← The Artificial Intelligence Show
The Artificial Intelligence Show
#160: OpenAI Hits $12 Billion in Revenue, ChatGPT Study Mode, More AI Job Losses, AI Is Coming for Consultants, Big Tech Earnings & Gemini 2.5 Deep Think
01:19:38
Loading summary
Transcript64 lines
- [00:00]
Mike Kaput
It's a really, really good time to be an expert who has a lot of real world background and context and I don't know how long that'll last. If you're in knowledge work and you are an expert that has all this domain expertise and background, don't waste this moment.
- [00:15]
Paul Raitzer
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 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 160 of the Artificial Intelligence Show. I'm your host Paul Raitzer along with my co host Mike Kaput. We are recording Monday, August 4th at 9:20am and we may be in the week of GPT5, so timestamping might be highly relevant this week. I'll keep an eye on Twitter while we're on here, Mike, and see if anything drops while we're doing this. All right. This episode is brought to us by AI Academy by SmartRx. The 3.0 version of Academy is launching on August 19th. So we have been working on this for, well, I've been kind of conceiving of this for a couple years, but intensely working on this since November 2024. So we have completely reimagined our AI Academy and our AI mastery membership program. They were first launched in 2020 and it's been a part of what we offer, you know, online courses, professional certifications, but this is on a whole nother level. So on August 19th at 12:00pm Eastern Time, we will hold a launch event that's going to share the vision and roadmap for where we are and where we're going. Preview all the new on demand courses and professional certificates. Introduce AI Academy Live, which is a new component. We'll give a preview of the new learning management system. It's an AI powered LMS that's going to be coming out later this year. Talk about personalized AI learning journeys and how you can build yours within, you know, the online education space. Take a look at new business accounts. This is something that we're introducing that's new. Five or more licenses can be part of our business accounts. So we'll preview all that and then we're Going to have an Ask Me Anything session with me and Mike and Kathy on our team. And so Mike and I thought we would be talking about this for the last couple episodes. That'd give you a little bit of a preview of what we've been doing because this has been the better part of my professional life for the last three months in particular building all this new content. And so the three main things that I've been creating for this August 19th launch are our AI foundations category. So AI fundamentals, which is a brand new series. This is an eight course on demand program. So there's Intro to AI AI Concepts 101, which is a brand new course I'm really excited about. It's actually one of my favorite ones. We built state of AI that goes to the five things everybody needs to know, the AI timeline, which takes a look at sort of AGI and beyond generative AI 101 prompting 101, which is actually a fun one to build. That was. I found it super helpful myself to go through that one, AI agents 101 and then AI and you, which is sort of like a personal look. Then I just yesterday finished the new Piloting AI. This is actually this was our flagship course series a few years back. So this has been completely reimagined. So Piloting AI third edition. And so that one's four courses. It's Piloting AI in Business, the Use Case model, the Problem based model, and how to build your CO X which is all about building AI assistance. And that one was again like kind of one of my favorite ones to build. I think it's super actionable for people. And then I'm finalizing this week the second edition of our Scaling AI series. So this was first launched in June, July 2024. So I'm doing a refresh of that series. That's probably the most evergreen of the courses we've created, but I'm going to do a refresh of those this week. So that one's eight courses. You have AI forward organization, the AI gaps, which is a new one, the AI Academy, the AI Council, generative AI policies responsible, AI principles, AI impact assessments and the AI roadmap. So in total, I've recreated or updated 20 courses for this August 19 launch. But that is just the beginning. Mike, what have you been working on the last couple months?
- [04:45]
Mike Kaput
Yeah, so Paul, I've been working on the first couple installments of our AI for Industries and AI department for Departments course series that we're putting out. So these are just the tip of the iceberg. But on launch date we'll have AI in professional services, which is, which is a four course series actually that is going to go through kind of your opportunity in AI as a professional services professional or a leader or an owner of a firm. We're going to talk about. Course two is AI and the future of the professional services firm. The third course is all about finding kind of your AI advantage in pro services. So really finding your own specific use cases in that industry. And then finally we are going to go through a ton of sample use cases and tools in the fourth course of that series. And then kind of a similar cadence. The AI in marketing series is going to go through kind of the high level opportunity marketers have to increase productivity and performance with AI in course one. Course two we do this great deep dive into the state of AI for marketing. So even if you're kind of newer to this topic, you'll come away with like a really good grounding in the actual state of things in our industry. Course three is about the AI forward marketer and really how you can invent your reinvent your career and your work very practically intangibly using AI. Then we also in course four go through a ton of AI tools and use cases specifically for marketers. And last but not least, in course five we do a whole applied AI for marketer section where we just kind of put you in the deep end. Not really. We give you plenty of instruction but you get started with some of the top tools out there. We just go through sample prompts and projects where you can just go kind of from zero to 60 very quickly with AI.
- [06:31]
Paul Raitzer
And then you've got the Genai app series which is brand new to AI Academy as well, which is going to be weekly product reviews like 15, 20 minutes each. We're going to drop those every Friday. And I think you've got a few queued up.
- [06:41]
Mike Kaput
Yeah, so we've got a couple queued up. You know, as of right now, the plan, because we, you know, these are meant to be almost in real time being created and published. So we are going to be creating one on custom GPTs. You know, with the caveat for our valued listeners if GPT5 comes out and blows up custom GPTs.
- [07:00]
Paul Raitzer
Got it.
- [07:01]
Mike Kaput
Which I really hope they they don't because that would really ruin my week. But we might swap in one of the other tools. But we're going to be doing things don't, you know, quote me on this order. But like NotebookLM from Google, Google Deep Research, OpenAI Deep Research some of these really core capabilities that business professionals in any function can get a ton of value out of. That's gonna be our focus for the first few of these.
- [07:24]
Paul Raitzer
Yeah. So we just wanted to give you a little bit more perspective since we've been talking a lot about this. And as Mike said, this is sort of the tip of the iceberg. There's a ton more planned. There's a whole AI for Department series. There's a who AI for Industries series, AI For Businesses, AI for Careers. So the whole idea is to. We've reimagined AI Academy to allow people to build these personalized learning journeys that really move them to, to a point of like, mastery over this topic. So, yeah, so appreciate, you know, everybody giving us a few minutes here up front to talk a little bit more about this. It has been a massive lift, but, you know, I think at the end of the day it's going to be incredibly valuable to people. And we're really excited to get these in everyone's hands come August 19th. Okay. So you can learn more about that at SmartRx AI. We will also put the link to the webinar, the August 19 launch webinar in the show notes. So that's what's coming with AI Academy. And then the second is brought to us by our Macon Marketing AI conference coming up October 14th to the 16th. As I said last week, we're trending in a really strong direction from a ticket sales perspective. We're expecting very strong attendance. We had 1100 last year. It's far outpacing that 1100 number this year. So we'd love to see you in Cleveland August or I'm sorry, not August. October 14th to the 16th. That's at the convention center in Cleveland, right across from the Rock and Roll hall of Fame and Lake Erie and the Cleveland Brown Stadium, at least for a couple more years before they move out. But come join us. You can check out the agenda. It's Macon AI Maicon AI. You can use Pod 100 for $100 off of your ticket. Most of the agenda is live. I'm. We're going to have some announcements coming up soon on the general session. The kind of featured talks and keynotes. We. Maybe next week we'll see. We may announce a few of them next week. But that's, that's taken shape and the main stage is really all about like the macro level. So I, I like to do talks on like education and the economy and the future of jobs, like bigger picture things. And they all obviously have relevance to marketing. But I really like to use that main stage to sort of expand people's minds and introduce new topics and speakers that maybe they wouldn't otherwise see at events like this. So love to have you in Cleveland, Macon. AI Again, it's M A I C O N AI. All right, Mike, we're going to. It was a busy week, like lots of big topics, and so we decided on Friday we're going to go with the rapid fire style again. There's a couple of these I made expand out a little bit past rapid fire, but idea is to try and go rapid fire on these because there's a bunch going on last week.
