually IPO and you know, become one of the most valuable companies in the world. It's not a given though. Like there's, there's all this stuff happening behind the scenes. So the joint statement, Mike, that you read was, I mean, literally, we'll put the link in the show notes. That's it. It was.
- [08:58]
B
That's the whole thing.
- [08:59]
A
Yeah, that is the post. It's like the, I don't know, 50 words that Mike read. Now it does link to a post from Brett Taylor who we talked about, I think last week on the podcast, if I'm not mistaken, who is the chairman of the board CEO of Sierra. We were talking about his AI agent, startup Sierra, former board chair of Twitter. So Brett Taylor put out a little bit more extensive post about OpenAI's nonprofit and public benefit corporation vision. So I'll read a couple of excerpts from his post and we'll again include this in the show notes. So OpenAI's planned evolution will see the existing OpenAI nonprofit both control a public benefit corporation or PBC, which is also how anthropic is structured by the way, and share directly in its success. OpenAI started as a nonprofit, remains one today and will continue to be one with the nonprofit holding the authority that guides our future. This new equity stake would exceed $100 billion, making it one of the most well resourced philanthropic organizations in the world. This recapitalization would also enable us to raise the capital required to accomplish our mission and ensure that as OpenAI's PBC public benefit Corporation grows, so will the non profit's resources, allowing us to bring it to historical levels of community impact. We'll come back to that in a minute. That's an important part of this. The structure reaffirms that our core mission remains ensuring AGA benefit. AGI benefits all of humanity. We continue to work with California and Delaware Attorneys General as an important part of our strengthening our approach, and we remain committed to learning and acting with urgency to ensure our tools are helpful and safe for everyone while advancing safety as an industry wide priority. So that sentence there is very intentional, based on what Mike alluded to of the growing concerns, including I think it's the FTC we'll talk about in a minute that's exploring the use of ChatGPT as a companion and the impact it's had on some recent cases of suicides. So there's a lot of very, very strategic language in this post from Brett Taylor and then the final excerpt I'll read, which leads into the other point I want to make here. So Brett continued to write, as part of this next phase, the OpenAI nonprofit has launched a call for applications for the first wave of a $50 million grant initiative to support nonprofit and community organizations in three areas. And now we're about to get our preview of what this hundred billion dollar stake in OpenAI is going to be. For the three areas Brett calls out AI literacy and public understanding, community innovation and economic opportunity. He writes, this is just the beginning. Our recapitalization would unlock the ability to do much more now, the $50 million grant initiative. They allude to links to another post from September 8th on the OpenAI website. I don't remember if we talked about this, Mike. This is the first time I recall seeing the name of this fund. But this fund that's going to provide the $50 million grants is a People First AI fund. That is the name of it. People First AI fund. So what is the People First AI fund going to do? Which, by the way, this is part of the reason we on the messaging we are very intentional about not always saying AI first because it implies people aren't first. So it's interesting to me that OpenAI is sort of like leaning in this People first direction with their messaging. Okay, so this is now excerpt from this post, which we also will link to the Literally the URL is People First AI Fund. We believe AI should help solve humanity's hardest problems and that we should listen to and learn from Organizations already leading that work on the front lines today. We are excited to share that applications for the first wave of grants are open. Grants will be unrestricted, reflecting our commitment to support the expertise of nonprofit and community based organizations. Application window will close on October 8, 2025. So if this is you, like if what we're reading here fits you, you got three weeks to get your application in for the first slug of 50 million. But there's 100 billion more coming, so don't worry about it. Grants will be distributed by year's end. So what they're basically doing here is they're racing to distribute money to show their positive impact on society and people. So as they're going to California and Delaware begging for the permission to do what they need to do, they're already taking action to say, look it, if we have access to this money, this is the kind of thing our nonprofit will be able to do. We've already done it. Now mind you, there's probably like an actual like human good intended behind this, but this is all very intentionally being accelerated to show a positive impact on society in, in my opinion. So the People First AI Fund will support organizations directly working in the three areas that we we called out from Brett Taylor's post. AI literacy and public understanding. They now this is from their post. We seek to support organizations that help communities build the knowledge, skills and confidence to navigate the age of artificial intelligence. This includes education programs, media initiatives and opportunities for people to engage with and better understand the technology. They're specifically interested in equipping people with practical skills. That may involve training local leaders such as educators, faith leaders, youth mentors and artists. The community innovation side they say the priority is to take is to back efforts to ensure AI strengthens civic life and helps people stay healthy, connected and thriving. And then economic opportunity. This could include programs that prepare people, especially young people, for the jobs of the future, future tools that support caregivers and local businesses, and initiatives that help workers build economic security. They do say the fund is the early step in a larger vision to ensure the intelligence age is shaped by listening, learning and building with not for communities. We look forward to working with our grant partners and learning from it. And then they do call again at the end that The People First AI Fund is intended for US based nonprofits with valid 501C3 status. Organizations may only apply once to be considered for the fund. So a lot going on here. Like I said, the biggest issue is they have to change structure to IPO to raise the amount of money they envision needing to do that. There's a whole lot of stuff that happens, has to happen behind the scenes, including a lot of politics that has to happen behind the scenes. And they need to be seen. Like I've used this, this line, I don't know if I've said this publicly, but I've used this line with some technology companies. You can either be the hero or the villain here. Like this technology is going to disrupt a society that is inevitable. It will be viewed as a negative by large portions of society as they are impacted by it. The way as a technology company, again, part of this is putting my communications marketing hat on. You have to, you have to get out ahead of this. Like your tech is going to disrupt people's jobs. It's going to do a lot of negative things in society, but it's also going to do a bunch of whole like incredible things in society. And you have to be leading in that way. You have to be proactive in that way to be viewed as someone doing good for humanity while your technology might be doing the opposite sometimes. So this is where lobbying efforts are going to be massive. It's where building this kind of program where you're giving tons of money away, the economic opportunities, opportunity, I can almost guarantee you, has some element of universal basic income envisioned into it where they're going to pay people to not have jobs like it. This is, we're leading off with this episode because this is a very, very far reaching topic that will, if you understand what's going on with OpenAI, you will have a greater grasp of what's going to be happening in society for the next like 10 years, basically.
- [16:44]
B
Not to mention, this might be going a bit out on a limb here, but it is not simply, in my opinion, a story of a massive high growth company trying to get ahead regulation or government overreach. At some point the scale of this gets so large, this is intimately entwined with the government at some point. We've talked about that a bit in the situational awareness from Leopold Aschenbrenner. I'm not saying that OpenAI gets nationalized, but what you're talking about when it comes to universal basic income, that inherently becomes, whether it's OpenAI doing it or a consortium horsem of companies that becomes very intimately linked with society and civic life, not just market.
- [17:29]
A
Yes, and, and we've talked a lot about jobs and the economy recently. We've got another topic today related to this. All of this is intertwined. Yes, you're 100 right, Mike. The executive order to prioritize and fund AI literacy through the government. You're going to see, you know, an avalanche of, of state and federal initiatives around AI literacy and reskilling professionals. Because again, they all now know what's happening. Jobs numbers are starting to indicate it. And that's what we've been waiting for. Well, not we. So what they have been waiting for is like actual data to prove this is all happening. So, yeah, I mean, I think we're, we're kind of entering this point of no return where all the people in power now realize the disruptive nature of AI and how quickly it could happen. And so there's going to be this massive effort, both private and federal, to, you know, private and government to try and solve for this before it becomes like a runaway train, basically.
