← The Artificial Intelligence Show
The Artificial Intelligence Show
#148: Microsoft’s Quiet AI Layoffs, US Copyright Office’s Bombshell AI Guidance, 2025 State of Marketing AI Report, and OpenAI Codex
01:24:55
Loading summary
Transcript75 lines
- [00:00]
Paul Raitzer
I just feel like people generally have no concept of how fast stuff's going to change because this change already lives in their labs and that's usually we're getting things that's 12 plus months old of what they've already been able to unlock internally. So there's always a smarter version. 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 148 of the Artificial Intelligence Show. I'm your host Paul Raitzer along with my co host Mike Kaput, who is fresh off his stint at Saster this week. So we are recording Sunday morning. This is an unusual recording day, so if you're new to the show, we normally record on Mondays, drops on Tuesdays, but we're recording Sunday morning, May 18th because I have to leave for London tonight. So Mike and I are like passing ships in the night here. So Mike was at Saster. Give me the quick rundown. Mike, you and I haven't really caught up. How was Saster? What was that like? I've never gone to the conference.
- [01:33]
Mike Kaput
Incredible event. It was my first time being there. Even in the short time I was there, kind of helping out, moderate a panel with Google Cloud. Just made a ton of great connections. Show floor is amazing. Tons of thousands of people there. Definitely, as far as I can tell, all in on AI. AI is everywhere. So yeah, I got to moderate a panel on their CMO Summit track which was kind of a private event within the event. Got to talk to leaders at Google Cloud and Demand Base about the future of AI for more CMOs and marketing leaders. It was awesome panel, awesome conversation. Had some fans of the podcast come up afterwards and take photos and stuff. It was fun.
- [02:17]
Paul Raitzer
I, I will say, like, I, I, I know you get out and do some events as well. It has become, I haven't quite adjusted yet to the fact that anytime I go to events there, there are like a dozen or so people that come up and like just want to talk about the podcast. So shout out to all of our listeners. I mean it's, it's still wild to me that this podcast has grown so, so big and we have so many people who are regular listeners to the show. I, like, I was at one, I think it was a week or two ago. It was on a Tuesday, so the episode had just come out like an hour earlier and someone came up to me and started like, talking to me about something from that episode. I was like, how do they know this? Like, this isn't even out yet. And then I realized, like, they must have listened to it in their hotel room before they, like, came down for breakfast.
- [03:03]
Mike Kaput
Oh, my gosh.
- [03:03]
Paul Raitzer
Just wild. Yeah, that, that's awesome. Glad you got to do that. Google Cloud has been an amazing partner of ours, so it's, it's always fun to get to do events with them. Yeah. And then I'm heading to London. I have a series of talks this week on, I think it's Tuesday and Wednesday maybe. So I'm, I'll be back in town later this week. But yeah, this is my, my last, like, big trip before the summer. And then I only, I only have like a couple of Eng for, for the summer, so I get to be home with the kids and Cheryl. So that should be nice. All right, so before I jump on my plane to London, let's get into episode 148. It is going to be a wild week ahead. We're going to. We'll give a little preview of some of the stuff that's likely coming. But we've got Google IO where we know some stuff's going to be dropping that's happening May 20th, 21st, which means that OpenAI is probably going to drop another model. We may get the anthropic model this week. It's going to be a wild end to May, I think. I think we're going to go into June with a bang when it comes to model releases and news. But we've got some bigger items to talk about going into this week. So this week's episode is brought to us by the AI for B2B Marketers Summit. This is one of our marquee virtual summits through Marketing AI Institute. This year's event is presented by Intercept and it is packed with incredible sessions from top B2B marketing experts. It's all happening virtually on Thursday, June 5, starting at noon Eastern time. You'll learn real world strategies to use AI to grow better, create smarter content, build stronger customer relationships and much more. So thanks to our sponsors, there is a free ticket option. So as with a number of our virtual events, you have the choice to do the free ticket option. There is a private registration option. I think that one's $99 if I' not mistaken. And that means your contact information is not shared with the sponsors. And then there is an on demand option as well. So if you happen to be at a different time zone or just can't make it, there is an on demand option there. So that is B2B Summit AI. Again, that is B the number two B Summit AI. To learn more. And this episode is also brought to us by Macon 2025. So this is our flagship in person event through Marketing institute, the Marketing AI Conference. We started this in2019. So if you are ready to drive personal and business AI transformation, join us in Cleveland October 14th to the 16th. On the shores of Lake Erie, right across from the Rock and Roll hall of Fame is where the convention center is that we will be at. With dozens of speakers already announced, dozens of breakout and main stage sessions you can check out, and four handson workshops, including one Mike is teaching and one I am leading. This is our sixth year for the Marketing Institute to bring forward thinking marketers together to share, learn and connect. We had about 1100 last year, so we're expecting more than that this year. Prices go up May 31, so grab your ticket now at Macon AI. That is M A I C O N AI. Okay, let's jump in again. If you're new, I know every week we have new listeners. So what we do with this weekly is three main topics is the normal format. So we kind of pick the three main topics topics. You know, usually we go 5, 7, 10 minutes on each of those topics. And then we go through rapid fire items and there's usually seven to ten rapid fire items. We do it all in about, I don't know, I always think it's going to be an hour. It never is an hour. It's always like an hour 15. But you can go up to 2x on your podcast listening network of choice and and get done in under an hour. So. All right, Mike, let's go.
- [06:49]
Mike Kaput
All right, so first up, Microsoft is cutting about 6,000 jobs, roughly 3% of its workforce, in a move that the company says is aimed at removing layers of management. And we're kind of seeing this as another example of the quiet AI layoffs that we've been tracking. Now, it sounds like from some of the reporting, the cuts are hitting software engineers the hardest. In Washington state alone, nearly half of all the layoffs that Microsoft did there were engineers, but product managers and technical program leads were also disproportionately affected. It seems like frontline sales and customer support roles were somewhat untouched. And Microsoft is kind of leaning into this messaging about saying we need to shed layers of management and build a more efficient business. And AI is playing a big role in that. In a few ways. Microsoft is spending big on AI, an estimated $80 billion this fiscal year on data centers alone. And the tools it's building seem to already be reshaping internal workflows and expectations. For instance, CEO Satya Nadella recently said AI now writes up to 30% of the code on some projects. Now, some are looking at this a little more starkly. So we found a post on X from someone we determined is an employee at Microsoft. A software engineer at Microsoft AI named Armin Khanker posted about this development saying, quote, my company just laid off thousands of tech workers to invest in AI. This is Endgame. So, Paul, starting back on episode 145, we began really talking in earnest about this phenomenon. You had indicated at the time it was your belief that quiet AI layoffs had been happening for the last six to 12 months, that they're now accelerating. Sometimes they're under the guise of other cost cutting. Certainly seems like Microsoft is an example of more of this. I mean, AI is a key factor in the efficiency plays they're making. Yet the stated reason for the cuts is about shedding layers of management. How are you looking at this and reading between the lines here?