- [09:56]
Mike Kaput
Well, you know, as I want to start saying, you know, any rapid fire could be a main topic if you try hard enough, Paul.
- [10:03]
Paul Raitzer
So.
- [10:04]
Mike Kaput
So shoot for the stars over here. But yes, we're going to go all rapid fire. I kind of see this as almost like the qualm before the GPT5 storm, since I imagine that'll be a leading topic.
- [10:14]
Paul Raitzer
And I think Google's holding their next model. Like, I'm starting to get the sense that it's like, who's going to release first thing right now with OpenAI and Google, there's like a little game of chicken going on.
- [10:26]
Mike Kaput
Okay, so first up, Open Air just crossed a huge growth milestone. They are now tracking somewhere around 12 billion according to the information in annualized revenue, or 13 billion according to the New York Times. Slight difference in those numbers, but this is nearly triple their pace of growth from the start of the year and breaks down to about a billion dollars in annualized revenue happening per month at the moment. And this comes with some equally aggressive funding. They have already raised $8.3 billion as part of a $40 billion round, five times oversubscribed. And they've got now some institutional heavyweights like Blackstone and T. Rowe Price jumping into this round. And the single largest check as part of this funding is $2.8 billion from Dragoneer Investment Group. And this is one of the biggest VC bets basically in history. Behind this are some pretty serious user base numbers. The user base has grown to about 700 million weekly active users, plus 5 million paying business customers. That was 3 million just a couple weeks ago when we had this. Also as a topic, the to keep up with all this, OpenAI has upped its projected cash burn to $8 billion this year. That includes massive spending on chips and new data centers, some in partnership with SoftBank. There's also a strategic shift happening which we've Talked about a little bit. It's not just a consumer chatbot anymore. You can see ChatGPT kind of evolving into a productivity suite that's starting to directly target some of the wheelhouses of Google and Microsoft. So, Paul, I'm never surprised that OpenAI is thriving. I guess I am surprised that they keep hitting these crazier and crazier speed and scale of growth numbers. Like what is driving just this massive jump in revenue which is tripling its pace. And the cash burn is up to $8 billion from, I think they projected like 1 billion. So what is going on here?
- [12:31]
Paul Raitzer
Yeah, the, the cash burn I would expect is just going to keep going up as long as the demand, you know, long term is there. And that's largely going to be coming from, you know, what they're building out, the future infrastructure. Plus, it's just the demand on, you know, the cost of delivering this intelligence. So as more individual users, you know, want it, that costs money to serve up that intelligence in every chat that happens, especially as you get into video and image and reasoning, which draws on more compute than a standard text chat. And then the revenue is coming obviously from the business user side. So to go from 3 million in June to 5 million, here we are in August, like, those are crazy numbers. And that's just probably the surface of where they're going in terms of enterprise adoption. And so that's where the revenue is going to come from. But, you know, I don't think that they expect to be profitable anytime soon or this decade. Like, they're not. That's not the goal right now. It's to stay ahead of the cash burn and sort of hit escape velocity when it comes to especially the business user side of things. So, you know, the other thing is GPT5 does appear to be imminent. We don't know, like, there's been rumors that actually might come out today, August 4th. There's been other things I've seen online that say, you know, it could be later this week, but it does seem as though like we're entering this phase. Sam Altman tweeted a couple days ago, we have a ton of stuff to launch over the next couple of months. New models, products, features and more. Please bear with us through some probable hiccups and capacity crunches. Although it may be slightly choppy, we think you'll really love what we've created for you. So lots more to come. And then Sam is just like flaunting the fact that he has GPT5. Like, he's not hiding it. At all. He talked about on the podcast with the Theo Vaughn podcast, I think last week we mentioned. Then he did a tweet, I think this is on Saturday or Sunday. Said, pantheon is such a good show. A user replied, Did GPT5 recommend this? And Sam says, turns out yes, with A screenshot of GPT5 as the model chosen. Now, this was a really interesting tweet because I didn't know what Pantheon was and I thought he was just doing like a cutesy thing, like they'd named their next model Pantheon and he was just like, you know, doing what Sam does and having some fun with it. So then I went and did like, well, what is Pantheon? And I don't know if you're familiar with this, Mike, but it's a new Netflix show. So Pantheon is an American. This is straight from Wikipedia, by the way, because the description on Netflix was like 10 words. So. Pantheon is an American adult animated science fiction drama television series based on a series of short stories by Ken Liu. Set in a world where mind uploading technology is on the verge of mass adoption, it follows a disparate trio of protagonists. Maddie Kim, a grieving teenager whose father was uploaded without her knowledge. Caspian Keys, a gifted teen unknowingly raised in a constructed environment, and Vinod Shah Nada, a brilliant computer engineer uploaded against his will. As they place themselves at the center of a global conspiracy, they also deal with societal consequences and existential crises brought forth by rapidly evolving technology. The series has received praise from critics, particularly for its animation, voice acting, emotional and philosophical death, and portrayal of the Singularity.
- [15:47]
Mike Kaput
I'm in.
- [15:48]
Paul Raitzer
I watched. Dude, I watched the first. It's like a minute and a half, two minute trailer on Netflix. Oh my God. Like dystopian probably, but, like, just chills. Like, I was like, oh, no. Like, I don't. I don't know if I'm ready to watch this. So Sam doesn't tweet stuff like this by accident. This is like, you know, Preludes to. He's obviously become obsessed with the Singularity and super intelligence. But yeah, so apparently it was a 2022, like, AMC show that got dropped and then it was picked up by Amazon Prime Video and then dropped from there. And now it's like, got New Life on Netflix. So I will. I will be watching. I will tune in, but it's, you know, I don't think it's a coincidence that he's sharing tweets about things related to the uploading of intelligence and Singularity and stuff.
- [16:38]
Mike Kaput
Yeah, it sounds like I'M going to watch this. Love it, and then immediately lose sleep over it.
- [16:43]
Paul Raitzer
No, I might sleep over the trailer. You'll go watch it. It's eerie. Like the trailer itself is like, whoa.
- [16:53]
Mike Kaput
All right, next up, some more OpenAI related news. Microsoft and OpenAI we've been talking about are deep in talks to rewrite the terms of their relationship. And a lot of it comes down to AGI. Because right now, Microsoft's $13.75 billion deal with OpenAI gives it access to OpenAI's models until 2030, or until OpenAI declares it has reached AGI, artificial general intelligence. This milestone is very vaguely defined as maybe AI that outperforms humans at most economically valuable work. Now, if that happens, if there is some agreement that AGI has been reached, Microsoft could lose access to OpenAI's technology. Now this becomes a problem for Microsoft because they've built Copilot, Azure, GitHub, and much of their AI strategy around OpenAI's models. So currently the two are working on a new deal, one that would let Microsoft keep using the tech even after AGI is declared, while also negotiating a potential equity stake in the reportedly low to mid 30% range. However, there is plenty of friction and details to work out here. OpenAI wants more revenue, looser constraints on who it can sell to, and stricter guardrails on how Microsoft deploys its models. Microsoft, meanwhile, has blocked some of OpenAI's acquisitions and may not be afraid to walk away if the terms don't work. So, Paul, I mean, we've been following this back and forth for some while now. Like, seems like there's plenty to work out, but it does actually seem, based on this newer information, there's some movement happening. So do you think they're going to work this out in any timely fashion?