- [18:31]
B
All right, our second big topic this week, AI coding platform Replit just raised $250 million and tripled its valuation to 3 billion. But the real headline is that it also launched something called Agent 3. This is a next gen AI developer agent that can build apps almost entirely on its own. So Agent 3 doesn't just suggest code, it actually tests. It fixes bugs and clicks through your app like a real user to make sure everything works, all without needing constant human input. So CEO Amjad Massad said in a post on X quote, Agent 3 is 10x more autonomous. It keeps going where others get stuck. The full self driving moment of software. He also says it has 10x longer fully autonomous runs than its predecessor Agent 2. So Agent 3 can run autonomously, fully autonomous for over three hours straight, which is about 10x more than Agent 2 could do now. Also behind the scenes, Replit's growth has been honestly insane. Annual revenue, this is not a typo, jumped from 2.8 million to 150 million in under a year. There are now 40 million users and they have enterprise clients like companies like Zillow and Duolingo. So, Paul, first maybe talk to me about Agent 3, about Replit's claims here about it being 10x more autonomous. Like, is that legit and what does that actually mean? Has there been some breakthrough in agents here?
- [20:08]
A
So this is a really important concept to understand and also to delineate between some of the ways we've talked about previous measurement of autonomy and how they're kind of positioning it. So at a really high level. Like the main takeaway here is the way these labs are thinking about the future of AI development and the impact it'll have on society and the economy is how long can These things work reliably without human intervention. And so if you remember again, if you've been listening to the podcast for a while, episode 152 in it was June 10th of this year. And then earlier in episode 140 which was in March of 2025, we talked about this seven month rule which is meter metr. It's model evaluation and threat research. So it's an organization, CEOs Beth Barnes and they, what they have is the seven month rule which is they look at the model's ability like a 50% chance of successfully completing a task of how long it would take a human to do it. And what they're saying is every seven months it's doubling. So right now the latest meter research was in March of this year was that AI models had the 50% chance of successfully completing a task that would take an expert human one hour. So now this was specific to coding and that's, that's a really important criteria. Anything we're talking about right now, whether it's related to ReKit, Agent 3 or this meter rule is all related to like computer programming. In Replit's case it's like building apps, you know, writing software, basically, not doing legal work or doing marketing work or things like that. So we're like separate out that we're talking specifically about software at this point in coding. So in the, in the meter research, so it was an hour in March, seven months, prior to that it was 30 minutes, seven months prior to that it was 15 minutes. So what they're finding, and it's kind of like a potential scaling law, is that every seven months it's, it's doubling. So you know, the theory then would be by August you should be at basically 2 hours, 50% chance of 2 hours. So now interestingly when GPT5 came out, they were given meter was given early access to it a couple of weeks in advance and they did find that it was basically continuing to double. So they found that GPT5 is capable of executing in their specific testing any of three main threat models. They use their time horizon evaluation. And this 50% chance of success had jumped to 2 hours and 17 minutes. So it went from 1 hour to 2 hours 17. So the scaling law was now continuing. So we had like four different milestones within it. But they did this going back six years and so it has held. That is different than what we're seeing with Agent 3. So again the meter is saying 50% chance of it completing something that would take a human one hour. What REPLIT isn't telling us is how long would it take a human to do what you're saying it's doing? All, all replit's giving us is runtime, meaning the agent does this thing. I don't, I don't think. I didn't see anywhere. They said reliability like 50% chance. It was just it can do something for 200 minutes straight without kind of losing its thought process, losing its planning structure, things like that. Now, is that 200 minutes of agent work equivalent to 20 hours of human work? Like what would it have taken a human to do it? That's what they did not share. And it was interesting because Amjad actually tweeted about the meter research. He retweeted a post from like March about the meter research and said, yeah, we're actually seeing something different. So his tweet was the meter paper says that the length of tasks AI can do is doubling every seven months radically undersells the scaling that we're seeing at replit. It might be true if you're measuring one long trajectory for a single model class, but this is where an agent research lab's alpha is at. We build multi agent architecture and use different models from various providers to tap into their latent abilities across varied tasks. So I'm going to zoom out and explain what that means. What he's saying is meter's research, which is the best we've seen publicly so far, that projects every seven months, it doubles. So if you take a human and they. It takes them one hour to do something, in seven months it'll be able to do two hours of that human work. What he's saying is we actually rapidly scale beyond this by using multiple agents. So we're going to have an agent that verifies things. We're going to have an agent that creates the plan. We're going to. So they're using multiple model providers because Replica doesn't build their own models. They're basically using, let's say they imagine they're using like an anthropic model for something, an open AI model for something. Maybe they're using Gemini for something. I don't know what the architecture is, but they're able to use different agents to do all of these things simultaneously. So what they've their Agent 1 in September 2024 did two minutes of runtime. Agent 2 in February 2025 was 20 minutes of runtime. Agent 3 in September 2025 is 200 minutes. So they're kind of following more. What is this? Every six months it's 10xing. Is that right?
- [25:24]
B
Roughly, yeah.
- [25:25]
A
Yeah. So they're, they're on a whole nother level of scale of runtime. Now the missing piece of this equation is what is that? 200 minutes equal in human time. All of that extracted. If you go to their Website it is 10x autonomy. Like they are all in on. We are automating human work in this specific instance. Automating computer programming is what they're actually doing. The key for the economy is when do we see these kinds of scaling laws in healthcare, in finance, in marketing and sales and customer success and operations. We don't have that research right now. There aren't evals that are looking at that. That is the thing. And Mike and I have had, I don't think I've publicly said this before. Mike and I have had these conversations internally of like the need for those level of evals at an industry level or at a job specific level. Because everything we're able to share with you today is coming from people who are doing this for AI research and computer programming. It's when you translate it over to your job and you say, wow, Mike spends two hours a week on this task. This AI agent is able to do that in two minutes with 90% accuracy. Things change when those the kinds of statements we can make. And so that's the significance. And why again, like some of you, if again if you're newer to all this, like Replit is not a household brand in enterprises. Like that's not probably, it's not on the open AI level, the anthropic level where you've probably heard of them. They matter though and like what they're doing matters. And Amjad is a very intelligent CEO and they're doing incredible research and this is the kind of stuff you will see. Google's working on the same kind of thing. OpenAI, everybody's working on this. They just kind of seem to be the one that's like publicly leading the way, talking about it and they've found ways to deal with the inefficiencies of computer use. Agents from anthropic and OpenAI and Google that aren't working very well, they claim they've found ways to fix that. So very, very interesting research direction and potentially important breakthroughs. When we like look back on this a year from now and really important.
- [27:35]
B
To note I was digging into where did all that growth come from? Because 2.8 million to 150 million in less than a year is crazy. It's largely driven by agents so that can tell you A, I think they know what they're talking about, but B, there's a vested interest here, right. In promoting what agents can do because that is driving their business. I think at one point their business was really on the rocks before all this. So it's amazing to see. I have nothing against replit but just also important to note, like before we go tweeting that chart where they show 10x autonomy. Right. Keep in mind that we are all who have vested interests in certain things and theirs is an agent's being. Amazing.
- [28:15]
A
Yeah. And it doesn't mean it's that for your job. And again, like, is it. The whole problem with autonomy is when it's talked about people and even agents, you know, more specifically, they assume it's transferable. That like, oh man, if they're 10xing autonomy, that means my job as, you know, an HR professional or a lawyer or a CEO. I can 10x my performance, my productivity if I go get replit. No, that is not what we're saying at all. But it's a prelude too. The breakthroughs happen where the greatest value will be created, which right now is in AI research and engineering and programming. And that's why all the labs are starting there. But when they saw how to do it there in reliable ways, then it trickles out into all the other industries.
- [29:05]
B
And I'm super bullish on agents long term, like we've discussed. I would just say if you are in a non coding role, go try out OpenAI agent mode on three different, very different tasks and then come back and you'll have a much better, a much more sober look at what's possible and what's not. And it's amazing. I like it. But it's, you'll see the limitations pretty quick.