- [09:03]
Paul Raitzer
Yeah, I mean, I definitely think that that's what's happening now. We have to accept, like, big tech companies lay people off all the time. Like, there's always the cyclical nature of these layoffs. But given the environment of where we are and the advancements of the technology and the comments from executives in other venues, other, you know, news outlets, it's not hard to connect the dots of, you know, kind of what's going on and what they think is coming. So just to like, call out a couple of elements from the Bloomberg article, Mike, that you were citing. So we continue. This is a quote from their spokesperson. We continue to implement organizational changes necessary to best position the company for success in a dynamic marketplace. In Bloomberg, they go on to say, Microsoft, which employed 228,000 people in June 2024, deploys periodic layoffs, often to reorient its headcount towards priority areas. So, for example, they laid off 10,000 people in January 2023. Does talk about the company being under pressure in recent years to keep a lid on costs amid the massive spending for the data centers. Like you had highlighted the 80 billion they have committed this year in capex. But then I thought what was telling was there was a excerpt from the article where they said last year CEO Satya Nadella said AI was helping the company save on labor costs. The theme came up again on Tuesday during a J.P. morgan conference when Microsoft finance executive Bill Duff said the company is, quote, saving hundreds of millions of dollars a year by using AI for customer support and reducing the need for human interaction. Duff said Microsoft is deploying AI across multiple divisions to help personnel analyze deals for compliance issues, write marketing materials and other tasks. The company routinely reorients its workforce. That is a very popular way of explaining what happens away from legacy products and toward growth initiatives. So again, it is common for sure. But the thing we've talked about previously is this idea that the software engineers who are writing code are basically the canaries in the coal mine right now for AI job disruption. Because that's the one that all the executives are talking about, how much code is being written by the AI. And as they look forward, they expect that to become dramatically more. So we'll drop a link in the show notes to a TechCrunch article in which Mark Zuckerberg was actually interviewing Satya Nadella. Last week might have been the week before on yeah, this was at the end of April at the llamacon conference which we talked about on previous episode. And so Satya said in that one, which you had just cited, Mike, that 20 to 30% of code inside the company's repositories was written by software, meaning AI. Microsoft CTO Kevin Scott previously said that he expects 95% of all code to be AI generated by 2030. So it's not like we're drawing some difficult to find conclusion here. If the CTO of the company is saying that within five years we expect 95% of all code to be written by the AI, then what do you need a bunch of engineers for? So this is just a prelude too. And it's basically looking on saying, okay, like as it gets better and better, more accurate, we're just going to need fewer software engineers. Sundar Pichai, CEO of Google Alphabet, said AI was generating more than 30% of their company's code. So now it's not clear exactly like the impacts can have. But Zuckerberg's on record saying they're going to have a junior AI engineer, software engineer by the end of this year. So, you know, it's just that's why I say the canary in the coal mine is like you look at this. And it's the one they're all building for because there's compounding value to automate with AI, the writing of code and the creation of products and the improvement of research and all these things. But if you then take that out and start looking at other industries and say, well, there's no reason they couldn't do this in a bunch of other professions, they're just choosing to first do it with engineering. So you'll remember this Mike, like back in 2014 when I wrote the Marketing Performance Blueprint, which was the first time I publicly shared any of my research on AI. And Mike worked with me on the research project to write the section of the book that was related to AI. So there's a section in my 2014 book called Algorithms and Artificial Intelligence. Now in that section I cited a study from the University of Oxford. So this was a study done in 2013 called the Future of Employment. How susceptible are jobs to computerization? Now the, that study was before gen AI. So this was just looking at machine learning. Deep learning was emerging like we had had the breakthrough in image recognition in 2011, 2012. So like they were starting to think we could see deep learning become a thing, but it wasn't a thing by any means yet. And at that time, that study found 47% of total US employment was at risk of computerization, which they called job automation by means of computer controlled equipment. And at the time the high risk jobs were telemarketers, insurance sales agents, real estate agents, market research analysts, marketing specialists, PR specialists, graphic designers, again before gen AI. And so if you now look at and say, okay, if we look at what's happening with software engineering and we know the labs are pushing on automating, coding, what happens when the startup world emerges and says, well let's go do this for accounting and bookkeeping, let's go do this for legal assistance, let's go do this for HR recruiting, let's go do it for diagnostic medical roles and all the content creation. It's all it takes now is just like someone to say, okay, let's push on this one, let's get those to 30, 40% with a view to 75 plus percent just like is happening with engineering. And so that's, I don't know, that's again why I think it's just so important we have these conversations and we start being realistic. I get the, that people are like doubt that this is going to happen. Like I understand that it's hard to like comprehend, but we're watching it Happen like it's, it's literally happening right now.
- [15:18]
Mike Kaput
Right.
- [15:18]
Paul Raitzer
And people still want to pretend like this isn't going to come and affect jobs in the economy. And I, I just don't understand that perspective. Yeah. So I don't, I don't know it Again, it's a topic we'll probably keep talking about because I think we, we just need more conversation in different industries. So like, if you're in the legal industry, look at this and then try and like start thinking ahead. What does it mean to you? It's just critical that we're doing that.
- [15:48]
Mike Kaput
You know, kind of as a counterweight to all that. We've kind of also talked about the fact that in the rush to potentially automate or augment workforces that there is a danger that you're using AI for things that it can't adequately do yet you're getting rid of people too fast. Kind of that difference between AI first versus AI forward. And we kind of did see an example of that this week because, you know, as early as late last year we've been talking about the company on and off Klarna, it's a fintech company and we've been talking about them because their CEO has been like all over the headlines bragging about how the company is freezing hiring and automating away hundreds of customer service positions thanks to AI While it came out this past week, he's now reversing course, going on a hiring spree because AI was not producing the quality needed for the roles that he needed it for. He actually said to Bloomberg as cost unfortunately seems to have been a too predominant evaluation factor when organizing this. Meaning his big push to have AI automate away humans. What you end up having is lower quality. Really investing in the quality of the human support is the way of the future for us. So Paul, maybe you could just give me a quick bit of context there. Like it seems like this isn't as black and white perhaps as some of the AI futurists want us to believe.
- [17:13]
Paul Raitzer
Yeah. So you know, we talk about all this automation and job disruption, but what we also have to do is have business leaders within these enterprises that are being realistic about what the capabilities of the AI truly is is because we do see too much like rapid like disruption happening because assumptions are being made that it can replace people. And one, like, as we've said many times, I'm not an advocate of that approach at all. Like, I'm fine with like, maybe we don't hire as many people moving forward, but we shouldn't be thinking first and foremost, like, let's just get rid of all the people. But I think that a lot of companies are just assuming, you hear all these AI agent talk and everything and they just assume it can actually do the job of people. So in this case, yeah, as a high profile person, they got over their skis a little bit, you know, in terms of making all these claims and now they're, you know, kind of circling back around, realizing people aren't all that bad. I did love the futurism. We'll put a link in the show notes the headline. I don't know if AI wrote it or if there's actually like a human writing these headlines, but the headline was Company regrets replacing all those pesky human workers with AI Just wants it. It's humans back. It's so perfect. And I can imagine seeing a lot of that. I, I think there could be these situations where almost like you see with Doge right now in the US Government, yeah, it's like, oh, let's just get rid all these pesky people. Oh wait, they were like fundamentally important to what we were doing and critical to our operations. Let's bring all those pesky humans back. And so I think you could probably see something like that occur. And maybe tech is again, tech's always going to be kind of leading on this, good or bad, where maybe 6 months, 12 months, and it's like, oh, actually we do kind of need a bunch of people and customer success because humans actually like talking to other humans sometimes. So I don't know, I think it's going to ebb and flow. I don't think there's this like cutoff point. We just fall off this cliff and all the jobs go away. I think it's just going to ebb and flow by industry and some people are going to move too fast and a bunch are going to move too slow. And that's probably actually the greater risk, honestly right now, is that you just sit around for three years and wait and see what happens with all these other companies that are figuring this out. That trial and error is going to lead to the breakthroughs in the innovations that are going to obsolete your company if you do nothing.