- [18:37]
Paul Raitzer
I. I don't know. I mean, I. Without being in the room, obviously, and only being able to read whatever, you know, the information or Bloomberg, Bloomberg gets access to you, you have to assume they eventually just find a way to get this deal done. Like both sides need this deal. But there's so many complexities here. There's the definitions of AGI and who gets to decide when it's been reached. There's Microsoft's access to OpenAI's models. There's the fact that OpenAI's compute needs are what's driving largely Azure's rocket growth, like the growth of the cloud computing business for Microsoft. Billions of dollars per quarter probably being spent with Microsoft. There's Microsoft's own AI ambitions under Satya and Mustafa Solomon, who's the CEO of Microsoft AI, former Google DeepMind co founder and Inflection AI founder competition for business customers. They're increasingly coming up against each other and selling against each other. OpenAI's desire for more compute beyond what Microsoft can or is willing to provide, which means they're having to go to people like Google and Oracle. OpenAI's IPO desires their need to change their business structure. OpenAI's funding, which is dependent upon them changing their business structure. So, like, there's all these things that we've talked about over the last year and a half on the podcast. It's not an easy deal to get done. There's lots of variables here. And then the other thing to throw into the mix here, Mike, is I was listening to this podcast last week. I'd never actually listened to this podcast before. It's Lenny's podcast. It's called Lenny Ratisky. I don't know if you've ever listened to this one before. Yeah, but he had Benjamin Mann on, who's a CO founder and AI safety researcher, Anthropic. He started Google DeepMind. But he was talking about the economic Turing Test. And I think he might be the guy who kind of coined this, like the way I've seen it now talked about on a couple. He was a no priors podcast. He talked about this as well. And I don't know that he gave attribution to someone else. So if someone else came up with this concept, you know, we'll mention them in a future episode, but he's the guy talking about it right now. So his idea of the economic Turing Test is sort of like everyone's trying to figure out, how do we define AGI? And they're basically trying to move the goalpost and say, well, let's just like, not even try and do that. Let's just set an economic Turing test. So the Turing test being, does the human know if they're interacting with the machine or not? Going back like 70 years, and that's sort of been achieved. Like, we've passed that. So what they're saying, the economic Turing Test is this is how he described it. He said it's this idea that if you contract an AI agent for a month or three months on a particular job, if you decide to hire that agent and it turns out to be a machine rather than a person, then it's passed the economic Turing test for that role. So what he's saying is like, a human hires someone virtually. They don't they don't know if it's a human or a machine. That human or machine does work for a month or three months on a particular job. And then the human, the person doing the hiring has to decide, am I going to hire, you know, professional A, professional B. And then unbeknownst to them, they choose to hire the agent over the human. And so that's what they're saying the economic Turing test is, is where a human doesn't know that it's an agent doing the work. And he said, and if the agent can pass the economic Turing test for like 50% of money weighted jobs, then we have transformative AI. So he's saying as an economy, once we get to the point where a human in a blind taste test basically chooses the agent over the the human worker 50% or more of the time for like 50% of jobs, then we have entered this age where it is inescapable. Like we are in true transformation of the economy, 10% plus GDP growth every year. And he talks about this as 2027, 2028 and anthropic tends to usually be more conservative than others. So you know how all that kind of stuff factors in. Whether this is part of the negotiations with Microsoft and OpenAI, I don't know. But it seems like this sort of more concrete quantitative test to say, hey, we're past AGI. This is what's missing from the AGI conversation basically. So it's just interesting to note and it's a really, really good episode. Like I will put the link in the show notes. It's one of those I'll probably listen to multiple times and take notes on next time.
- [22:53]
Mike Kaput
Well, as a funny corollary to that, that's probably happening in some way with humans getting jobs they're not remotely qualified for because they're using AI to gain to game hiring. Like we've talked for sure.
- [23:05]
Paul Raitzer
And probably like Fiverr and sites like that where you have humans that are applying for jobs, getting jobs and then they're agent do the work.
- [23:12]
Mike Kaput
Exactly.
- [23:13]
Paul Raitzer
And like the person hiring them has no idea that that's what's happening. I, I can almost guarantee you that's happening a ton. And people are making a lot of money having agents doing most of the work.
- [23:23]
Mike Kaput
All right, next up, OpenAI has just rolled out a new study mode in ChatGPT. So instead of just solving a problem or answering your question, study mode will walk users through concepts step by step, using prompts, hints and checks for understanding to guide the process. So this is meant to help students specifically actually learn instead of just getting a quick answer to an assignment or a question. And this actually leans on insights from teachers and learning scientists. It uses techniques like Socratic questioning, cognitive scaffolding and metacognitive prompts and these are all tailored to the student's level and memory from past chats. It even includes quizzes and open ended questions to help ideas stick in. Some early feedback reported by OpenAI students described it as quote, a live 247 all knowing office hours and some praised it for breaking down tough topics into something they could finally grasp. Study Mode is free for all logged in users. It can be toggled on and off during chats and for now it's basically just powered by custom instructions layered on top of ChatGPT. But OpenAI says the long term plan is to bake these types of behaviors directly into its models. So Paul, I thought this was really great to see. I mean just based on my own usage and how I get value out of AI models. I've been so bullish on AI for real personalized one to one learning, like the amount of things I've just been able to learn, even hacking together my own prompts has been incredible.
- [24:55]
Paul Raitzer
Yeah, this is a cool advancement. The post they put up did say that it was created with college students in mind. Yeah, be interested to see if college students actually use it in this way. I know some college students who, you know, don't really want it to function as an advisor, they just want it to do the work. So they did say in their post also was the first step in a longer journey to improving learning in ChatGPT. It is currently powered by custom system instructions, meaning they didn't change the model at all. Like the underlying model is still the same, they're just giving it specific instructions. But they said that, you know, if this works and once they make improvements, they plan on training this behavior directly into their main models. They're also exploring functionality to make study mode more engaging and helpful, such as clear visualizations for complex or text heavy concepts, goal setting for progress, tracking across conversations, and deeper personalization tailored to each student's skill level. So this is great. And I, I do think that there's, you know, two, three years out this is just how you learn. Like I think it'll be adopted pretty quickly. Interestingly Mike, I was actually working on and you were talking, you and I were talking about this before we got on today. So my daughter's 13 and she's a very gifted artist, like you know, illustration and painting and things like that. She gets that from my wife, but she's taken a keen interest in creative writing and so I've been working with her on kind of teaching her how to use chat GPT, but for this exact way. It's like, hey, don't have it right for you. Like talk to it. Like, hey, I want you to help me. I want to give you my writing. I want you to like tell me how to improve it. Don't just rewrite it, like, tell me what you're doing. And so I've been trying to figure out a way to like do this without her unintentionally using it as a crutch to learn the process herself. And so ironically I was yesterday actually going through, it's like, how do I build like a GPT for her that functions in this way that doesn't do the work for her, but actually helps her be a creative writer? Because I've, I mean I've written three books, I write for a living, but I'm not a, I don't write fiction. Like I don't really know the process of writing great fiction. So I, I went into study mode this morning and I said I'd like to create a GPT using study mode to help my teenage daughter develop her writing skills. How do I get started? And it said, hey, this is a great idea. Study mode can make learning fun, personalized and effective. And it said, you go through and define the role and tone of the GPT. You know, guide, not give answers. Use questions and prompts and feedback to be build skills, match your daughter's age and personality. And then it gave me a sample. It's like you're a creative writing coach for a teenage girl. Your tone is warm, encouraging and curious. You follow study mode rules, meaning you guide, ask questions and help her build her writing voice and storytelling skills through prompts, feedback and gentle challenges. You never write for her. You help her express her ideas more clearly. So this is, this is interesting. Like this is the direction. I'm really excited. But then it actually said like in when you're building your GPT in the capabilities section, choose study mode and then this will force the GPT to do it. And I was like, oh my God, did they build this into GPTs? And so I went and checked. That is not actually possible, but that is like if they go there, which I assume they have to, like, this is a no brainer to make this work. So right now if you go into ChatGPT, you pick study mode in a normal conversation but you can't set that as the default. But if you can build GPTs where you can choose Study Mode as a capability, now I can build that creative writing assistant for my daughter and know it's going to follow Study Mode. Or teachers can build GPTs for their class and set Study Mode as the default capability. So I assume OpenAI is going to do that and it's on the roadmap because that it seems like an obvious thing and I think this is great. Like, I would, I would build GPTs for my kids all day long if I could set them in Study Mode.