- [29:24]
A
And deep research is another great example. Like it's incredibly impressive, but it still just has its flaws. And that's a great way to experiment with an agent is like go build a deep research project in Gemini or in ChatGPT and you'll see an agent at work, you'll see it do its planning, you'll see it go through a chain of thought. You'll see it, but it's not doing those like self correcting and verifying the outputs. But that's all going to come. And that's, that's the example here. That's basically what they've solved in programming is it can do the planning, it create the, you know, the plan to follow. It can go through a chain of thought. But now they have agents within this architecture that, that then look at the work and find flaws and then self correct the work and then they like keep going that when they hit a blockade, they fix it themselves. You don't have that in like deep research right now in Gemini and chatgpt, but it's coming. Like all of this is a prelude to what will happen to the general tools the rest of us use.
- [30:16]
B
All right, our third big topic this week is answering the following question. Should your company's next thought leader be an AI avatar? We've been kicking around that question internally because someone in our network who know well, Databox CEO Peter Caputa has debuted a new video course taught entirely by his AI double. So he posted about this on LinkedIn pretty extensively. Both his experiments with AI avatars and now that one is going to be teaching a course from him. And this avatar looks and sounds like him. It's powered under the hood by the popular AI video and avatar tool HeyGen. Caputo wrote the script and it's trained all on, you know, not only his visuals, but all of his knowledge and expertise that he's teaching in the course. So he said after some trial and error with lighting and camera angles, this AI avatar now delivers hours of content on his behalf. And he says it saves time while preserving that unique value he has to share because it's been trained on his content, his expertise. It's not just coming up with it on its own. Now what's interesting is plenty of people seem to agree that using hey Gen for this kind of thing is useful because they just raised $60 million at a $500 million valuation and they've got more than 40,000 businesses using the technology. So Paul, like, the question here is like, just because you can do this, does this mean you should?
- [31:41]
A
This is an interesting one. So for context, Peter and I go back a really long time. It's actually hard to imagine, but it's like eight, 18 years we've known each other. Oh, wow. So Pete was the architect of the HubSpot Partner Program. So anyone who doesn't know, my agency that I sold in 2021 was HubSpot's first partner back in 2007. So we were the origin of their HubSpot partner ecosystem. Pete was the guy internally within HubSpot that pushed heavily to build around outside partners that he, he believed that agencies could build, you know, be a value added reseller network. There were other people within HubSpot that believe this also. This is, I mean, this is like year two or three of HubSpot. This is the very, very early days. And so I would sit in meetings and we would talk about these things and, like, they would look at what we were doing at my agency and think, wow, could we scale that to like, thousands of agency partners that can resell and then add value added resellers, value added services to the software? So Pete was the guy who had an agency, then he came to HubSpot, and so he saw the potential for this. So in the very early days of HubSpot, like 2007, 2008, Pete and I worked very closely on what we were doing at my agency and how that could scale. And then my first book, the Marketing agency blueprint in 2011, became kind of a catalyst for the growth of that HubSpot program, in part due to Pete's efforts to, like, push me to share what we were doing. So, backstory. I know Pete very well. So Pete has been the CEO of Databox now since. I don't know. I mean, I feel like it's like seven, eight years. I kind of lose track of time these days. But he's been there for a while. So I see this post From Pete on LinkedIn and I'm like, I don't. I don't know. I don't know. I'm like, sure how I feel about this. But to Pete's credit, like, Pete is a. He loves to stir things up. Like, he is very comfortable with, like, creating conflict and, like, letting people kind of argue out a point. And so knowing that about Pete, I'm like, all right, I'm just kind of. I'm either going to get drawn into, like, posting a comment on here or whatever. So I didn't comment. But then I was thinking about it. And so when I went to write the editorial for the Exec AI newsletter this weekend, the SmartRx executive newsletter, I was like, you know what? I'm going to address this. Like, I think this is a really important topic. And my premise here is I don't agree with him. Like, I actually, I feel the opposite. I think that the human component of the course is. Is maybe the most important part. And so I. I can't even fathom using an AI avatar in my place to teach a course. And as someone who just spent hundreds of hours of my life as a CEO who doesn't have the time to spend hundreds of hours on this, but I just spent hundreds of hours creating 20 courses and recording them myself. Like, no AI avatar involved. I can't even imagine having used an avatar to do it right. That doesn't mean Pete's wrong. And so this is where I kind of land on this. And this is why we wanted to address this as like, a main topic. All of us have to make this choice. The technological limitations of avatars are going away. So the nuances of like, the hands, the, you know, the blinking of the eyes that just kind of like uncanny valley feel where it's like we're kind of there. Like it feels like it might actually be Pete. I'm not sure right now if it's Pete or not that's going away. Like we're going to get to the point where you just don't know and you can't know. Like, unless you have access to the metadata and you know where the thing came from. Videos in the very near future are. Are going to just be in. In discernible from reality. You could argue in some cases, they're already there. There was an. Actually, part of the reason I decided to do this as the newsletter last week was there was a point last week with the president where there was a video in the Oval Office and it went crazy on. On X because people thought it was a AI avatar of him. And I'm not convinced it wasn't. Like there. There was some nuances to it where you're like, yeah, that actually might be. And so now we're in this realm where even in politics, we're not sure. Like, we have to question it and you have to really analyze to know whether it was or not. So my whole point was in. In my newsletter editorial, I'll just read kind of the end because it kind of makes the point here. So for me, per personal connection, authenticity is essential when communicating with my audiences. That doesn't mean I disagree with Caputa's choice and strategy. For him and his brand, it is a subjective decision. There isn't necessarily a right or wrong. If you go to the LinkedIn post, you'll see the comments is like, 50% love it. 50% are like, ew. Like, this feels wrong. But the point is, as you alluded to Mike, like, hey, Gen is blowing up 40,000 business customers worldwide, so more brands and business leaders are choosing the AI avatar path. I think we'll talk about. We have the podcast topic later, Mike, where their AI is using to be generated Podcasts, Podcasts.
- [36:37]
B
Yeah.
- [36:38]
A
So the tech is getting there. We are all going to have to choose how we use AI to create our content, our thought leadership, our expertise. Can't fake me standing on a stage like, you know, that's me. But you're going to, as a brand or as an individual creator or leader, have the choice to fake the other thing. And I'm not saying fake in a bad way, like deep fake the thing. And I know of brands, education brands that are choosing to do this, that are scaling up content with AI avatars. I actually think, well, I don't. It wasn't part of Coursera's announcements, but Coursera really recently made a bunch of like, AI powered announcements about how they're infusing into their platform. And as someone who's the CEO of an AI education company, this is like, we have to think about this, we have to address this with our own team. Like, will we use AI avatars? The answer is no, in case our team is wondering. But, like, this is going to become part of what you're doing and then you as the, the consumer of that content, have to decide if you're okay with it. Like, is knowing that, like he said, literally, like, I spent nine months writing these scripts. This is from 25 years plus of scale, you know, scaling businesses. It's all my experience that is all true. Like, no one's going to take that away from Pete, that you spend a bunch of time on the scripts and you put 25 years into this content, but at the end of the day, it's the AI avatar that's presenting it to me. And so if you go into the comments on LinkedIn, some are like, yeah, the fact that you couldn't take the extra, like five hours just makes me not want to take the five hours to watch it kind of thing. Like, and so again, you're going to have people who are just, I want the information. I don't care if it's AI avatar, whatever. Like, just give me the information. And then you're going to have people feel like, no, this doesn't feel real. Like, I don't, I don't have that same connection with you. No right or wrong. Like, again, the whole point of this conversation is to bring it up as a conversation topic that people might not be aware they have to solve for or might not even know that they could create an AI avatar, their CEO and save a bunch of time. So I don't know. How do you feel about Mike, you and I actually haven't had this conversation of should we use AI? I've just said we aren't, but, like, you know, should we be using avatars in our academy at all based on.