- [19:24]
Mike Kaput
Our next big topic this week is that President Donald Trump has fired the head of the US Copyright Office just days after the office released a report questioning how AI companies use copyrighted material to train models. Now, this report notably pushed back on the idea that scraping vast amounts of copyrighted content always qualifies as fair use, which is what Big AI labs tend to argue in this debate. Now, the firing of Shira Perlmutter, who had held this role since 2020, came right after Trump also ousted the Librarian of Congress. Some industry sources have speculated that this was a bit of a tech industry coup, possibly linked to Elon Musk's vocal disdain for intellectual property. And that's because this new report is seen as a pretty significant win or, you know, planting a flag in the ground for copyright advocates because it's not legally binding. But it provides a bunch of guidance from the US Copyright Office on how it's viewing AI companies using copyrighted materials. And at the core of this, like I mentioned, and at the core of this overall debate, is fair use doctrine, which major AI labs use to justify the fact they've scraped copyrighted material to train models and produce outputs. Basically, the thinking is if your use of the copyrighted work falls under these fair use rules that, that are enshrined in U.S. law, you are not infringing on the work's copyright. So this report is 113 pages and over it. It appears to challenge the idea that all this AI training is fair use. So according to AI copyright expert Ed Newton Rex we've talked about before, it basically concludes that some uses of copyrighted work by AI aren't what they would call transformative enough or changed enough in their usage from the original material. It also concludes that AI use of these materials in their training may seriously dilute the market for similar works. That's a big consideration when lawmakers judge whether or not something constitutes fair use. So Newton Rex actually said, quote, many of us have been saying this for a long time, shouted down by a chorus of AI boosters. It's great to see the US Copyright Office, the government body responsible for providing guidance on copyright law, agreeing. So, Paul, let's maybe first talk about the report's conclusions, then get a little bit into the political fallout here. Again, this is not legally binding, but it does serve as pretty strong guidance on what they're thinking related to copyright concerns related to AI. So how big of a deal are the conclusions in this report?
- [22:13]
Paul Raitzer
Yeah, I don't know what kind of long term impact they're going to have, you know, in terms of is it going to change anything. But I mean, obviously big enough that within 24 hours they fired them. So like the people involved. So, you know, again, just to rewind here, all the AI labs building these models absolutely used copyrighted materials. They absolutely knew it was a legal gray area and most likely illegal at the Time based on current US law. But they knew the other labs were doing it. And if they didn't train on the copyrighted materials, their models would not be on par with, with their competitors. And so they basically took a financial and legal gamble that whatever the, the outcome of this is, even if we end up losing in courts, one, it's going to take five to 10 years to all play out and the game's over by then if we wait, two is, you know, if we end up having to pay billions of dollars in fines, we're pursuing a market worth trillions of dollars, so we gotta go. And so they all kind of had this window to figure this out, you know, and this is part of what happened back in 2022 when ChatGPT came out and did something Google what hadn't been willing to do to that point. Google was under too much regulatory scrutiny. They couldn't have basically done what ChatGPT and OpenAI did. But once that was, you know, that happened in the market, it triggered this massive competition between Google and Meta and all the other players to just go do it too. Now again, they had all probably already been training models on these things, but they hadn't released a high profile consumer product with these models that had been trained on this stuff. So I think I'll call out a few excerpts that I assume is what caused for the firing. So this is straight from the document. We will put the PDF in the show notes, throw it in notebook LM and you know, explore it. It's like 113 pages if I'm not mistaken. So there's a lot to it. But I'll try and call out the things I think are the triggering points. So in the doc it says in providing this analysis, the office rejects two common arguments about the transformative nature of AI training. As noted above, some argue that the use of copyrighted works to train AI models is inherently transformative because it is not for expressive purposes. We view this argument as mistaken. Again, they didn't mess around with their language. They're very direct. Language models are trained on examples that are hundreds of thousands of tokens in length, absorbing not just the meaning and parts of speech of words, but how they are selected and arranged at the sentence, paragraph and document level. The essence of linguistics expression. Image models are trained on. Oh wait, then we go on to image models. Image models are trained on curated data sets of aesthetic images because those images lead to aesthetic outputs where the resulting model is used to generate expressive content or potentially reproduce copyrighted expression. The training Use cannot be fairly characterized as non expressive. Goes on to say, we do not agree that AI training is inherently transformative because it is like human learning to begin with. Because that's when you hear all the time, well, this is how humans learn. We go online, we absorb things, and then we like create things. So it's just like imitating how humans. So they're saying that is not true. They say to begin with. The analogy rests on a faulty premise, as fair use does not excuse all human acts done for the purpose of learning. A student could not rely on fair use to copy all the books at the library to facilitate personal education. Rather, they would have to purchase or borrow a copy that was lawfully acquired, typically through a sale or license. Copyright law should not afford greater latitude for copying simply because it is done by a computer. Moreover, AI learning is different from human learning in ways that are material to the copyright analysis. Humans retain only imperfect impressions of the works they have experienced, filtered through their own unique personalities, histories, memories and worldviews. Generative AI training involves the creation of perfect copies with the ability to analyze works nearly instantaneously. The result is a model that can create at superhuman speed and scale. In other words, of Professor Robert Braunness, generative model training transcends the human limitations that underlie the structure of the exclusive rights. And then two quick excerpts from the conclusion. Making commercial use of vast troves of copyrighted works to produce expressive content that competes with them in existing markets, especially where this is accomplished through illegal access, goes beyond established fair use boundaries. In our view, American leadership in the AI space would best be furthered by supporting both of these world class industries that contribute so much to our economic and cultural advancement. Effective licensing options can ensure that innovation continues to advance without undermining intellectual property rights. These groundbreaking technologies should benefit both the innovators who design them and the creators whose content fuels them, as well as the general public. Now imagine when this goes to the Supreme Court, that that is basically the argument being made on the side of fair use or on the side of the creators. And so that's why the administration's pissed. They basically know there's a bunch of legal battles involving Mark Zuckerberg, Sam Altman, Elon Musk, all the people who are giving the money to the administration who are fighting for, you know, American innovation. They don't want this to be the outcome. If it does, it impacts their ability to drive innovation in America. And the U.S. copyright Office just laid out the argument that could go against all of that. That's why they get fired. Now what happens? I have no idea what my best guess is someone gets installed into this position that rewrites this paper so that it does not when it goes to the courts. This is not the position of the US government because as right now, this is now the official position of the US Government that can be, you know, used in legal proceedings.
- [28:15]
Mike Kaput
So at first I was excited. We had more clarity on this, but it seems like that might not be the case.
- [28:22]
Paul Raitzer
Yeah, I, I don't know. I mean, we'll, we'll see what happens. I did see an article. I'll try and find what they're on the show notes. That was like the firings backfired because, yeah, somebody got installed into the position who's actually like anti tech and it might actually be worse for the tech companies.
- [28:37]
Mike Kaput
Yeah, there was some reporting from the Verge that actually what happened as of right now is that people who are installed in these positions are actually still part of the, the Trump administration's coalition, but less on the tech acceleration side and more on the kind of anti big tech side. So they're a bit hardliner on those. We'll see how that actually plays out. But not the ideal people you would expect to be in these positions if you're against copyright law.
- [29:05]
Paul Raitzer
Yeah, we'll try and keep an eye. I'm very curious to see how fast before we see these exact words quoted in an ongoing case. My guess is the briefs have already been written and it's like been submitted as, you know, evidence or supporting materials in existing court cases. Like, I can't imagine this wasn't within a day already being injected into actual briefings.
- [29:28]
Mike Kaput
Yeah, we had a. Obviously won't name names, but we had a partner of ours who was. Sent me an email the other day after we had done some work together and was like, hey, could you just like, what's the latest on AI copyright? We were thinking of these questions and wondering about some nuances here and I just like, wanted to cry. I was like, I'm so sorry, I don't have a good answer for you here. This is a highly paid lawyer's question because I can't answer.
- [29:52]
Paul Raitzer
Yeah, we are not giving legal advice. We say this all the time. We are not lawyers. There's no legal advice. And we're not providing investing advice. We are providing commentary on what is happening in the world.