- [28:42]
Mike Kaput
Yeah, for sure. It seems definitely something to keep an eye on. Next up, according to some new reporting in the Wall Street Journal and Gizmodo, we are seeing the trend further cemented that we've been tracking for a while now. They're Talking about how CEOs are not only saying the quiet part out loud when it comes to AI's impact on jobs, but in some cases they increasingly seem to be referencing cutting headcount. Thanks to AI efficiency gains as basically just a badge of honor, executives are openly celebrating smaller workforces as signs of efficiency, technological progress, and investor discipline. For instance, they mentioned the example of Wells Fargo, whose CEO told investors the bank has shrunk for 20 straight quarters, down 23% since 2019. Verizon says its workforce is going down all the time. And bank of America, which once had 300,000 employees, is now sitting closer to 212,000. It sounds like from some of these examples and people they talk to, these aren't mass layoffs in the traditional sense, but a lot of time companies are letting attrition do the work. They're leaving roles unfilled, combining responsibilities or automating tasks. And AI is often the rationale here. So at bank of America, AI now reconciles trades, summarizes client info, and writes code, which reduces the need for new hires. And one AI consultant who spoke to Gizmodo just kind of had a really money quote here. Whether you love it or hate it, and it says he said, quote, CEOs are extremely excited about the opportunities that AI brings. As a CEO myself, I can tell you I'm extremely excited about it. I've laid off employees myself because of AI. AI doesn't go on strike. It doesn't ask for a pay raise. These things that you don't have to deal with as a CEO. Now, Paul, this kind of reminded me a little bit of maybe the darker side of AI adoption that you've mentioned in the past. You know, companies may get only focused on short sighted, near term efficiency gains and just like go use AI to chop headcount as much as possible and not really consider the full consequences of something like that. Is that what's happening here?
- [30:53]
Paul Raitzer
Yeah, I mean I, I think there'll be times where there's some pullback and they realize they probably went too far. But it's, it's just amazing to me how quickly this has gone from no one talking about this to everyone accepting this as the norm. I mean, I, I, I don't know what episode it was, but I remember vividly like saying, I don't get why this isn't a conversation. Like this is the inevitable outcome. And it was probably like early this year, like 2025 or end of 2024. Yeah. Where we're just like pleading with people to like accept that this is going to happen. Why isn't anyone saying anything? And because we were hearing it in conversations with executives. So it was, it was always inevitable that as C suites, boards and investors increased their awareness and understanding of AI, that we would have layoffs as a result of it or reductions in hiring. The thing I would caution though is we're still very early, but the financial pressures for publicly traded companies, VC backed companies, private equity owned companies to reduce human labor costs is going to grow. Like AI agents aren't even reliable yet. And we're already seeing CEOs straight up saying, yeah, we're going to cut workforce 10% because of AI. It's like it's not even that good yet. So as it becomes more reliable in the next 6, 12, 18 months, go to that economic Turing test idea as we, we enter these phases where it's actually reliable, like this isn't going to slow down. Like we're going to see continued disruption of this. And I think we're going to go through a very challenging period for jobs and an extended transition period for the workforce. I would expect this to pick up steam in 2026. As I said before, I think this is going to be a major topic of discussion and sticking point in the midterm elections in the United States starting in spring of 2026. We saw it just last week in America. The jobs reports wasn't what the current administration wanted to see, so they fired the person in charge of publishing the data. So I think it's just kind of inevitable that this is going to happen. And I, again, I don't want to be like the, the bearer of bad news, but I, I think a full blown economic and societal crisis around this. I don't want to say, like, it's totally inevitable, but it's greater than a 50% probability by 2027, 2028. And it's because when you go listen to the heads of all these AI labs who all believe this, they all think that this is going to happen. None of them have a solution. Like, they're all starting to look at it and research it. We'll talk about the Microsoft paper next. So they're looking at it, but nobody has a plan. And that's the thing that worries me is like, I think they all now realize what's going to happen and they don't have a plan, so they can't really talk transparently about it. But yeah, this is why I always said, like, I just didn't comprehend why people were ignoring this. And I, I think it just came down to they didn't know it was going to happen yet, and now they do. And so now we get earnings reports where they talk about how many people they get rid of and how efficient they can become because of AI.
- [34:03]
Mike Kaput
And I don't want to be a doomer here as well, but I had to just check this again because, you know, I think people sometimes lack perspective on some past periods in history that we are starting to, you know, make analogies to. Right. Like, I think we all think like, okay, the Great Depression, which none of us live through, we're thinking like, oh my God, nobody has a job. It's the worst thing ever. It's such a big anomaly. The unemployment rate during the Great Depression in the United states peaked at 25%.
- [34:31]
Paul Raitzer
Is that right?
- [34:32]
Mike Kaput
That's not that high. I mean, that's huge and has huge effects. But that's one in four people. Right?
- [34:37]
Paul Raitzer
So like 3 or 4% now? Something like that.
- [34:39]
Mike Kaput
Yeah, I think it's like 3 or 4%. So just to show, I mean, that's an enormous number relative, you'd have to, you know what, multiply 8x the amount of unemployment. So I get that that's a huge anomalous event, but it doesn't have to be like 9 out of 10 jobs go away due to AI for there to be a massive crisis like the one you're talking.
- [34:59]
Paul Raitzer
Any, anything touching 10% is crazy. Yeah. And so I don't. Again, I'm, I am not an economist. I don't know where the breakpoints are. Like, I don't know when we get. All I know is like family, friends, things you hear just in conversations. It's different. Right. Now, like I do with the job numbers wrong, right. All I know is it's starting to feel pretty different. And when I talk to people who are unemployed or underemployed, the prospects of jobs just don't seem to be what they were before. And I don't know, like I. It's just a really, really important conversation and it's something we need to be following closely on this podcast. But like you all need to be thinking about this in your own companies and starting to kind of try and look out 6 to 12 months and see what the impact in your company and your industry is going to be because it's going to be uneven. Like this isn't going to happen everywhere all at once. But I can just start to sense that it's once the C suite and board and investors get it, which they seem to be getting it now, then the dominoes start falling way faster than they have the last two years.
- [36:16]
Mike Kaput
So related to this we actually also got pretty big new study from Microsoft that was analyzing how exposed jobs are to AI. And what they did is they actually analyzed 200,000 real world interactions with Microsoft Copilot. And the researchers dug into what people actually do with generative AI at work, what the AI does in return, and which jobs that touches the most. And the results they found are people in this study are using AI most often for gathering information, writing, explaining things in other words, classic knowledge work. And AI for its part, tends to act more like a coach or assistant rather than at this time a full replacement. In fact, they found 40% of the conversations they studied showed no overlap between the user's task and what the AI actually did. And what they did is they looked at all these different occupations and how the highest and lowest ones exposed to what AI was being used for. And some of the top affected jobs according to their methodology include interpreters and translators, sales reps, writers and authors, customer service reps, news analysts, slash reporters, slash journalists and editors. And Paul, you know, I mean it's always good to see research like this. I like they're taking data, they take data right from actual usage of copilot and they like that, they share in the methodology. They also align that data with occupation and work activity info from onet, which is this public database of occupational data. And that actually kind of mirrors some of the ways we actually think about this stuff when we're building courses and when we're doing kind of some of our work. Trying to look at that, looking at Bureau of Labor Statistics and kind of total addressable markets there as well. The one thing is this data was gathered 1-1-2024 to September 30, 2024. So it kind of falls into that same thing we always kind of complain about which is the data is not exactly new.
- [38:13]
Paul Raitzer
Good is this right?
- [38:14]
Mike Kaput
Right. Though I do like it's based on how people are actually using one model at least though again it's not going to be the most powerful model that we have today.