- [38:50]
B
How much time and energy it takes to record courses? I Would love to, but I'm like, really against it personally, because of what you hit on. This is not a knock on Pete. I'm sure his course is amazing. But if you couldn't bother to, like, show up to the studio, it's like, what am I paying for now? Again, the context might matter. Like, what if it's a quick team onboarding training? Okay, maybe, like, you could tell me on that. That's like, oh, okay, Mike's out for two weeks or something.
- [39:16]
A
Here's.
- [39:17]
B
And even that would feel weird to me, though I could certainly grow to accept it because I think that is where we're going. I think it gets a little murky just because I'm so focused on the course thing recently and the value we're creating. I would feel the same way as if someone sent their AI avatar to a meeting. If you're not taking the time to bother to engage with me, even if it's asynchronously via on demand course, I don't have time for it. That's just my personal perspective. I would love to. I know that. I feel confident in that perspective because I'm looking for a reason to want to be able to do this, given how much time it takes to record courses.
- [39:54]
A
Yeah. And I think I was going back to the human to machine scale for writers. Like we talked about. I don't remember what episode that was on, but I shared, like, in. Was that in March or April this year we did REI for Writers Summit, and I did a keynote on, like, when should you use AI to write? And my whole premise was sometimes it's fine, like, product descriptions, things like that. Like, who cares? People just want the information landing pages. But when it's like a keynote presentation or an editorial piece, like, you want to know that that's coming from the person. And so there's. There is a scale of, like, when it's cool to use it, but again, it's not. It's the same for everybody. It's not prescriptive. Like, it is up to you to decide where that comfort level is. And a lot of it comes down to what does your audience expect? And if your audience expects you to show up and be authentically there and to have put those extra two hours in to record the thing, you got to show up and do it.
- [40:52]
B
Yeah.
- [40:52]
A
If it's something that, like your point about, like, I don't know. I don't even know what use case we do internally. But if someone, like, if someone came to me and said, hey, we've Got this onboarding thing. We got this other thing we want to like, you know, teach how to build game plans or use asana. Is it cool if we create an AI avatar of you that tells the story? It's like, I don't know, like, I would step back and think about, well, do the employees feel like it's actually supposed to be me? Like, is that an instance where I could see it being valuable? But again, people are making the case for this every day of, hey, gen. Seeing that kind of growth, it's, it's becoming a real thing. So, yeah, I have, I shouldn't even talk about this right now. So I have, I haven't talked to Mike about this. I'll probably seed this. My team's going to want to kill me. So I want to actually start doing like an AI Pulse is what I was thinking it's going to be. We'll do like, survey our listeners on stuff. And it's interesting because Sunday morning I was, I was sitting on my front patio, like, thinking about how to do this and what it would look like. And I was trying to think like, what would be like the ideal things to ask in these pulses that basically it's like real time research that Mike and I would then turn around and share the next week. This might be actually the perfect thing to kick it off with Mike, is how do you feel about AI avatars? Like, would you. So I'm not saying we're going to do that survey today, but we may in the next week or two. Because I was actually going to schedule a meeting with, with Mike and our team this week to talk about this idea. We may do kind of these quick one to three question surveys of our audience and then turn around and like share the results the next week. Because I would actually be fascinated to know how do people feel about AI avatars? And like, would you create one of yourself? Would you allow it to teach a course? So maybe, maybe that'll be our first AI Pulse survey.
- [42:32]
B
That'd be awesome. All right, let's dive into some rapid fire this week. So first up, OpenAI just signed what might end up being one of the bigger cloud computing deals in history. They have a $300 billion commitment with Oracle to buy computing power over five years, starting in 2027. Now, to put that number in perspective, OpenAI currently brings in about $10 billion a year, so it's committing to spending six times that every year on compute alone. The Oracle deal would require 4.5 gigawatts of power, which is roughly equal to what two Hoover dams generate. Oracle stock soared by as much as 43% on the news, which briefly pushed Chairman Larry Ellison into the top spot as the world's richest person. But this is also a massive gamble. OpenAI is losing money. They don't expect to turn a profit until at least 2029. And Oracle is likely to have to take on significant debt to build out this infrastructure. So Paul, these numbers are pretty staggering. Like how big a win is this for OpenAI and for Oracle?
- [43:40]
A
I mean, stock wise? It was great for Oracle, so I have to laugh about this one. So ironically, this morning I'm driving my kids to school, as I do every day, and somehow the national debt comes up. My kids are seventh and eighth grade, so the 12 and 13. I have no idea of how we got on the national debt topic. It's only a 10 minute ride, so this is a pretty heavy topic to have to cover and explain to them. So my son asked the question about, you know, how does that work? And, and then my daughter simultaneously is like looking up what the national debt is on her phone and everything. And so I'm explaining this, this idea that like we're, you know, basically spending more than we make and like at some point you got to kind of pay that off. And then I actually started explaining the whole Doge initiative with Elon Musk and cutting, you know, spending and, and then the big beautiful bill shows up and throws it all into chaos and then he fights with Trump. So like all of this was like a seven minute conversation on the way to the school. So I'm laughing though, because there was this great tweet that I thought encapsulated this OpenAI Oracle deal. So well, it's from Yuchen Chin, who's a co founder and CTO at Hyperbolic Labs. So we'll put the link in the show notes. This is his tweet, how money works. 1. OpenAI signs $300 billion GPU deal with Oracle. Now keep in mind, OpenAI doesn't have $300 billion, so they sign this $300 billion with the overall. 2. Larry Ellison gains 100 billion net worth. No GPUs shipped. 3. Larry invests in Open Eyes $1 trillion round. Now this is hypothetical. They haven't had a trillion dollar round yet. But four, Sam uses 300 billion to pay Oracle. Five, Oracle stock pumps again. Six, Larry makes another 100 billion. Seven, Larry invests in OpenAI. So basically we're creating $300 billion out of thin air. And so that like I Had then explained to my kids, like, the idea of the stock market and how stocks, like, move. And I was giving these examples, like, say, you know, a hundred dollars, and then you announce this deal and now you have $130 per share, and then that 30 all. I'm actually kind of shocked we had this conversation now that I'm thinking about it. And they seem to actually understand it, which is the incredible part. But that's kind of what this feels like, is like this is literally manufacturing $300 billion out of nothing by signing a deal that pumps the Oracle stock, which then makes Larry Ellison richer, which allows him to then invest in the next round of funding for OpenAI, which then raises the valuation of OpenAI, which then gives them the money, which then they pay to Oracle, which raises their stock again, which they then invest in. It's hilarious. Like, this is how this stuff works. So how you create the money out of nowhere is. It's just part of capitalism, I guess.
- [46:18]
B
Yeah. And interestingly, and if they follow through on this, all of this made up imaginary money could result in huge amounts.
- [46:25]
A
Of real world physical infrastructure, actual infrastructure, like, oh man, how the world works is amazing. When you understand this is like how it all works. Like, you just look at the world differently.
- [46:38]
B
Yeah. I mean, it's easy to understand all this and then be like, yeah, everything's kind of made up, isn't it?
- [46:43]
A
None of it makes sense. That's how in the conversation. So I drop them off, we're like walking to the front of the school and Balin's like, so the money doesn't really exist, but it comes. I'm like, all right, buddy, just have a good day.
- [46:56]
B
Yeah, right.
- [46:57]
A
We'll talk later.