- [30:02]
Mike Kaput
All right, so our third topic this week is about something we actually released, which is we released our 5th Annual State of Marketing AI Report from Marketing AI Institute and Smarter X And in this year's report, we had almost 1900 marketing and business leaders answer 23 different questions about AI usage and adoption. And we actually found a ton of really useful insights as a result about where the marketing industry is actually at when it comes to AI. So Paul, in this segment I was going to share just a few of the key findings from the report. Kind of get your thoughts where you have them on certain stats here. I would highly encourage everyone go download the full report@stateofmarketingai.com there's a ton more in here than what we'll go through here. So I'm just going to dive in.
- [30:52]
Paul Raitzer
And I think the webinar is on demand too. Like they could probably.
- [30:55]
Mike Kaput
Yeah. So if you go to marketingai institute.com under resources and click webinars, you'll be able to quickly see where we have the webinar on demand. Diving further into this, definitely also drop.
- [31:07]
Paul Raitzer
The direct link for that in the show notes too.
- [31:09]
Mike Kaput
Absolutely. So we did find, you know, to kick things off that marketers are racing ahead with AI. We ask about various aspects of the marketing team's AI usage every year. And this year actually 60% together either said they were piloting or scaling AI. You could choose from three phases, understanding piloting or scaling. And 60% are now either prioritizing and running AI pilot projects or say they are pursuing wide scale adoption of AI. And what's really cool is because we've started to benchmark this over several years, we can see that that's actually an 18 percentage point jump since 2023 where only 42% were in either phase, which is super interesting to see that kind of growth and adoption. When marketers, when we ask about what are you trying to achieve with AI, marketers say more. They're more likely than ever to say that they're trying to actually save time. So this year, 80 said they're reducing the time they spend on repetitive data driven tasks was the primary outcome they're trying to achieve with AI. Not only was this the top answer, it was the highest percentage of responses for this question that we've recorded so far. There was some cool data on tools. Respondents overwhelmingly said ChatGPT was their favorite tool. When asked, it was like 57% of people. We also asked if their organization provides licenses to either ChatGPT team or Enterprise, Microsoft Copilot or Google Gemini, or none of the above. ChatGPT was also a leader here with 61% saying that their company was most likely to provide them with a ChatGPT license. Now what's interesting is that actually steadily declines in popularity as firms get bigger. So it's much higher on the low end. One to $10 million, for instance in revenue had 70% of people saying this. It falls to 37% when you look at firms with a billion plus. And that gap is often filled by Microsoft Copilot now a big finding. We cite this every single year. A lack of training is a huge barrier to AI adoption. So once again, marketers cited a lack of education and training as their top barrier. When we ask about this, that's been the case every single year. We've published the report this year, 62% said it was a top barrier. Not to mention, we also ask, do you have AI training at your company? 68% of respondents say no. Then that consists of 44% saying they don't have it, 21% saying it's in development and 3% unsure if it exists. I mean, Paul, we talk about that stat all the time because it's really not changing fast enough.
- [34:01]
Paul Raitzer
Yeah. And just a reminder, you know, we've touched on this before, but like this is almost 1900 people from our audience. So this is like marketing AI Institute subscribers. It is people who listen to the podcast. This is like likely the leading edge in terms of adoption. So when you're hearing this data, it, it's not necessarily representative of the entire market because people who aren't doing this stuff likely aren't listening to the podcast or subscribing to Marketing Institute coming to our events following SmartRx. So this is like in my assumption, this is probably the best of the best. This is like more indicative of what innovators are doing than the laggards in the market. Yeah, I actually think the data is like skewed and has bias because it's a forward thinking audience already.
- [34:55]
Mike Kaput
Yeah. And if you check out the full report, we go in depth into the methodology, the respondents so you can see exactly kind of who answering these questions. A couple more data points I want to call out here. So unfortunately, marketers seem more pessimistic than ever about AI's impact on marketing jobs. So this year, 53% said they believe more marketing jobs will be eliminated by AI then will be created. We started asking this question in 2023 and it's the highest number on record. It's risen by 13 percentage points in just two years. Now, it should be noted marketers are a bit more nuanced when we also asked this year, what do you think the impact will be on Your job specifically, the Highest proportion, about 33%, say they're neutral in their concern about AI's impact on their job. They see both risks and opportunities. That was the highest percentage answer. Now, last but not least, we actually found through a series of data points that when you look at the data, at the end of the day, it's clear that things like AI roadmaps, AI infrastructure is the unlock. Because we ask about do you have an AI roadmap or strategy for the next one to two years that kind of outlines your use cases and direction for AI? 75% lack one. But when we look at who does have one, the people who do have an AI roadmap are also twice as likely across several data points to have training available, to provide prompt training, to have an AI counsel, and to have generative AI and AI ethics policies. So, Paul, I was just wondering to kind of get your thoughts on that. It really does seem like the way forward is just more formal AI infrastructure within your organization.
- [36:41]
Paul Raitzer
Yeah, I mean, there's so many interesting data points in here and the fact that we now have years of benchmark data is always so fascinating to me. To go back through this, I'll just kind of highlight in a couple of these. You hit on Mike. But. But when we, you know, every month I do this free scaling AI class where I teach these five steps for scaling AI and it's actually the basis for the entire Scaling AI course series. You can go to scalingai.com and see more about that. But we teach these five steps. And so one of the ways I actually assess AI transformation within an organization is to look at how far along they are with each of these steps. So the first is education and training. Do you have an Internal AI Academy? 32%. Yes. So that's only a third have it. Do you have an AI council? Only a third. 33%. Do you have generative AI policies? 38%. AI ethics policies? 41%. We don't ask directly have you done AI impact assessments, but to show the need for those, 53% now believe that more jobs will be eliminated because of AI in the next few years. Interestingly, when we asked the question, which I think was a newer question this year, how do you personally feel about AI and the impact it's having? 70% said positive. So despite all this, like, people are actually optimistic. And then the last one, Mike, that you had really zoomed in on is this idea of an AI roadmap. Only a quarter of them, 25%. So to me, this screens opportunity because again, these are probably more likely to be forward thinking AI, forward professionals and leaders. Yeah. And this is still where we're at. Like a third are doing the things that they actually need to be doing. So for you, regardless of your job, regardless of a company or what industry where you're at in your career, the opportunity is so massive to like move, like stop being overwhelmed by it and having all this anxiety about it. Just go do something. Because this data just keeps showing. We're not seeing leaps in any of this stuff. We're seeing, you know, 5 percentage point, 8%, which, you know, maybe 10%, 20% improvement each year. We are not seeing this, this, the, the switch flip to where it's like, okay, now 80% of people have generative AI policies and 90% of roadmap. And I don't think we're anywhere close. Like we've been doing this for what, five years? Did you say six years now?
- [38:56]
Mike Kaput
Five years, yeah.
- [38:57]
Paul Raitzer
Yeah. And like it's incremental every year. So I, to me it just screams opportunity and I would, I would race ahead to integrate these five components that I just highlighted.
- [39:08]
Mike Kaput
Yeah. Clearly doing something is better than what the vast majority of people appear to have been doing, which is nothing at the company level.
- [39:16]
Paul Raitzer
Yeah, Wild.