- [38:22]
Paul Raitzer
So yeah, so those are great points. You know we've talked about Anthropic has done some work in 2025 with similar approach where they're looking at actual usage. Problem with Anthropic's data is it's predominantly used for coding so you don't really get this great subset of data. Microsoft Bing is obviously a more broad data set. 200,000 anonymized users. That's, that's great. The fact that it was June to September 2024, we didn't have reasoning models yet. So 01 wasn't introduced until September 2024. And we know that the vast majority of people still don't. Reasoning models are how to use them and that's the most disruptive technology potentially to, to high level knowledge work. So this is great. We, we need more study, we need more real time data like this. Like ideally we would see more data like this from Google and OpenAI. It cannot be year old data like we, we have to get this stuff in more real time. This is why I created the AI exposure key. So when I created Jobs GPT which we'll put a link to in if you, if you haven't seen it or used it before. So Jobs GPT is a custom ChatGPT that was meant to enable people to do impact assessments on Jobs, actually trained it on the O NET database. So in essence what happens is you can take any jobs if you've never used the O NET database. It's great. Like you can go up there, there's like seven or seven to 900 jobs occupations in there and it'll give you the standard tasks of every job. And so the whole concept behind Jobs GPT is to break a job into a bundle of tasks and then do an assessment of how that job will change as the models get smarter and more generally capable. So what Jobs GPT does is it doesn't just look at what current models can do, which is what the Microsoft study is looking at. It's like how are people who probably aren't even trained to use gen AI using gen AI with Today's models. That's good. What we actually need, though, to figure out the impact this is going to have on the workforce in the next 18 months is we need to project out where are these models going and how could different knowledge workers use these models once we get there? So the AI exposure key I created looks at as image, as video, as audio, as voice, as advanced reasoning, as persuasion, as AI agentic capabilities, as AI vision. As all of these abilities are built into these models and become more reliable, then what happens to writers and attorneys and consultants? So, yeah, it's important data, but again, you always have to look through the lens of when was it taken, who were the people that did it, and are they considering future known improvements to these models? Like, we know what's going to happen in the next 18 months, roughly. So I would say, like play around with jobs GPT. I actually built a capability in a couple months ago that lets you future cast any job or college major. And so you can put it in there. And it considers that exposure key against that job.
- [41:23]
Mike Kaput
And somewhat related to this, we also got a report in the Wall Street Journal talking about how McKinsey, the consulting giant, is kind of facing a bit of an existential crisis because AI can do much of its work faster and cheaper. And that reality, says the Journal, quote, is pushing the firm to rewire its business. So they have quietly deployed over 12,000 AI agents that write in McKinsey's signature tone, draft presentations, summarize interviews, and even check the logic of arguments that their consultants are making. And the firm's global managing partner told the Journal the goal is one agent per employee in the not so distant future. Now, meanwhile, since 2023, its headcount has dropped by about 5,000 people. Now, what's interesting here is kind of how the math has changed, because traditionally, McKinsey built teams of about 15 consultants per project, and they were aiming to, like any service business charge based on the scope and duration of that project. But AI is both speeding up the consulting work and means that fewer people are needed to work on each project. So today McKinsey puts three consultants on the same project he used to use 15 for, plus bots. One consulting industry insider even told the Journal that junior employees will likely be most affected immediately in consulting by these kinds of factors and that you can expect slimmer staffing to ripple through the entire consulting industry food chain. The insider said, quote, you have to change the business model. You have to make a dramatic change. Now, Paul, this definitely seems to align with, I mean, what We've seen, we have a huge professional services audience. You know, I mean, I was reading through this, just nodding and like saying out loud yes, because it's literally just proved I was on the right track with the AI for professional services course I'm doing. Because it literally touches on these factors. These are the structural issues that are sending the consulting industry and professional services at large towards a cliff that it needs to navigate. You know, so can you kind of unpack for us, like, what's going on here? What do consulting firms or other service firms need to be thinking about?
- [43:35]
Paul Raitzer
It is tough. Like, so I, I've said this before and we've talked about, you know, consulting firms and agencies. It's a great time to be an AI native consulting slash professional services firm. So if you're building one from the ground up and you can start fresh with a, you know, a pricing model and a talent structure and a service mix that's adapted to, you know, where the market's at today, and you can build from there, it's great. You can build a more dynamic, more efficient one. You know, fewer people, more revenue, and you can, you can be adaptive to the market much easier. It's a very difficult time to transform an existing one into an AI emergent firm. I owned a marketing agency and consulting firm for 16 years, sold it in 2021. Mike worked with me there as a senior consultant and leader for nine of those years. So this is our world. Like we, we lived this for a long time. The economics of that model are being reinvented. There's no obvious answer to what that's going to look like. The impacts on staffing and compensation models for your staff. Like, there was pretty standard ways you determined how much you could pay somebody based on what their billing rate was and how many hours per year they would do and things like that, that's being kind of tossed upside on its head. Service demand is shifting faster than ever before and the AI models are advancing like we just talked about, like, if you built your service mix not knowing that AI models could do reasoning tasks, and then all of a sudden they can do reasoning tasks pretty reliably, what does that do to your service mix? And so as your clients become more educated on these capabilities, their expectations of what you're going to deliver to them and at what price you're going to deliver to them, as is changing every day. And that's a really hard environment to manage. A firm of that size with all that legacy stuff and all the people who built their careers making a million dollars a year or something close to that, who maybe that expertise isn't as valued anymore. And like, they don't want AI to come in and do their job. They don't think it's capable of that. There's just going to be a tremendous amount of friction, and I'm, I'm sympathetic to it. Like, I, it's hard. There's going to be a lot of change within the professional service world in the, in the next, like, one to three years. A lot of turnover of top firms that, you know, kind of get disrupted by AI native upstarts. I think it's a massive opportunity. And if, if you're at one of those big firms, like, I mean, I, I want to provide career advice to people per se, but, like, there's never been a better time to do your own thing. Like, I, I, I, I believe that for a lot of people that's going to be the path is to, like, start fresh, you know, be dynamic, be nimble and building a more dynamic model. That being said, listen, there's great people working at these firms and there's great leaders at these firms. I'm just saying it's going to be hard for them to solve this. But, you know, this is why you get paid the money as a leader to kind of right that ship. We're going to talk about Apple next. Like, Tim Cook's going through this right now. It's hard to be the leader of a big established company, not just in consulting, in any firm, in any professional service firm, in any business. It's a very, very disruptive phase. And there's very little historical precedent that leaders can look back to that's going to get them through the next few years. It's unprecedented.
- [47:01]
Mike Kaput
All right, so let's talk about Apple because some speculation is heating up that Apple may be getting serious about AI. So on the heels of a $94 billion quarter, CEO Tim Cook said Apple is open to AI acquisitions and is reallocating a fair number of people internally to focus on AI features. Now, this obviously makes a bit of sense because we've talked about Apple's Siri revamps behind schedule. Meta is poaching talent like nobody's business. Apple has reportedly considered deals with OpenAI and Anthropic. They've floated acquiring Perplexity behind the scenes. They reshuffled their leadership this spring, which we've talked about. They moved Vision Pro head Mike Rockwell to lead Siri and AI efforts. And Siri is still struggling with some AI issues. However, Apple's Fundamentals remain strong. We'll talk a bit more about their earnings in a sec. But their iPhone sales are up. Services revenue hit 27.4 billion. There's a new iPhone on its way this fall. So like Paul, we've been talking about Apple's need to catch up here for some time. Is an acquisition or acqui hire of a leading AI company. The answer is a partnership with one of them. How does this play out?