- [46:58]
B
I love that. All right, next up, Anthropic has agreed to pay at least $1.5 billion to settle a landmark copyright lawsuit, which is the first major AI class action case in US history. So we've talked about this a couple times on the podcast. The company allegedly downloaded over 7 million pirated books to train its Claude Chatbot authors argued this amounted to industrial scale copyright theft. Anthropic denies wrongdoing, but is going to end up paying around $3,000 per infringed work. Roughly 500,000 titles have been implicated so far. So a federal judge, when we talked about this last, had ruled that the training on copyright books was probably fair use by his interpretation. But the issue was they acquired all these books from piracy websites and libraries. So basically that was theft in the judge's estimation. However, now what's happening is, though they have settled, this Same federal judge, U.S. district Judge William Allsup, also criticized the settlement, saying he might actually reject it because he is questioning whether half a million pirated books is truly the final number and whether the claims process will fairly reach all the eligible authors. So he demanded basically a final list of works by today, when we're recording September 15th, and a reviewable claims process by the 22nd. These aren't resolved as issues. This could collapse and could go to trial or the bill could be even larger. So, Paul, when we last talked about this, the latest update we had had a few episodes ago was the settlement had been reached, but we didn't know how much it is. Now we see a massive 1.5 billion, but that number sounds like it could go up or this case could even go to trial. I don't know how likely that is, but the fact they're questioning this already seems tough.
- [48:48]
A
Yeah, the 3000 PER seemed really low because wasn't it like 150 PER is what they were potentially on the hook for? Yeah, yeah. So we talked about it at the time as like a, like a potential, like, you know, extinction risk event for anthropic if they had to pay 150,000 per stolen book. So the 3,000 just seemed like a slap on the wrist, especially given the numbers that they're currently raising at. So, yeah, I don't know. It'll be interesting to see where this goes. There was like two other, you know, cases related to this stuff that we didn't even put into the show last week and this week, and I noted for Mike, maybe next week we do a summary of some of these high profile cases. There are a bunch of these things going on. Like, this one's pretty far along, but I don't know, like, this whole space is crazy right now with like, what's going on. And. And now it seems like we're getting to the point where over the next six to 12 months, we may see some, like, landmark cases come through that start to play out where this goes. Now, the other variable we've talked about before is the federal government currently is not on the side of the copyright holders when it comes to this argument. And they want the AI labs just to take whatever they want basically to train and build these models. So there's always that risk that the government steps in and asserts some leverage here. So I don't know. It is interesting to follow. I kind of, like, I was laughing to myself when I was thinking about this one. It's like, all right, so they're going to, you know, let's say the fine goes up to 10 billion and you know, 10x that, okay, every chauffeur gets 30,000, whatever. All right, just go do a deal with Oracle and, and then their stock will go up and then they'll invest in your next round and they'll pay the fine for you. And like that's how this all works. Like I, I can't see anthropic going under as a result of whatever this number is at the moment, they appear to be able to raise whatever they want. And so if it looks like they're going to get hit with a 10 billion dollar fine, you just raise the extra 10 billion in the next round and you take care of it. Like, I, I don't. It is a pessimistic view I think of like how the legal system works. But I, that's kind of how I've always felt about this is what I've always said is they're going to pay fines and they stole the stuff. It is plain as day, they all did it. Meta, Google, OpenAI, Anthropic, every single one of them stole copyrighted material to train their models. We all know it, they all know it, the courts know. Eventually just gets solved somehow is what I truly believe happens though. And probably through a bunch of really large payments to creators.
- [51:18]
B
All right, next up, after a high profile recruiting spree, Meta has as of right now poached more than 50 AI researchers from rivals like OpenAI, Google App and X AI. Many of them were lured with huge pay packages and promises of abundant compute. But now some of those hires are already leaving, frustrated by status battles and internal politics. So this the company secretive TBD Lab, which they've created as part of their kind of Meta Super Intelligence Labs initiative is now working just steps from Mark Zuckerberg's desk and it's become a flashpoint because it requires special badge access, it's not listed on any internal org charts and it's seen as a really important part of their super intelligence ambition. So this is sparking resentment, according to some reports we're seeing especially in the Wall Street Journal. Like legacy employees are demanding raises or threatening to leave. Meta says that their moves here were already planned. It's not about, you know, poaching talent or anything or kind of comparing to legacy talent, but we are starting to see kind of here, Paul, these cracks where like all these all stars that they've recruited and paid a ton for, there are some ripple effects Here, not only are they not getting along with everybody, but now your legacy employees are starting to demand more.
- [52:36]
A
Yeah, I mean, maybe they didn't see this coming. I don't know, like, do this in any organization, take any department and say, like, Mike, say we did this to you. Like, all right, we're going to go get some new podcast hosts and. And some people who can create courses and, like, create content. And, Mike, you've been amazing, but we're going to pay them seven times more than you, and it's going to be in the media that they're getting seven times more than you. And they're going to get offices closest to mine. We're going to move you just. Just a little bit over there, maybe about 100 yards away so I don't have to see you. And, like, Mike's me. Like, yeah, that sounds great. Like, let's do that. I feel really good about this situation. Like, of course you're gonna have resentment and frustration, and the people who are there have been building the thing, and, like, they're gonna want more money and they're gonna want to be paid. Like, those people. I. I don't comprehend how this wouldn't have been foreseen, but I don't know, like, they didn't prepare for it properly. But, yeah, the fact people are already leaving and I don't know, like, we've talked about this openly on the podcast for. For the last couple months. Like, this just seems like a train wreck waiting to happen. Like, you don't put 10, 12, 50 people in a room who are all now paid ridiculous amounts of money, who are all the best of the best and think they're all going to get along and work really well together. And all the A players who are already there met a stack with talent. Like, they're just going to be like, all right, cool. Like, we're now not important. And like, I. I don't know. It. It's. It's pretty predictable. It's, I guess, an entertaining to watch to a degree, but, yeah, I don't know. See what comes of it.
- [54:21]
B
Yeah, I guess. I can't say I'm shocked. As. As amazingly competent as Mark Zuckerberg is, the fact he didn't account for personalities and feelings and interpersonal dynamics does not really surprise me.
- [54:33]
A
Maybe they don't care. I don't know. Yeah, I don't know. Yeah, it's just. Yeah, I mean, that was the Wall Street Journal article. Arco was like, this is Meta's facing a quintessential management problem, how to recruit and retain top talent while keeping remaining employees satisfied and maintaining harmony across the organization. Like, I don't know how you do that in that environment.
- [54:52]
B
All right, next up, the job market may currently look fine on paper, but for millions of workers, it's becoming a real problem. So according to the Atlantic, there's low unemployment and rising wages, which is great. But behind the scenes, hiring seems to have nearly frozen. They document how for job seekers, especially young ones, the experience trying to get a job is kind of hellish. Like, applicants are flooding the market with AI generated resumes. Employers are overwhelmed and using AI to read them. And as a result, nobody's getting hired. For instance, one graduate applied to 200 jobs, got zero responses. They have a bunch of anecdotes like that in their reporting. And it's kind of this vicious AI driven loop, it sounds like where job seekers are using things like ChatGPT to sound professional. So companies deploy AI filters and chatbots to weed out that noise. Now, the reason we mentioned this, and we've talked about that phenomenon a little bit in past episodes, is because on top of this, at the same time, we're getting potentially a new preliminary revision from the Bureau of Labor Statistics saying that job growth through March 2025 has been overestimated by about 911,000 positions. So if this goes through, this would mark the largest annual downward revision in US history. They're confirming if this will happen next February. Now, on top of all this, researcher Eric Brynjolfsson, we've talked about the past few weeks who studies AI's impact on labor, had this to say about this phenomenon. He said, quote, the big revisions in the jobs numbers of two important implications. One, the economy is in the midst of a bigger disruption than most people realize. Two, productivity is growing faster than most people thought because productivity equals GDP divided by hours worked. If there's a smaller denominator, it means productivity growth may be 0.5% faster than previously estimated. So, Paul, we don't know why the job numbers necessarily were actually so much lower in reality. But Brynjolfson's kind of proposition here saying that we're in the midst of a bigger disruption than many people realize and productivity is growing faster than most people thought. That's kind of almost exactly what we've been waiting or expecting to see show up in economic numbers as a result of AI's impact, isn't it?