- [39:17]
Mike Kaput
All right, let's dive into this week's rapid fire. So first up, OpenAI has launched a major new tool for developers called Codex. This is a cloud based software engineering agent that writes code, fixes bugs and submits pull requests all on its own. Unlike a chatbot assistant, Codex appears to work more like a full fledged coworker. It takes assignments, run tests and returns finished work inside a secure cloud sandbox. It's powered by a model called Codex One. This is fine tuned with reinforcement learning to match real world developer habits. It can handle multiple tasks at once, work with actual repositories and follow instructions from project specific agents. MD files that guide how it navigates and tests code. So OpenAI says that companies like Cisco are already starting to test out this tool in real world environments. OpenAI engineers say they rely on it daily to reduce context switching and handle repetitive tasks. Right now Codex is available to ChatGPT Pro Team and Enterprise users and apparently plus and edu access is coming soon. So Paul, I wanted to get your initial thoughts on Codex. Obviously you and I are not programmers, but this definitely comes at an interesting time as we're seeing more and more people start to wake up to the fact that truly agentic AI like this could have a pretty serious impact on knowledge work.
- [40:46]
Paul Raitzer
Yeah, I mean a few quick takeaways here. People who build things already are going to build smarter and faster. So people who can code are going to code faster. Just as we talked about with the Microsoft story up front, probably not unrelated that these kind of advancements are happening, Microsoft has a unique relationship with OpenAI, likely has had access to this kind of tooling for potentially months. And so yeah, I mean I think things that get built are going to get built faster. So as a consumer of these products, as a business leader, as a practitioner, a knowledge worker, you, it's just going to explode. Like the apps that it can be built, the things these apps can do, the smart, the tech's going to get smarter. That's what this means to most of us. To the non developer audience, it just means developers are going to build faster and smarter. This is not a product that you or I are going to go play around with, Mike. It's a product for developers.
- [41:41]
Mike Kaput
Next up. In a recent chat with Sequoia Capital, Sam Altman reflected on OpenAI's journey so far as a company and as part of this discussion highlighted a really interesting aspect of the company's larger ambition. He talked about how they want to become the core AI subscription. Those are their words for your life. Basically the central interface used to interact with the world, powered by AI, hyper personalized to you and your preferences. Now related to this in this interview he also notes that kind of crazily there's already this, what he calls a generational dividend when it comes to how different age groups use AI. He said he was seeing that older people tend to use Chat GPT as a Google replacement. People in their 20s and 30s are using it as a life advisor and people in college seem to be using it like a full fledged operating system. Now Paul, taken together I thought the points about becoming a core AI subscription for your life and the fact that these different groups are already using this in different ways in that context I found all that really fascinating to kind of think about where we're headed. Like is this the direction we can expect AI development to go?
- [42:59]
Paul Raitzer
This is definitely meant to be a rapid fire item. However, as I was kind of prepping this morning to go through the notes, I actually realized this connects very deeply to another podcast I'd listened to earlier this week. So I watched this interview with Sam and It's like a 30 minute Q A. It's a Q A with the moderator, but then largely with the audience who's made up of Sequoia portfolio companies. I, I Assume like you know, CEOs, founders of tech companies funded by Sequoia. So they get this inside access to Sam as part of that portfolio. So I would recommend watching it. He tells an interesting story of Chat GPT that I actually hadn't heard told in the way he told it, like the origin of it and their own doubts around it and things like that. So it's kind of like an interesting first like seven minutes where he sort of got goes into that. But I want to, I want to call out a few quotes he had and, and provide them under categories because I think this is actually, it ends up being really important. Like I think six months from now we could probably come back to this episode. What I'm about to tell you I think will make a lot more sense. So I'll do my best to explain why I think it's really important right now. So first you got to ask about enterprise adoption. And he did say it's going to reshape everything but that basically companies move really slow. He then he really nailed on this like 20, 30 year old thing. He kept coming back to this theme, but he basically said a bunch of enterprises are just going to get obsoleted, that the startups are just going to move so fast and these companies aren't going to move quick enough and they're basically all screwed is kind of what he was saying. But he said, you know, this basically happens to people too. That's where he said talk to a 20 year old about how they use ChatGPT and then go look at a 35 year old and it's just completely different how the uses is. And this actually ties back to an episode or two ago when we talked about them having to pull back the updated 4O model because they didn't realize how the younger generation was using it fully. And then they realized once they did that like they hadn't really accounted for that usage as like therapists and, and friends and relationships and things like that. When he talked about the young people use, he, you know the, the quote that you kind of alluded to, he said gross oversimplification. But older people use ChatGPT as a Google replacement. People in their 20s and 30s use it as a life advisor and people in college use it as an operating system. The operating system is the key phrase here. I will come back to that. I thought it was really interesting. They asked him about an importance of voice and we had all this, you know, drama last year when the, the voice agent came out and it started it sounded like Scarlett Johansson from her and. And it like made you think that they totally cracked the code on voice. Well, he said, quote, we just haven't made a good enough voice product yet and that's fine. It took us a while to make a good enough text model. We'll crack that code eventually and when we do, I think a lot of people are going to want to use voice interaction a lot more. When we first launched our current voice mode, the thing that was most interesting to me was it introduced a new stream on top of the touch interface. You could talk and be clicking around on your phone at the same time. I continue to think there's something amazing to do with the voice user interface. We just haven't cracked it yet. But before that we'll make voice really great. And when we do, not only will it be cool with existing devices, but but I think a voice will enable a totally new class of devices if you can make it feel truly human level. So that's a prelude to them getting into the device business, which we know that they're looking at. They talked about future innovation and he was basically okay, so is it algorithms? Is it data? Is it these massive data centers? Is it more Nvidia chips with compute? Like what is the thing? And he did say that he thinks there's still big algorithmic breakthroughs to be TAD that'll unlock 10x to 100x improvements. He said, I don't think there's very many, but there might be one or two like the invention of the transformer, like these truly, you know, breakthrough things that change the future of development. And then this is the one I wanted to center on when he got asked about customization. So we've talked many times about how OpenAI's vision is truly personalized chat GPT experiences to where, you know, Mike, you and I have totally different experiences with the ChatGPT, right? So he said, the interviewer said, I'm curious how you're thinking about customization. You mentioned the federated Sign in with OpenAI bringing your memories in your context. I'm just wondering if you see customization and these different post training application specific efforts as a kind of band aid or a step toward making model core models better. How are you thinking about that? And he said in the platonic ideal state is a very tiny reasoning model. So imagine like oh three with trillions of tokens of context you put your whole life into. The model never retrains, the weights never customize, meaning the underlying model just doesn't evolve. They just have a great Reasoning model. But that thing can reason across your whole context and do it efficiently. Every conversation, again, this is still Sam. Every conversation you've ever had in your life, every book you've ever read, every email you've ever read, everything you've ever looked at is in there. Plus it's connected to all your data from other sources, and your life just keeps appending to the context. Your company does the same with all its data. We can't get there today, but I think of everything else as a compromise off of that platonic ideal. And that's eventually how I hope we do customization. That is the operating system vision. So when he talks about it being the operating system for your life, for your business, that's it. So when, when we get to the ability to have a trillion tokens of context come back to that moment and he just laid out for this is like, I almost. When I heard him say this, I was like, this is like when he wrote the Moore's Law for everything in March 2021. And people just, like, nobody paid attention. This is it. This is the same thing. Like every. When they introduced memory three weeks ago, when they introduced the agentic abilities, all of it is for this. So then that led me to this other interview that I. I didn't even realize it was going to be connected. But on a flight last week, or whenever it was, I was listening to Logan Kilpatrick, who we've talked about on the show before Google DeepMind. He was interviewing, like, a legendary researcher, Nikolai Savanov, at Google DeepMind. And the whole premise of the podcast episode was about long context windows. And so I'll call out a couple of things because it's extremely relevant here. So he talked about, like, in the early days, how like 200,000 tokens of contacts. Now, again, if you're new to this stuff, a token is a piece of a word. And so a given word may have like two tokens in it, three tokens in it. So we're talking about 200,000 tokens. You're looking at. It's like roughly 70, 80% of that is like the number of words that it can remember, basically. So 200,000. You're talking about, like 170,000 words. So imagine you're having a conversation with Chat GPT. It remembers things about you. It can roughly remember 170,000 words and talk coherently based on the knowledge that exists within that word. So our last book, Mike, was 50,000 words. So basically, like three of our books is like the context window of These models originally.