- [48:15]
Paul Raitzer
I don't know that it solves it. I think I mentioned this on episode 159. Maybe like the more I was thinking about it, if they go, I think they will do acquisitions. But I just wonder will those people stay there? I don't know. It's such a fascinating case study here because they haven't really been penalized from a market cap perspective for largely sitting on the sidelines since Nova November 2022 when ChatGPT hit like Apple intelligence is useless largely. They have incredibly failed at making Surrey any better. Like and yet they're crushing it. And so it just goes to show the strength of their brand and distribution and the quality of their products. I feel like they have like one more chance. It's like the market has given them one more chance to figure this out. In the all hands meeting that Tim Cook called, which is an unusual thing for him to do. Bloomberg says the executive gathered staff at Apple's on campus auditorium Friday in Cupertino. So this is last Friday telling them that the AI revolution is as big or bigger quote unquote as the Internet, smartphones, cloud computing and apps. Quote apple must do this. Apple will do this. This is sort of ours to grab. Cook told employees we will make the investments to do it. He then went on to say, we've rarely been first, there was a PC before the Mac, there was a smartphone before the iPhone. There were many tablets before the iPad. There was an MP3 player before iPods. I'm laughing like I'm wondering how many like of our younger listeners don't Never had an MP3 player. Didn't know there were tablets for that. I think Apple invented all these categories. They didn't. But he said Apple invented the modern versions of those product categories. This is how I feel about AI. Then went on to say, echoing comments he made during the earnings call, Cook told employees the company is investing in AI in a big way. He said 12,000 workers were hired in the last year, with 40% of the new hires joining in research and development roles. So my feeling on Apple is they have the money. Do they have the culture and vision because many of the top AI researchers, they want to work on AGI and super intelligence. They don't want to build consumer products or make Surrey smarter. So can Apple do enough to attract and keep those people? Even if they hire or aqua hire or straight up acquire, can they keep those people there and compete with other AI labs? But then the question falls back to like, well, do they need to. Do they just need a few hundred people who aren't the top billion dollar researchers that everybody else is fighting over? Do they just need great AI researchers to execute because they have massive distribution through all their, you know, Mac and iPhones and iPads and Vision Pro and all these other, you know, devices. So I don't know, like, I, yeah, again I'm not going to give investing advice here, but I'm like, so I don't know. Apple's a really fascinating play for the next few years and what they do with AI.
- [51:15]
Mike Kaput
Right. There needs to be some movement here though, I think in a positive direction.
- [51:19]
Paul Raitzer
Yeah, they gotta get this, like, I feel like they're Gonna really, by 2026, they're gonna have some problems if they haven't nailed this.
- [51:27]
Mike Kaput
All right, next up, we got a bunch of quarterly earnings coming out from some of the big leaders in AI. So Paul, I'm going to go through just a handful of these and then we can kind of talk about these in aggregate or if there's anything that jumps out while I go, feel free to stop me. But first up, Google Alphabet, their parent company posted some strong earnings, but there are some complications here. So their parent company saw revenue jump 14% last quarter to over 96 billion and profits were up even more. Sundar Pichai, CEO, credited AI for driving strong momentum across the business. But there's still like a much bigger question. Can Google stay on top as AI reshapes search? The company is apparently all in on AI mode, but they now expect to spend 85 billion this year on CapEx, especially around AI. 10 billion more than planned. For now, investors seem cautiously optimistic. Their shares have recovered from some earlier dips. Microsoft also just posted some jaw dropping numbers that show its AI bet is paying off big time. They reported 27.2 billion in profit last quarter, up 24% year over year. Revenue hits 76.4 billion, also beating expectations. And basically AI and the cloud are driving this growth. They poured 88 billion into new data centers this year to keep up with surging demand from AI surfaces services, especially through the OpenAI partnership. Azure, the cloud platform, brought in $75 billion over the last year. And despite already being massive, Azure's growth rate jumped from 26% to 39%. Meta also posted a blowout quarter revenue of 47.5 billion. They beat expectations 3.848 billion users across Meta apps. And Zuckerberg, which we'll talk about in a little bit here, is also saying that Meta is all in on building what he calls personal super intelligence. At the same time though, they are seeing a lot of capex, their reality labs, they're kind of VR ar augmented reality has a multibillion dollar loss, but Wall street loves the direction. They sent the shares up by more than 10%. And then Apple, like we just talked about, had its strongest quarter in years. There's a surge of iPhone sales. Revenue rose 10% year over year. That's the biggest jump since 2021. That's thanks largely to the iPhone 16. Their Mac lineup had a strong quarter. Their services is now $27 billion a quarter business which rose 13%. Cook obviously said the company is significantly growing its AI investments and wants to acquire to accelerate that roadmap and sounds like they've got the money to do it like we just talked about. So Paul, I am by no means an expert investment analyst, but if I had to boil down these trends we're seeing this quarter, it seems like if anyone was worried these AI bets wouldn't pay off, they shouldn't be worried because they seems like investors are rewarding the AI bets even though these companies are doubling down on these really large capex expenditures.
- [54:35]
Paul Raitzer
Yeah, and just so capex, we use that term a lot and everyone's good to stop and explain what it means. So capital expenditures, it's a big thing that Wall street looks at in these earnings reports because what it's doing is it's referring to funds that a company uses to acquire, upgrade or maintain long term assets that are expected to provide benefits for more than a year. So it's like forward looking stuff. So if they continue to invest in cloud infrastructure, data centers, research and development, acquisitions, these are things that they're like longer horizon. And so if they, if the companies like Google and Amazon and Microsoft and Meta are increasing their capex number, that means they're continuing to see value in these long term AI plays. That's like a synopsis of what it is. So every quarter investors are kind of holding their breath to see is this a short term AI bubble? Like is this just a frothy period where it's going to eventually like collapse and like all these investments aren't really going to deliver the kind of value that are expected, or is it a long term transformation of the economy with AI as the underlying operating system? So they watch for things like cloud computing numbers, revenue growth, Usage data of AI and these CapEx commitments this year and beyond. And they're looking for guidance from these companies. So when they're saying, hey, we were already at 70 billion, we're going to 80 billion this year, that's a really good sign for the bulls, the long term AI bulls, who think this is going to keep going for the next five to 10 years. So for me, like I bet everything personally and professionally back in 2016 that Wall street was missing the big picture with AI, that investors didn't realize what would happen to the economy as AI progressed and was infused into every profession and industry. And I'm not again giving investing advice, but more a commentary on the state of AI. So I think generally speaking, most investors and business leaders still don't fully comprehend how early we are in this intelligent explosion and what the implications will as we move into the age of AGI and beyond like we've been talking about all throughout this episode. I think there's going to be downturns, I think there'll be doubts in the months and years ahead and there will certainly be some friction and resistance as AI starts to have a greater impact on jobs. But the end game for these labs is omnipresent intelligence. Like it's infused into every piece of software we use, every device that we use. And I think that we are still at the base of an exponential growth in consumption of energy compute from data centers and the underlying models that these companies are building and serving up. So I'm not saying there aren't going to be like quarters where earnings don't meet expectations or where capex doesn't like hit that number that Wall street wants to see. But I think when you zoom out, we are still at the very base of this exponential that this is going to. It's hard to comprehend because the human mind thinks in linear paths. Like when we think about why didn't people get it in 2016 when I was thinking this was obvious, why didn't people get it 2022 when it seemed obvious? Like there's been these moments where you look at it and it's, it's because it's really hard for the human mind to think about something that looks totally different than what we see today and tomorrow. And when you look at the exponentials though, of the scaling laws that are driving all of this and the seemingly insatiable desire for intelligence that consumers have. Those two things kind of indicate we are, we are just at the base of this continuing to grow. So I don't know, like I talked about in the Road to AGI episode, there are obstacles, there's things that could slow this down, but overall I think we're just at the start.
- [58:17]
Mike Kaput
So our next topic actually kind of looks at a bit like how quickly we are progressing here because Google just launched what they call deepthink inside the Gemini app and this is a souped up version of its AI that is designed to reason more like a mathematician. It is actually based on a variant of the Gemini 2.5 model that recently hit gold medal performance at the International Math Olympiad. That version took hours to solve problems. This one's much faster and aimed at real world use. It still hits bronze level performance on the same math benchmark. The trick here is something Google calls parallel thinking. Rather than sprinting to an answer, DeepThink explores multiple ideas at once, revises them and even combines them before landing on the best solution that makes it useful for way more than just math. Google says it shines at web design, scientific reasoning and algorithm development. Basically anything that requires building up ideas step by step. So this also tops a bunch of leading benchmarks for code generation and reasoning. The only catch here is right now you need to be a Google AI Ultra subscriber to use it in the app. That is the Ultra subscription that costs almost 250 doll bucks per month now. Paul, I, I don't know why I'm surprised, but it is pretty incredible to me to see we can, we literally went from talking about this like experimental cutting edge model winning the international mantholipiad gold one week, like a week or two ago, then we get consumer access to a version of it literally a week or two later, even if it is hundreds of dollars per month. I find that incredible.