- [57:09]
A
Yeah, it's definitely what you would expect if the AI was starting to have that impact. And to remove the politics from this Because I know this has become a hot button issue in politics. It is a broken system. Like these revisions are weird. Like if you don't follow along how this works, you can all of a sudden think that one side is maybe doing this to the other side. They're making, you know, jobs political so that we go into midterms next year. It, it creates some friction. That being said, it's how it's always worked. Like it wasn't like this system was invented three years ago and, and now it's like, you know, hurting the current administration or anything. This is the flawed system America has been using for a long time to, to do it. It's a hard thing to get accurate numbers. So it is a process. They go through these regular revisions. So cnn, there's an article we can link to that talks about these revisions coming. It says prior to Tuesday's release, economists predicted that a large downward revision was likely due to three primary factors weaker than inferred job creation at new firms. Which could be AI related. They're not saying it is, but that that could be sampling errors resulting from declining survey responses. So they, they depend on people who run companies to report this stuff. And if they're not getting the same reporting rates, then there's higher error rates basically in the data and to some extent adjustments for asylum seekers and other undocumented workers. So that could all play a role. So nowhere are you at least this article saying specifically AI. But that is certainly the undertone of all of this is that like maybe it is starting to have its impact. And then there's a Atlantic article that we'll link to that is the Job Market as Hell is the title. And I'll just read a couple of quick excerpts on this one and then we'll kind of move on to the next topic. So as right now millions of would be workers find themselves in a similar position. Corporate profits are strong, the jobless rate is 4.3% and wages are climbing in turn. But payrolls have been essentially frozen for the past four months. The hiring rate has declined to its lowest point since the jobless recovery following the Great Recession. In a recent survey, chief HR officers told the Boston Consulting Group that they are using AI to write job descriptions, assess candidates, schedule introductory meetings, and evaluate applications. In some cases, firms are using chatbots to interview candidates too. Prospective hires log into a zoom like system and field questions from an avatar going back to our thing. Their performance is taped and an algorithm searches for keywords and evaluates their tone. Okay, two other parts I'M going to read, but I'm going to zoom out for a second. If you haven't been in the job market or if you're not involved in the hiring process, you may have no idea that this is how HR is working right now. And that's why we thought it was an important thing to sort of bring to the forefront. Here is this is what's happening in the world. I have talked with people recently who've been looking for jobs and this is the kind of experience that they're seeing. So the Atlantic article continues. Still a lot of job applicants never end up in a human to human process. The impossibility of getting to the interview stage spurs jobless workers to submit more applications, which pushes them to rely on ChatGPT to build their resumes and respond to screening prompts. And the cycle continues. The surge in same same AI authored application prompts, employers to use robot filters to manage the workflow. Everyone ends up in a tinderized job searched hell, which I thought was a pretty funny way to describe it. And then the final four months, the economy has been in a low hire low fire equilibrium. Virtually every sector of the labor market, except for healthcare, has been frozen. The amount of time a worker has spent looking for a job has climbed to an average of 10 weeks, meaning that Americans are spending two weeks longer on the job market than they were a few years ago. The share of American workers quitting a job has fallen to its lowest level in a decade because of concerns about rising prices and jitters about slowing growth. So again, you everything may be gravy to you. Maybe you're working in the AI space and you're in high demand and you're, you know, you're seeing your salary going up and like things are good. And maybe you live in a bubble where things are really good, but when you get out of that bubble, things aren't so great. And it is really hard to find work, especially in that younger worker level. We talked a couple weeks ago about ages 22 to 25. Like unemployment rates are like what, 13 was it, Mike? And then underemployment, who knows? Like it's probably in the 20s. It's a very, very delicate market right now that could sway. And again, the politicians are very, very well aware of that. So anything you hear related to jobs, the economy, you have to understand where that information is coming from and what the intention of the distribution of that information is. Because we are entering midterms in the United States in like three months. Basically we're heading into that, that cycle, this is going to be a major, major topic.
- [62:12]
B
Somewhat related, we had a listener of the pod kind of ask a question to us on LinkedIn that we wanted to address.
- [62:20]
A
And this was public, by the way.
- [62:21]
B
This was public post.
- [62:22]
A
He tagged us in.
- [62:24]
B
So Matt Brooks, basically, after listening to us talk through some of this stuff in episode 166, kind of was asking, look, if companies replace too many human workers with AI, which some people are projecting, who is left to buy what they're selling? He said, what's interesting to me is that we're predicting tremendous value for companies as they adopt AI and reduce human workers. But if we replace humans with AI, who is going to buy the products they're trying to sell? I mean, if they're losing income or being, let's say, underemployed, even making less due to AI. How did. Paul, how are you looking at that question?
- [63:00]
A
Yeah, so I thought this was a really smart question, and I actually did comment on this one. So it's like sometimes I, I see these questions and I, I don't have the brain power available to, like, think deeply and provide a good response. So Matt's I thought was a great question and something that we're seeing a lot of. And I get this question in, like, private talks. So I'll go and do talks for executives and I will get these kinds of questions quite often. So I thought it was great that he was, you know, put it out there. So, again, like, for our listeners, if you post stuff on LinkedIn or even on. Actually, there's a pretty good chance, Mike and I do see it sometimes I don't have the time to, like, engage with it. But, like, we appreciate these kinds of commentary and the whole idea that this podcast is intended to help drive these kinds of questions throughout society, like, get people thinking about these topics and then asking the hard questions so people in their network start to think about these things. So, you know, kudos to Matt. It's a great question. So I'm just going to read my response because it's probably the best I can, I can do to respond to something I said. So you're asking one of the most important questions. The first challenge was that economists were largely in denial about the impending impact on jobs because the data didn't support it. Yet, as we've discussed in this one, I feel like that mindset is shifting. We are now seeing economists asking the hard question. The second challenge is that the AI labs and leaders building the tech that will cause the disruption didn't see it as their jobs to address the impact and help solve for it. It that was, in their view, the job of philosophers, sociologists and economists. They preferred to talk about abstract ideas like universal basic income, but with no actionable details or roadmaps. As we've talked about today. Mike, the People First Fund from OpenAI is a shift in the mindset here. So two years ago, you know, OpenAI is doing a study on universal Basic Income. That's going to be the answer. They realize, oh yeah, that's probably not going to be the answer. We can't just like assume that that's going to show up and solve everything. We need to invest heavily in AI literacy in the economy. Let's build a people AI first fund and let's start doing something. So again, we're now seeing the shifts. The third challenge was that the government leaders didn't care because AI wasn't moving the needle on votes. As the impact on jobs accelerates, politicians are realizing AI may become a central topic to upcoming election cycles. So, in short, we have no idea the answer to your question, Matt. But the people who need to care to create a sense of urgency seem to be coming around to the idea that it could be a near term reality they all have to solve for. So as, I mean, again, we didn't set up the podcast this way to build up to this topic. But as you will see, all three of those things are things we've already talked about on the podcast, that all three of those stakeholders are now realizing what's coming and they're all now trying to do something about it. Whether or not they're successful, I don't know. And I don't know what this looks like when people aren't making the money to buy the products and services, buy the goods so we can increase the gdp, but like, who's buying it if they're not? They don't have jobs? That is the great question. If, if unemployment overall were to reach 13%, go like what we're seeing at entry level workers. If, if, like across the economy of 13%, we got major, major problems. And that's the stuff that I think economists are starting to realize, which I've talked about this. I've had leading economists laugh at me. Like three years ago when I was talking to him about this and saying, hey, maybe you guys should be doing more to model this and prepare for this. They I literally had one leading economist tell me this isn't in his top 10 concerns, like 18 months ago. And I was like, I don't think you should be on a stage right now like this. Yeah.