- [49:47]
Mike Kaput
Right.
- [49:48]
Paul Raitzer
But then DeepMind ships one that does a million context. So. And I think 2.5 pro is 2,2.
- [49:55]
Mike Kaput
Million I think at the moment.
- [49:57]
Paul Raitzer
Yeah, yeah. So imagine like if you wanted to use this in your business and you want to give it like a bunch of proposals you've written, you can give you know, up to say 1.5 million words of context and it's going to like, remember. So the way Notebook LM works, you throw a bunch of stuff in there. The reason it can cite things, it's because it has a really strong context window. So here's where it gets interesting though. When they did the research release in spring of 24, it might have been at the IO conference. I don't remember when they previewed this, but they said in the lab we tested this on 10 million tokens of context and it works still. So they told us a year ago they could already do 10 million. So this is the first time I heard this explained though. He said when we released 1.5 Pro, we actually ran some inference tests, meaning testing it like actual chat usage at 10 million. And we got some quality numbers and single needle retrieval. Like we're trying to find a really small thing within there. It was almost perfect for the whole 10 million. We could have shipped this model. So they're saying like we had a market ready 10 million context window a year ago, which means they probably have a hundred million right now, 10x, you know. But he said it's pretty expensive to run it and basically no one had a use case for it. So like if we built this thing, we put it out in the world like no developer had anything that they would use it for. So like, ah, let's just like ship some other stuff. So I was like, okay, so this starts getting really interesting. But then he says, what I think is going to happen is when we achieve close to perfect million contexts, then it's going to unlock incredible applications, like something we could never imagine, like the ability to process information and connect the dots. It'll increase dramatically. What I think is going to happen, and this is Nikolai again, what I think is going to happen is these superhuman abilities, they are going to be more persuasive. The better the context we have, the more capabilities that we could never imagine are going to be unlocked. So that's step number one. The quality is going to increase and we're going to get nearly perfect retrieval. After that, what's going to happen is the cost of the long context is going to decrease. So to do the 10 million is going to cost as the same as the 1 million today. And that's going to. Every 12 months this will happen. So to get to a billion in like two years, it'll basically cost them to do what they're doing today. So I think, so the cost will increase. So I think reasonably soon we will get that 10 million contacts window. I actually, I have no idea if it's gonna. I would not be shocked at all if that actually gets announced at IO this week that the 10 million is available to build on. That would be really logical actually, now that I'm thinking about it. So the 10 million becomes the commodity. So OpenAI chatgpt today is 300,000. I don't even know what chat GPT is. I don't even think they're at a lot.
- [52:36]
Mike Kaput
It's a lot lower because.
- [52:37]
Paul Raitzer
Yeah, so Google's way ahead on the context window thing. So he says when this happens and we get basically the super cheap 10 million, that's going to be a deal breaker for some applications like coding. So go back to the coders. Do we really need people writing code by 2030? He's saying no, because the flaw right now is code databases are massive. But once you can put an entire code base into it and it's perfect at retrieval, then you don't even, you don't need the humans to do the code. So that gets interesting. And he says, we're going to hold all this information in the memory at once and they're going to be reproducing any part of this information. Precisely. Imagine this in the legal industry. Not only that, they will also be able to connect the dots. They will find the connections between files better than coders. Basically, I imagine we will very soon get superhuman coding systems. They will be totally unrivaled. They will basically become the new tool for every coder in the world. And so when this 10 million happens, that's the second step. Then it gets into agents that can operate effectively because now they have basically infinite context. They can keep track of everything. They don't lose previous actions, all these things. And then he says, so to keep all these previous interactions in memory, so we get back to memory, you need longer context. So that's where long context, nature. So to, to realize, to wrap this up, to realize the vision Sam is laying out for an operating system, which Google probably has the same vision. It's why they're turning on memory within your emails and calendar and all these things and your search history. Once they have perfect retrieval of everything and everything you've ever done that you've allowed them access to is in the memory, in the context, because it's basically infinite and they have perfect retrieval ability. Instantly everything changes. And so, like, the key with this interview is this is two tech people. Like Nikolai is not the CEO trying to juice the stock price or raise the next 40 billion or whatever. This is a dude living within the labs building long contest. He's like the expert on long context. He has no reason to hype this stuff. He's, he's just talking about the reality of what they're building. And that to me is like, so I guess demonstrative of what is likely to come and why. I just feel like people generally have no concept of how fast stuff's going to change because this change already lives in their labs. And that's usually we're getting things that's 12 plus months old of what they've already been able to unlock internally. So there's always a smarter version. So we always say, like, this is the dumbest form of AI we're going to have Notebook lm as incredible as it is, Deep research, as incredible it is, they have smarter versions of it 100% within their labs that they're testing right now.
- [55:15]
Mike Kaput
Yeah, it's fascinating to that last point, because I don't think this applies to our audience, but there's so many people out there that I feel like just stopped updating their mental models. Like, I still hear so much about, well, it hallucinates well, it can't do this. Well, it can't. Documents in memory, whatever. And it's like, sure, I'm not denying any of that, but you really need to start paying attention to what's already existing.
- [55:41]
Paul Raitzer
Yeah, I, I just being realistic, like, I, I don't, I, again, I don't think it's ignorance most of the time. Like, I, I think it's denial. Honestly. Like, I, I really think that the majority of people who aren't willing to like, realize that this is a distinct possibility, that the, the next few years looks nothing like, like today. I think it's just denial that like nothing can change that fast. It's not possible that it can do what I do. Like, I think it is, and I'm not a psychologist by trade. I think it's like largely rooted in that, that we just can't accept it that it could possibly be that different that fast.
- [56:19]
Mike Kaput
I agree. All right, our next topic. There is some more drama shaping up between Sam Altman and Elon Musk and Musk's Grok AI chatbot is in the middle of this. So a new spat between the two started off when Musk posted about Altman online. He reposted something and appeared to call out Altman for past tweets critical of the Trump administration way back in 2016. Altman responded saying, hey, I got it wrong back then, and then posted an article from 2022 that reported Musk was being Trump critical of Trump at the time. Then Altman posted, hey, anyway, see you next week, in apparent reference to both of them being in the Middle east for a state visit by Trump. And he then said, quote, let's be friends. AGI is too important to let a little feud get in the way.
- [57:11]
Paul Raitzer
Now with the sad emoji and with this hilarious.
- [57:17]
Mike Kaput
It's crazy. These are going to be like the. The pieces of content that are, like, enshrined in history here.
- [57:23]
Paul Raitzer
Emojis.
- [57:24]
Mike Kaput
Yeah, and Musk, I don't think responded to that one, but Altman took another opportunity to needle him a bit because it came out just after this that GROK spent several hours over this week for some reason pushing a bizarre fixation on, quote, white genocide in South Africa. It was injecting this topic into all sorts of unrelated posts and responses. XAI comes out blaming the incident on a, what they call an unauthorized modification to 3am prompt. At 3am according to Xai, someone altered Grok's instructions to force this politically charged response in direct violation, apparently, of their internal policy. Altman then kind of gets the last laugh here, saying he posted about this and said, quote, there are many ways this could have happened. I'm sure Xai will provide a full and transparent explanation soon. But this can only be properly understood in the context of white genocide in South Africa as an AI program to be maximally true. And then he trails off, like, totally trolling. So, Paul, I kind of want to focus on, like, the. There's obviously always this drama, this back and forth, but this is not the first time something like this has happened with grok. And like, I guess I just have to ask because both times it's been seemingly, like, issues that one person in particular really cares about. And that person is Elon Musk. Like, is he just going in and making these changes?