- [59:55]
Paul Raitzer
Yeah, and this is the sort of advancement that's really going to impact these high level knowledge work jobs and consulting firms like we just discussed. You know, if you think about it, one senior strategist or researcher with these advanced capabilities for 250amonth, which is nothing talking about like if you know how to use them, will be able to do the work of 10 or more people. Like so if you're a McKinsey firm and you have access to this kind of technology and you can highly train people how to use this stuff, or a law firm or you know, a marketing Agency or a business with a C, you know, your C suite, your director level, VP level and you train them how to do this stuff. You're talking about transformation of work. Like there's no there, there is no like 1 2x thing. Like this is 10x stuff. And again, I just don't think that most business leaders are even aware stuff like this is possible. Like they don't really know how reasoning models work and how they can augment or in some cases replace human labor. Now it doesn't solve for the verification gaps like the AI gaps we talk about like verification, thinking and confidence. So if you go Back to episode 155 where we kind of preview this idea of AI verification gaps and I mentioned I built a whole course now on this in AI Academy, but you have the verification where someone still need to validate the work that comes out of it. You have the thinking gap where someone's got to apply the critical thinking to it. And then the confidence gap of like you actually have to understand the material to be able to present it and talk about it. But the AI labs are trying to solve for verification and thinking with other agents that are trained to do verification and thinking. Like, and that's kind of what deepthink does. It's this self improving mechanism that checks its own work and verifies it and then creates a, you know, a more polished finished product, I guess that then just needs the human to do some level of oversight. So I don't know, I mean I. These are those little product announcements that, you know, OpenAI has got something like this, you know, Anthropic's got something like this. Six months from now it'll just be commonplace that you can use models like this. I don't know, like these are the kinds of things that are, I think going to end up being way more disruptive than most people realize in the moment.
- [62:08]
Mike Kaput
Yeah. And just one more note about the verification gap. Like it is a. We've talked about this. It's a really, really good time to be an expert who has a lot of real world background and context. And I don't know how long that'll last. But I echo the advice you gave in the consulting industry that now is a good time to start something. Whether you start something or not, if you're in knowledge work and you are an expert that has all this domain expertise and background, don't waste this moment because there is at least a gap here where you are very, very, very valuable, more so than before.
- [62:40]
Paul Raitzer
I agree and I'll I mean, I'll think out loud here, and maybe this is a little more proprietary information than I should probably be saying out loud, but so like, Mike heads up our AI content studio within SmartRx and it's like an emerging area within the company that oversees the creation of all of the content, all the research, all the courses. And like, I'm wondering, Mike, like, do we need like a AI verification team? Like, is one of the things we build actually just a team of experts who verify the outputs within, you know, the research and things like that? Because you're going to have the high level experts, the lead researchers, the course instructors who need to have expertise in this, need to do the deep thinking, need to have the confidence in the presentation of the material. But it's possible you actually have a team of people whose job is largely to verify the outputs and work with the models and do some additional prompting. And so those are the kinds of things that I think about for like future. Org charts. And again, Mike and I have, I'm literally, I'm thinking of this in real time. Like, we've never had this conversation, but yeah, that's the kind of stuff I think people are going to solve for now. Is that needed five years from now? I don't know, but like, it's certainly needed right now and for the foreseeable future.
- [63:46]
Mike Kaput
Well, we've even talked about just how valuable it can be in certain contexts and we, you know, eat our own cooking in this respect of people having really hardcore journalism skills. Because while journalism as an industry is very economically struggling, I can translate those skills really well with some AI literacy to becoming a very good AI verifier or someone with those skills can. So it's interesting to also just think about instead of even, will this job exist? Like, how do we, I guess, retrain or reframe some of the existing skills out there too?
- [64:16]
Paul Raitzer
Yeah, and if we have any, anybody at the university level who is involved in journalism schools, something to be thinking about like that, that may be a future role, very near future role for your graduates.
- [64:29]
Mike Kaput
All right, next up, Mark Zuckerberg has recently shared his vision for the company's AI future. And this focuses on building what he calls Personal superintelligence. So he released a statement, a video, and then kind of an extended statement titled Personal Superintelligence. That's kind of like a letter to employees and to the world, I guess. So he starts this letter by saying, quote, over the last few months, we have begun to see glimpses of our AI systems improving themselves the improvement is slow for now, but undeniable. Developing superintelligence is now in sight. He then says that while many in the industry are focused on using AI to automate work at scale, Meta has a different vision. They don't want centralized control, but personal empowerment. So instead of building this single AI brain to run the world, Meta wants to give everyone a deeply personalized assistant. One that knows your goals, grows with you, helps you become the person you want to be. And Zuckerberg basically explicitly calls this out as part of what they're building. He says this is distinct from others in the industry who believe superintelligence should be directed centrally towards automating all valuable work, and then humanity will live on a dual of its output. He rejects that vision and says Meta is going a different direction now. Paul, obviously, I mean, Meta comes with plenty of its own baggage here. It is not always the most altruistic company on the planet. But I did personally at least appreciate the tone of his vision. He is saying superintelligence, quote, has the potential to begin a new era of personal empowerment where people will have greater agency to improve the world in the directions they choose. I thought it's at least a nice idea.
- [66:08]
Paul Raitzer
Yeah. Okay, I'm gonna. I'm gonna come back to this mic for a second. So I was actually, while you're doing that narrative, I was scanning to see if GPT5 had been announced yet. It's not, but ironically, the thing that pops up is the information has a headline. Why Universal verifiers are OpenAI's secret weapon. So literally, we just talked about the verification thing. It talks about how they're using reinforcement learning to train models to verify the outputs of AI models.
- [66:40]
Mike Kaput
Wow. Our idea to project the future of work is already outdated, right?
- [66:45]
Paul Raitzer
Yeah. So like I said, they may just have AI agents that do all this checking. So episode 161. We'll come back to that article. But okay, so back to the Meta superintelligence thing. So self improvement. So if you're. If you're again, like kind of newer to this stuff, and I get into this in my AI concepts 101 course, kind of explain these concepts of how these models learn and how they're trained and things like that, and what the dimensions of improvements are. Self improvement is right at the top of the list for everybody. And that's the idea that it's like a key unlock. Once the models can improve themselves, improve their own outputs, improve their own training data, things like that, then we have potential escape velocity for the intelligence that we can truly get to super intelligence. I don't even know that there's anything beyond super intelligence. But like once we get there, we, we can unlock everything that's possible. It's also a very slippery slope because once they can improve themselves, it becomes harder to interpret what they're doing, why they're doing it, things like that. So just know that self improvement is a known thing that has been pursued for years in AI and by all these AI researchers and him alluding to the fact that they're seeing that. I heard something similar from Sam Altman recently. I saw, I think, Ben Benjamin Mann on the episode podcast I referenced earlier, he talked about it. This is something you're going to be hearing a lot about the ability for these things to sort of improve themselves. And the verification thing I just mentioned from OpenAI is one of those ways, the ability to check its own work and then improve based on that. So there was a couple of excerpts, Mike, I'll add to the mix that in addition to what you were talking about, that I think are just interesting. So he said, in some ways this will be a new era for humanity, but in others it's just a continuation of historical trends. As recently as 200 years ago, 90% of people were farmers growing food to survive. Advanced syntax have steadily freed much of humanity to focus less on subsistence and more on the pursuits we choose. At each step, people have used our newfound productivity to achieve more than previously possible, pushing the frontiers of science and health, as well as spending time on creativity, culture, relationships and enjoying life. He's very optimistic about superintelligence, which will help humanity accelerate our pace of progress. Intersection of technology and how people live is meta's focus and this will only become more important in the future. Again, to your point, Mike is met at the company. We really want determining this to be determined. If trends continue, then you'd expect people to spend less time in productivity software and more time creating and connecting. Personal intelligence that knows super intelligence, that knows us deeply, understands our goals and can help us achieve them, will be far the most, by far the most useful personal devices like glasses, which obviously they're making huge bets on that. Understand our context because they can see what we see, hear what we hear, and interact with us throughout the way the day will become our primary computing devices. So that's kind of putting a stake in the ground which we knew, but I don't know he's been saying it quite as directly. They think the interface of the future is going to be through things like glasses. We will change from sitting in front of our computers and using our phones to things we wear that just, you know, know everything that's going on around us and the rest of this decade seems like likely to be decisive period for determining the path of the technology will take and whether superintelligence will be a tool for personal empowerment or force focused on replacing large swaths of society. So, yeah, he's definitely taking the opposition to the other labs and it's going to be interesting to see how that plays out and honestly, the implications of if he's right and they win. I, you know, I don't, I don't know. Like, I don't. I, I will say, like, well, I don't even know. I, like, if I think about my kids, like, 12 and 13, I would rather at this moment in time that they use AI built by Apple than built by metal. Like. And I don't, you know, I don't know how to obviously, anything too truly controversial here. Like, I, I think I would rather at, at this time the thought that goes into Apple's devices and intelligence and the, the principles with which they build that for versus, you know, what a social media company has been built around, which is all about engagement, keeping people on their apps. Like, I, I don't know. Like, I. Again, like, I don't. Nothing against Meta. Like, I. Meta's done some great things too. I, I just, I think these are the kinds of things we're gonna have to grapple with as parents, as business leaders. Like, which companies do you bet on? Which companies do you believe in? Which companies do you think is most closely aligned to the values of, you know, your company and your family? And those are decisions we're all gonna have to make and we're gonna have choices. Like, they're all gonna be building this stuff and everybody's gonna make, you know, their choice around it. But, yeah, I mean, I. And companies change. People change. You know, maybe they had a good direction and this ends up going well for society. I don't know.