- [66:37]
B
So, yeah, it feels like at the very least, we need more robust conversation around it. Because I'm just so tired of seeing people say, well, AI is going to create jobs or like, UBI or whatever. Like, okay, great, what's the next. What's the next sentence? What's the next question?
- [66:55]
A
That was the last. We talked about this on the AI Forward CEO memo. Was it last week on the podcast where it's like, that's fine, you can state that how like. And so what the labs have done is. I mean, I've literally listened to interviews with Demis, who I respect more than any of them, and Sam, who are like, that's not our job. Like, that's. That's what philosophers do. That's what sociologists do. That's what economists do. That's not us. We build the tech. And, like, they could sit there and, like, think about it, but that's not what they're going to work to think about. They're going to work to think about how to build the next frontier model. And so it's true, like, it's really not Demis's job to think about this, but hopefully the labs are going to continue to do more around this.
- [67:40]
B
Well, like you alluded to, they now have the incentive to. Because it may not be your job to think about it, but when there's real world societal impacts, your job is to figure out how not to get regulated out of existence because of the backlash around this. Right.
- [67:53]
A
Correct.
- [67:55]
B
All right, so another interesting topic this week, a startup called Inception Point AI is betting that its business model, which is flooding the Internet with AI hosted podcasts, more than 5,000 shows and over 3,000 new episodes a week will pay off. So they're using AI to produce podcasts at an extremely cheap rate. Each episode they do costs about a dollar to make, and the dollars make sense if 20 people listen, they said. According to an interview in Hollywood Reporter, the episode turns a profit thanks to the programmatic advertising attached to the episode. And that's basically their entire pitch. Ultra cheap, endlessly scalable, fully automated audio content. So, like I mentioned, they've done more than 5,000 different shows. They claim to be churning out more than 3,000 new episodes per week. Each of those episodes takes like an hour to create. The company's digital hosts are AI personalities, and they're assigned to shows ranging from really mundane weather updates and niche biographies, and then also more in depth, longer form shows. They actually use AI to Generate topics based on trending searches, build out the scripts and customize the voices. They also say they've racked up 10 million downloads across this network since 2023. They're experimenting with short form video and influencer style social media as well. And the CEO of this company said to Hollywood Reporter, quote, I think that people who are still referring to all AI generated content as AI slop are probably lazy Luddites because there's a lot of really good stuff out there. So Paul, I'd be love to get your take on this. I'll be honest, this sounds like a terrible idea to me. Like, I'm not offended at all. Like, I know that's where this is going as a podcast host. Like, I'm not even, I'm a realist. Like none of this surprises or offends me. I'm just more like, I already think there's like crappy podcasts hosted by humans out there and I don't even have enough time to listen to the good stuff stuff. So like, I don't need more noise.
- [69:51]
A
Yeah, I hate this. So contextually, you know, I was talking about my days with HubSpot. So back in, you know, I started my agency in 2005 and we were a content agency. Like, that's Mike and I started working together at that agency. Mike joined us in what was 20, 2012. Yeah, 12. Yeah. So you, you'll remember this era very well then. Mike. So Mike came in as a content specialist, like he was a writer by trade and then, you know, developed into, you know, a leading strategist and eventually like, you know, played a key role in building everything we're doing with AI, co authored books, all these things. But in 2011, 2012, when I was building what we were building with HubSpot, when we were scaling a content agency, content farms were the thing where like, let's say that, you know, if you never hired a freelance writer, you know, it could be, let's say a dollar a word, $2 a word. You're hiring them to like create the research report or write an article, write a video script. You would pay roughly on like a per word basis. Like that was kind of a common way to do it back then. So then these content farms show up in like, you know, late, you know, 2009, 2010, 11, and you could pay two pennies a word. And people did it. It was because there was SEO value. Like Google hadn't caught up to this slop that was being created yet. And so you were rewarded in organic search results for the crappy content you were putting out. And so we were talking with companies at that time who were like, hey, listen, I know paying you guys 5,000amonth would, would help, but I mean, I can get like 10x more crappy content from these people and they tell me it's going to help my SEO. So instead of the four posts with you guys that are like really high quality that humans create, we're going to have these other humans who are willing to do it for 2 cents a word do it. And we're going to create 40 crappy posts. And you're like, okay, well that's going to crumble eventually. Like, that's not a model that's sustainable. That is, you know, cutting corners. It might, might work for 6 months, 12 months, 8. I don't know. But usually if there's like an ickiness to the strategy, it probably eventually falls apart. And I think this is one of those you're looking at. It's like, God, this just feels gross. Like, yes, you could do this. You could create a podcast. You can put it on YouTube and Spotify. You can juice it with like, you know, $1,000 a month in YouTube ads, which you can then turn into $3,000 a month and ads like ad dollars in totally. A viable model doesn't mean it should happen. And I can almost guarantee you there's no AI verification process in this. You have these models creating this crap that like, is a script that the podcast AI, you know, hosts read. Nobody's ever verifying if any of it's actually true. It all sounds good and maybe 80% of it's viable. Like, I hate this inevitable. But it's just a crap business model. And I understand that people do things to make money and that's cool. Like, we live in a capitalistic society that allows that to happen. Doesn't mean we have to like, like the fact that they do it.
- [72:54]
B
Yeah. And I, you could interest me far more in a larger conversation about how AI generated content can be good. It can resonate, can be interesting. There's nothing saying that that's not what they're doing. But this business is going to zero. It's an arbit pool of like the moment the platforms wake up to it going to be like Google's Penguin update, like those content farms went to zero.
- [73:14]
A
Yeah, you're at number two and then all of a sudden you can't even be found in the search results.
- [73:17]
B
Yeah.
- [73:18]
A
And, but again, it takes the, the fortitude of the distribution channels to do something about it. So it takes YouTube being able to verify and say, okay, yeah, AI generated podcasts just are not going to get, you know, shown up in the search results or Google or, you know, Apple podcasts, Spotify. It's going to take them saying, oh, wait, this is ruining our platforms by having all this AI generic. And again, I don't know this CEO, I don't know this company. Maybe they do have some verification process, and this is actually like, really valuable stuff that people want. And maybe this company is legitimate, but it's just a blueprint for somebody else who isn't, who just wants to make money to show up and do it. So it's one of those where you look and say, okay, if that's where this is going to go, it's going to ruin the fun for everybody. It's what, it's what marketers do. Entrepreneurs sometimes do. It's like, like you get a good thing and then, like, people show up and like, oh, I can make a bunch of money on this, and then they ruin it. And I don't want to see podcast podcasting ruined.
- [74:12]
B
Right. Same. All right, next up, the FTC has launched a sweeping investigation into AI chatbots designed to act like AI companions, especially when used by kids and teens. So the Federal Trade Commission is demanding answers from seven major AI players, including Meta OpenAI and Character AI, about how AI is being used to form relationships with users. So obviously we've talked about before, bots from these companies can simulate emotions, intentions, and friendship, and sometimes convincingly enough that users trust them like real people. Though we've talked about people from all walks of life, not just teens, forming deep attachments to AI chatbot companions, whether that's platonic or even romantic. So the FTC wants to know what steps companies are taking to prevent harm, especially to kids. So, Paul, we've talked about this on a lot of episodes. I. I fear we will be talking about it a lot more. When I saw this, even though I thought, like, this is still very early and we'll see how it actually plays out, I did have this kind of initial visceral reaction saying, like, thank God someone with actual power is starting to take this seriously.