- [58:57]
Paul Raitzer
I don't know. That is certainly the. What is being implied by Sam and other people online is recall. Like, two months ago there was this issue again where GROK got changed by an employee and they supposedly put a whole bunch of, like, guardrails in place so that couldn't happen. So I, I would imagine that the number of people who can at 3am make a change to the system prompt that guides Grox behavior is quite limited. And there's certainly one person in particular you would assume has that ability. So I have no idea. But for me at the bigger picture level one, it's funny obviously in the Middle east trip where they were there for like these energy deals and data center deals which you know, we're not going to get into in this episode, I don't think. But like there was massive deals being made to drive AI infrastructure basically. And so they were there together, which is what Sam had alluded to. And then this tweet kind of taking these shots at Grok happened six days later. So we're guessing they didn't make peace while they were, while they were there was not a peace summit, I would say. Yeah. The fundamental problem to me that this is demonstrated of is the bias of the models we are relying on to power society. So like it's basically five models right now that the vast majority of users use and one of them which is going to be embedded into humanoid robots, cars X, all these places, apparently someone can just go in and change them and it completely changes their behavior in the way that the human is trying to change it. It. So Grok, I got to give him credit. The thing is pretty honest, it's, it's training data has caused it to say all kinds of things that Elon Musk would not want it to say. There does just seem to be certain topics that really bother him that then somehow magically the system prompt is is changed to not do that thing. This is the cons. This is why there's all this push for like they need to be democratic. Like because the humans decide they're democratic, they provide the training data to make them democratic in nature. So this, it's just illustrative of the fact that we are reliant on human decisions and bias to actually build these supposedly unbiased models.
- [61:24]
Mike Kaput
Next up, a new survey came out from our friend SEO expert Andy Crestedena and the research firm Question Pro and it has some compelling new data on AI and search. Their survey pulls over a thousand people in the US across a range of questions related to AI and search. We'll put the link to the full recap that Andy posted in the show notes. But a couple interesting call outs. They found 72% of searchers use Google's AI overview when it appears, which has big implications for click throughs and organic traffic. 62% of people use an AI chatbot every day, 51% of chatbot users plan to use them more, and 49% of people think AI will eventually replace search engines. Additionally, it talks about how Google's AI overviews now appear at the top of half of all search results. And ChatGPT, which obviously has barely existed for three years, is now the dominant AI tool. It is outpacing by a long shot Microsoft and Google's AI tool. And it's not always clear when people are reaching for chatbots over search based on the data. But this data does provide some more nuance. So for local info or reviews it seems Google is still winning there. But for messy kind of multi part questions, think something like find me. Venues in Chicago with vegan food and river views. Chatbots seem to increasingly be the default for people on those types of searches. Now among all this, they point out very clearly search usage is actually up, but the ways in which we seem to be searching are changing quite a bit thanks to the existence of AI chatbots. So Paul, I found this to be pretty helpful and nuanced. Like I like that we're not just saying, you know, search or SEO is dead or alive or whatever. We're just getting a more balanced picture of the changes ahead. But I do think we just have to accept that this is going to change a little faster than we anticipate, don't we?
- [63:28]
Paul Raitzer
Yeah, I'll just give a quick like personal anecdote here. So I was planning a trip to New York. So for, for Mother's Day for my wife. She's you know, loves libraries and bookstores and rare books in particular. And I was like, I don't like that'd be a fun Mother's Day gift. Like let me like plan a, like a book tour basically in New York. So I actually went into chat GPT first I was like I'd love to like plan this like unique getaway for my wife for Mother's Day. You know, what are some cities we could do it at? And if it was, I started with cities and actually came up with New York and then I happened to have a talk in New York. So I was like oh perfect. Like let's do this. So it found the library hotel for me which I did not know existed that has Dewey decimal system floors and like rooms. Like it's awesome. It's across from the New York library. So I was like perfect. And I was like okay, what else would you see if you were going to go on a multi day book tour in New York. What are the places you would have to see? And it wrote up a summary of, like, six bookstores at why they would be important. I gave a little context of, like, the kinds of books she likes, things like that. So this is all living in ChatGPT so far. Everything's happening. It's giving me URLs, whatever. So then I was like, all right, well, let me map this out. Like, how is this in relation now? I went to Google Maps and I actually, like, popped in each one. Like, where are these things in the map? So that gave me a function for. For that. And then the only time I ever actually went into Google was to. To look at the actual websites of each of the places. It was not the search function itself. And then I think I did one Google search just to verify of, like, oh, what are the place? Basically the same as I gave you the prompt, like, what are the places we would have to see? What are the bookstores? And it just verified that, yeah, the six that I already had were, like, the places to go. So, I mean, that's. That's an evolved search for me. That is a, like, a total change in how I would have planned that trip previously from a year ago even.
- [65:13]
Mike Kaput
Right.
- [65:13]
Paul Raitzer
And honestly, it was amazing. And then I had chatgpt like, help me write descriptions of each of them and created like a, you know, a presentation with, like, about each one, and it was perfect. So, yeah, I think that more and more that's what's going to happen. And I do think the next generation, like, it's just. They're just never going to go to.
- [65:30]
Mike Kaput
Well, like we talked about with them treating it as an operating system, that changes everything.
- [65:35]
Paul Raitzer
Yeah, for sure.
- [65:37]
Mike Kaput
All right, our next topic is related to kind of a hot topic we talked about last week. Then we talked about a New York magazine article that went viral about how AI is causing a cheating epidemic in higher education. You know, in that article, the educators interviewed were kind of almost at the point of despair about AI's impact on classrooms. They were really worried it was replacing wholesale the need for students to do critical thinking and engage with the learning process. So we actually wanted to continue the discussion of AI in education this week with a couple other updates that provide some more context on this topic. Now, this week, it's actually about how AI could be impacting education in positive ways. So a couple developments that we've been tracking. Bowling's Green State University, which is in our backyard here in Ohio, is launching the first undergraduate degree of its kind in the US which is a Bachelor of Science in what they call AI plus. This is set to begin fall of this year. The program pairs core AI training with a second field of study so you it is offering students the chance to apply AI to one of six disciplines Computer science, math, physics, History, journalism or public relations. So the goal is to prepare graduates not just as people who are comfortable around AI technology, but also as domain aware technologists ready for a diverse set of careers. And then second, some more positive news A new study seems to indicate that ChatGPT may not be having as adverse an effect on students as some people think it does. A new Meta analysis of 51 studies found that ChatGPT significantly enhanced student learning performance perception and higher order thinking. The strongest effect was on learning performance. Students benefited most when ChatGPT was integrated into problem based learning models and skills development courses. The impact on learning perception, which is students attitudes and motivations, was moderate and highly dependent on duration. Longer engagements beyond eight weeks or so led to better emotional engagement, possibly due to getting increasingly comfortable with using ChatGPT in an education context. And then short term use unfortunately showed limited gains. But they say this is likely because students hadn't yet developed the prompting skills to use ChatGPT effectively.
- [68:07]
Paul Raitzer
Right.
- [68:08]
Mike Kaput
So Paul, we don't want to like diminish the very real disruptive impact that AI is having on education right now, but it's also important to highlight the positive which is when used in the right way, it seems like AI can have a really good impact on learners. How do you weigh these positives and negatives when it comes to this? Because we know we just see like the dark side so often here as well.