- [71:45]
Mike Kaput
And, you know, incentives matter as well. I think looking at how the company makes its money is a helpful way to start gauging that too.
- [71:52]
Paul Raitzer
And I will say, like, I have friends within Meta and, and I will say there, there's really good conscientious people working on these products who do care deeply about the human side of this. Like, you can't. Meta isn't just Zuckerberg. It isn't just like, that one person and that you maybe, you know, 50% of people love them. 50 people maybe don't, but there's tens of thousands of other people working at Meta and many of them are really good people with great intentions and great hopes for humanity. And so I, I don't want to like. So I don't really like saying I do or do not like meta. I do or do not trust Meta. It's not just one person. And sometimes meta more than many companies gets bucketed into that one person and how people feel about him in particular.
- [72:38]
Mike Kaput
Yeah, it's a good reminder, especially with how personality driven some of these places can seem in the media. Right?
- [72:44]
Paul Raitzer
Yeah.
- [72:45]
Mike Kaput
All right, next up though the issue is now resolved for at least a short time. Some ChatGPT chats started being indexed in Google search results. And this wasn't really an accident. This was happening when users clicked share on a conversation and opted to make it visible on the web, which is an option. But many apparently didn't realize that option meant the whole world could find it with a quick Google search. So there was kind of this freakout for a while where thousands of intimate exchanges, some of them discussing trauma, mental health, specific family details, were now publicly indexed. One user talked about their ptsd, another described their personal history in vivid detail. Some named people in their lives in different ways. They were conversing with ChatGPT. Now the good news is, as of August 1, OpenAI has now patched the issue. Shared chats are no longer visible in Google Search. Our good friend Chris Penn at TrustInsights AI posted about this. He recommended that people regularly and routinely check their chat settings by going to Settings, data controls, shared links, then manage and then get rid of anything you don't need to share or that you didn't mean to share in the first place. So, Paul, I have to imagine this is like quite a wake up call for some people because I know, I know for a fact lots of users are not paying as much attention as they should to their chatgpt privacy and security. And we've also talked more and more people are relying on ChatGPT for really personal stuff.
- [74:13]
Paul Raitzer
Yeah, yeah, I think it's just a user beware kind of thing. And I mean, I just generally take the position, and I think I've said this before, like anything you share online, just assume, yeah, you know, your parents can read, your boss will read, like whatever. Like if you think you're doing in a private forum, don't assume it's private. If you're sharing a link that only people with the link can access, don't assume only the person you sent it to is going to be the one accessing it. So I think this is just more of a general awareness about overall user behavior online and certainly a bit of a not great look for OpenAI that they enabled this feature without being clear about it. They did fix it quickly, but yeah, it's like it shouldn't happen, but it's going to happen more and more and I think people just have to be.
- [75:03]
Mike Kaput
Aware of that for sure. All right, Paul, we're going to end up here with some AI product and funding updates that I'm just going to run through real quick here. So first up, Anthropic says it is rolling out new weekly rate limits for Claude Pro and Quad Max in late August. Sounds like Claude code is to blame here, Anthropic said some of it biggest fans are running it continuously in the background 24. 7, which is very costly. They said one user consumed tens of thousands of dollars in model usage on a $200 a month plan. Anthropic, however, says these rate limits are only going to apply. They estimate to less than 5% of subscribers based on current usage. Now at the same time, Anthropic is also closing in on a massive new funding round that could raise up to 5 billion. That would push its valuation up to a staggering 170 billion, which is nearly triple where it was earlier this year. And finally, in some other anthropic news, HubSpot has launched their first ever CRM connector for anthropic's Claude. So basically this makes the AI assistant far more useful for teams already running on HubSpot. So Claude can now tap into real time CRM data and respond with tailored summaries, charts and next steps. Also, Microsoft is now testing something called Copilot Mode in its Edge browser. This turns the browser into a full blown AI assistant. So with Copilot mode, Edge can scan all your open tabs, summarize comparisons, book restaurants, do all sorts of stuff through natural language. And with your permission, Copilot can access your browsing history, passwords and saved credentials to complete tasks on your behalf. Ramp, the corporate finance startup, which is known for its AI powered expense platform, just raised half a billion dollars, bringing its valuation to 22.5 billion, which is up from 16 billion barely a month ago. This cache is fueling their push into AI agents. Their first agent launched in July and it's used by thousands of customers to flag expense reports and check compliance. Basically like a digital accounting assistant, one finance manager said. It's doing the job of an entry level clerk, and Ramp says its system will reason through policy docs and predict expenses, and future agents will handle tasks like procurement and budgeting. Last but not least, Google just gave Notebook LM a big upgrade and it's all about turning complex material into something you can understand better. This new standout feature is called Video Overviews and basically it uses AI to generate narrated slides that mix visuals, diagrams, quotes and key data from your documents. You can then go ahead and customize these videos based on what you know, what you want to learn, and who the content is. 4 All right, Paul, that's a wrap on a busy, busy week in AI. I have to believe that this next week is going to be a big one as well.
- [77:54]
Paul Raitzer
Yeah, one quick note. The continuing soap opera between the AI labs. When Anthropic tweeted and then posted that they were like rate limiting people the next day, they also shut off OpenAI's access to the model. So it kind of like appeared as though maybe it was OpenAI that was abusing it, that someone within OpenAI was logging in, like using their agent non stop to test it and stuff. And so OpenAI somebody at OpenAI tweeted like hey, we, we still give Anthropic access to ours, but anthropic said that OpenAI was using it against the terms of use and whatever. So just the constant like little digs back and forth at each other. It's always entertaining. All right man, good stuff. Thanks again for curating. Mike and I are both still in the lab all week creating courses for the Academy launch, so definitely join us on August 19th. Lots to share with you all. And yeah, I expect another busy I think we're heading into crazy season. I think August is going to be interesting. I think September may be on a whole nother level when it comes to AI news and product releases. So stay tuned everyone and always something interesting to talk about. Thanks for being with us. Thanks for listening to the Artificial intelligence show. Visit SmarterX AI to get 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.