- [75:24]
A
Yeah. And again, this might be one of those, you live in a bubble where this isn't a thing. You know, kind of like the jobs thing. Like, maybe everything's cool. Maybe you. You don't personally use them as a companion, and maybe you don't even know people who do. But it is like, what are the top three most popular use cases for ChatGPT. Yeah, and I thought Ali K. Miller tweeted a good anecdote here. I'll just read her tweet because, you know, I think it's representative of kind of what's going on. Again, if you're not familiar with how this is working. So she said, weird story about AI companions and spouses. A friend was chatting with a woman, let's call her Steph at the gym about mental health. Steph has been married seven years and loves her husband, but has found a second companion in ChatGPT. She would have long live voice chats with a male voice in chat GPT. First about basics like her workouts, then eventually about her whole life, including mental health support and discussing her husband. Her husband finds out she's doing this. They argue about whether she should continue and he demands Steph switches the voice to a female voice. Steph understands his points of point of view and switches the voice, but feels like she lost her friend. We will hear a lot about parent children dynamics as it relates to AI. We're going to hear a lot more from life and romantic partners. So I shared that and I said reality is increasingly feeling surreal.
- [76:42]
B
Yeah.
- [76:43]
A
So again, just if you're not aware of what's going on, if you have teenage kids, this, this is a reality, like the next generation, it's going to be normal to have these kinds of relationships with an AI. And again, like, there's no right or wrong here. We're not, you know, judging this person for their relationship with it. We are just presenting this as like, this is where society is going and you need to be aware of that. For many reasons, depending on where you are in your life and where your family and friends are, things that sort of seem like they're straight out of a movie are going to be part of your daily life.
- [77:23]
B
Yeah, no kidding. All right, next up, AI is transforming retail. And we've recently come across some fascinating case studies in this industry. So we wanted to quickly share those just because they're great examples of kind of what's possible here. These come from a recent report in Fortune. So three different companies using AI. First is Walmart. They've rolled out real time AI systems across the U.S. canada, Mexico and Costa Rica. These tools now spot consumer trends as they emerge, forecast demand and shift inventory before products run low. One standout system is called Trend to Product. It tracks signals from social media and search data, turns them into mood boards, and feeds those ideas directly into product development. Amazon has also deployed new agentic AI systems that can forecast demand, map global logistics, and coordinate robotics. They unveiled a tool called Wellspring, which is a generative AI that maps logistics networks. And an AI forecasting engine was also debuted that helps balance global inventory. Finally, the grocery chain Albertsons, which has over 2200 stores, built predictive models that estimate how many shipments are arriving each day and then sync staffing levels to meet them. So the result is less overstaffing, fewer delays. They stock shelves 15% faster during peak seasons. They've also started using AI to scan messy supplier emails and PDFs, extracting delivery changes or risk factors that might otherwise slip through the cracks. So, Paul, these are the exact types of case studies and leaders we've been featuring in our AI for Industries course series and other Academy content through AI Academy. Great to see more of these being published and publicized, in my opinion. Did anything in here jump out at you in particular about how retail is using AI?
- [79:08]
A
I just love to see the tangible examples. You know, the. The whole premise behind the name Smarter X is you can build a smarter version of any business. So X is the variable. Like, it was funny, my son was actually asking me, what does the X stand for? Smarter X. And the X is the variable. You can build a smarter organization, you can build a smarter marketing department, sales department, you can build a smarter version of your career. Like, that was the whole premise behind the name years ago when I created it. And so I. I love to see these transformations occurring. And we get asked all the time, Mike, for those examples of companies that are. Well, this is actually the kind of thing that's the inspiration for AI Transformation course series we're going to be launching as part of AI Academy Live. And then we're actually planning on doing an AI transformation series for the podcast, where we go interview the leaders from brands that are doing this kind of transformation. So we're going to make an effort to do a lot more of this kind of applied AI on the podcast and through our courses, so people can get inspired by examples they're seeing of other organizations and realize the almost infinite amount of use cases there are for AI in business.
- [80:08]
B
Yeah, no kidding. All right, to wrap up here, Paul, we've got a few quick AI product and funding updates I'm going to run through and then kind of bring us home here.
- [80:16]
A
Sounds good.
- [80:16]
B
So, first up, Cognition, the startup Behind Devin, the AI software engineer, just raised $400 million at a $10.2 billion valuation. This company is barely a year old. In a year, They've grown from 1 million to 73 million in annual recurring revenue. They also acquired Windsurf, which is an AI powered development platform, which doubled ARR again. Perplexity, the AI powered search startup, has locked in another 200 million in funding, this time at a $20 billion valuation. This is just weeks after its last raise at 18 billion. They pulled in about a billion total in funding every few months so far. 11 labs, which we have talked about in the past, is letting employees sell shares at a $6.6 billion valuation, again double what it was just months ago. The $100 million tender offer is led by Sequoia and gives longtime staff a chance to cash out without waiting for an IPO in some kind of product Lawsuit Related News A couple things that we'll be tracking like we talked about on future episodes, Perplexity is facing a lawsuit from Encyclopedia Britannica and Merriam Webster. They claim Perplexity has scraped their websites, plagiarized their definitions, and even misused their trademarks. Midjourney, one of the leading AI image generators is facing a serious legal challenge from Warner Brothers Discovery, which is suing the company for copyright infringement specifically for letting users generate images of its iconic characters. It is not alone. We talked about previously how Disney and Universal filed similar suits earlier this year. This next one might be something we end up revisiting. We'll kind of see how it plays out. A new wearable called Alter Ego just dropped and it's being described as near telepathic. So this is born out of research at mit and what it does is it picks up the silent signals your brain sends to your speech system before you even say a word. So it's not really your thoughts exactly, but like what you intend to speak. Which means using this device you can type, search or interact with apps using nothing but silent intent. The breakthrough that they're kind of touting is something called Silent Sense that captures everything from mouthing words to pure motionless intention. So it's like like AI as a mind extension that could eventually let you have a full conversation without making a sound. And then finally NotebookLM. Google has given its AI research assistant NotebookLM another upgrade, turning it into a full on study partner. It now generates flashcards and quizzes from your notes, lecture slides, or research papers. You can set the difficulty, share sets with friends and even ask follow up questions to understand what you missed and why. It's also helping you create smarter reports. You can upload a research article and it might suggest a blog post, glossary or even a character analysis tailored to the content, not just the format. They've also got a new learning guide feature that pushes deeper understanding, asking open ended questions instead of just spitting out answers.
- [83:13]
A
Mike, I'm going to add a request for our Academy team, the Gen AI app reviews. So we do these like every week. I know we've got a full calendar coming up. I would love to see app reviews of the learning guide and the guided learning that's available in Gemini because here's my use case. I've been using the guided learning to help me help my kids and I want to actually take it to their school and say, listen, you all should be integrating this, but I don't have the time right now to build the deck to pitch them on. Hey, you should actually be very proactively integrating this because you have students who could benefit from this. It's just different than a standard AI assistant. Like it's a. But the perception at schools is that AI is AI. Yeah, this is different. And so I would, I would love to see more schools very aggressively exploring Learned the Learning Guide within Notebook and Guided Learning within Gemini and ChatGPT because my personal experience has been it is a game changer for working with your kids through problem solving, not giving them answers, but actually showing them how to solve something so they learn in the process. So these are, they seem like really small announcements, like really small features, but when you understand the implications to them, they're potentially like massive transformations of the educational system. And I don't think it's being talked about enough.
- [84:40]
B
Yeah, for sure. We'll get those on the docket. I'm eager to explore.
- [84:44]
A
Put in a request.
- [84:45]
B
Yeah. All right, Paul. Well, thanks for breaking down another action packed week in AI for us.
- [84:51]
A
Good stuff as always. And we will be back regular time next week. Everyone have a good week. 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 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.