- [68:30]
Paul Raitzer
It just comes down to responsible use and teaching of responsible use. And you know, I think even my LinkedIn post about that article, I had some wonderful comments from higher education professionals who were like, yeah, here's what we're doing in our classroom and some of it was amazing stuff. So there are absolutely outliers here of, you know, entire schools and more specifically into like professors and teachers, administrators who are racing ahead to solve for this. And I think we're hopefully going to hear more and more stories like that. I've been an advocate since 2018, I think was the first time I proposed this to a university that AI101 should be required for every freshman. Like it should be the absolute thing when you come in and then you have the fundamentals and then like what BGSU is doing, which is phenomenal, like you know, kudos to them for doing that and thanks to my dad for sending me that article. I actually hadn't seen that one, so I think we're going to continue to see that. Now I will say in the meantime, because this is not going to be evenly distributed. I, I get asked all the time about advice from parents who have students who are in high school or college. And so I had someone actually reached out to me last week about this. So here's the general advice I give just for people trying to figure out what to do. What I say is like, regardless of major, get whatever AI experience and education you can to complement the major that they choose. Each university is handling this differently, you know, in terms of what they offer, as we see with bgsu. But I would look for any additional learning opportunities that are focused on the business side of AI, not necessarily the computer science side. Right. So as we talked about earlier, we don't know the importance of coding. I'm a believer that learning to code has far benefits, far beyond just building software. You know, I think I mentioned this recently, but it teaches you to do hard things, repetitive things. It teaches you to problem solve. It teaches you how to, you know, be a critical thinker. Like there is value to learning to code regardless of if you use it in your career. But there's tons of opportunities, free, low cost, from Coursera, LinkedIn, learning, Google, Microsoft, OpenAI like all the tech partners. HubSpot has the app stuff. So if you're looking for education, you can go, you know, add to your degrees and your studies with elearning platforms. Our AI academy that we're, you know, we've had for five years, but like we're reimagining and rebuilding now, we're doing a lot more work with things that'll be additive too. We actually have partnerships with some universities who buy licenses for students in classes and then they take our classes as like complimentary to what they're learning in schools. And you know, we're working on some much larger initiatives there under our AI Literacy project to expand our efforts in education. So yeah, we're not helpless here. There's good stuff being done and there's more that we each can do individually and then, you know, in partnership with some of these bigger organizations.
- [71:18]
Mike Kaput
Our next topic is from AI expert Ali Miller, and she recently shared a somewhat disturbing anecdote about where AI might be heading. So she recently posted on X Quote, a famous person wore a secret hidden wire while speaking with me about AI. According to her account, she was at an event backstage talking about someone talking with someone about her private views on AI. And she joked at one point the conversation should have been recorded. The person responded saying they were always miked up. And Miller said, quote, he had a secret mic under his shirt the whole time. No disclosure, no consent, just recording. I felt completely violated. She then pointed out the connection here to AI tools coming down the line. We have things like Meta Smart glasses, AI recording necklaces like Limitless that she says may lead to increased social friction. If they become the norm because they duplicate this very uncomfortable environment, they might erode trust and compel people to be more on their guard. Now, Paul, this is a wild story. It doesn't really, I guess, unfortunately surprise me, but is this the future we're headed towards where everyone records everything?
- [72:29]
Paul Raitzer
Yeah, this is go back to like the conversation around the Limitless pendant and all that stuff. And I was not a big fan of it and I remain not a big fan of it. I do worry a lot about this. I don't know the answer. I mean, just in the last week we saw that Google or Apple is supposedly planning on integrating recording devices, cameras into watches. They're working on glasses. Apparently Meta already has glasses. Google is going to have glasses. I'm guessing that we're actually going to see a demonstration of Google's vision for glasses at IO this week. I think there's a distinct possibility they reintroduced that project into the world. I don't know how we avoid it. I do. Honestly, I find myself uneasy around people wearing Meta glasses already because I have no idea if they're recording what we're saying. And it's not like I'm saying something that I wouldn't want recorded, but I don't want to have to think about it. It, like it's just, you just, you're just having conversations and you, you assume some level of privacy if it's a one on one conversation and that it's not going to end up somewhere. And so I, I, I feel very uneasy about that stuff. And if I saw someone wearing a pendant, I would naturally be guarded around them. I don't know, like, I don't know how this plays out. And I think I'm more conservative than some people when it comes to this stuff. I know a lot of people are like, who cares? Everybody's got their phones out anyway, everybody's recording everything. Like I get it, but it's not the same thing. Like that is, that's a bad counter argument. So yeah, this is a very, very slippery slope. I, I think it's a direction we go. I don't think the current administration does anything from a regulatory standpoint to stop it. I could absolutely see this, though, being a major legal thing, you know, a couple years down the road.
- [74:12]
Mike Kaput
Like a landmark case or two about this.
- [74:15]
Paul Raitzer
Yeah, yeah.
- [74:17]
Mike Kaput
All right, so as we come to the end of today's episode, I am going to quickly run through some AI product and funding updates PA and we're going to kind of have you wrap things up by answering our listener question of the week.
- [74:30]
Paul Raitzer
Cool.
- [74:31]
Mike Kaput
All right, so a bunch of updates this week. First, perplexity. They are reportedly closing in on a $500 million funding round that would value them at $14 billion. That's a huge jump from last year's $3 billion valuation, but notably lower than the 18 billion it initially targeted. Next up, TikTok has launched a new AI powered feature called AI Alive that turns still short animated videos directly inside TikTok stories. This is the platform's first image to video creation tool. Meta is not having a great week. Unfortunately for them, they are quietly pushing back the release of their flagship AI model, Llama 4 behemoth. Originally slated for spring, then bumped to summer, the launch has now been delayed again, likely into the fall or beyond.
- [75:22]
Paul Raitzer
Product.
- [75:24]
Mike Kaput
Generative AI video platform Higsfield AI has released a feature called Higsfield Ads, which turns a single product picture into a studio ad in seconds. You literally just upload a product photo pick from 40 different design templates and the tool will instantly create a cinematic ad spot. In some other video news video generation startup Hedra just raised raised $32 million in a series A round led by Andreessen Horowitz. The company focuses on character driven video, which you may have seen in one instance on display lately. With trends like taking the Internet by storm, like the AI generated videos of talking babies hosting podcasts. Seems like there's at least a lot of money in talking babies hosting podcasts.
- [76:14]
Paul Raitzer
Based on this.
- [76:17]
Mike Kaput
Google DeepMind has unveiled Alpha Evolve, an AI coding agent that designs entirely new algo algorithms. Powered by Gemini models, it will generate thousands of code variations, test them, and evolve the best performers to solve tough problems in math and computing. Interestingly, it's already saved Google Real Compute. It found a scheduling trick that freed up 0.7% of capacity across data centers, which is huge at global scale. And OpenAI just launched a unusual and ambitious initiative. It's called the OpenAI2Z Challenge and it's literally a real world archaeological treasure hunt powered by AI. Participants are invited to search for undiscovered ancient sites hidden deep in the Amazon rainforest using satellite imagery, LiDAR scans, colonial diaries and indigenous oral histories. Entrants must use OpenAI's latest models to analyze open source data, identify likely settlement patterns, and propose novel discoveries. And it kind of centers on finding this legendary lost city of Z, which is rumored to lie somewhere in north Brazil. The winning team gets $250,000 and the opportunity to collaborate with archaeologists to verify their findings in the field. Couple other pieces of news here. Anthropic is about to release two upgraded AI models, Claud Sonnet and Cloud Opus, which will take a bold step forward in how machines think. This seems to be from the early commentary Anthropic sentry into Test Time Compute, which is used by other models on the market to produce deeper reasons by taking more time to think about problems. Now some rivals claim what Anthropic is teasing is not that different from what's already out there today, but Anthropic insists their approach will be more robust and adaptive. And last on my end and Paul, I know you've got an update or two Google I o we've mentioned a couple times the company's developer conference is happening May 20th to 21st, so definitely expect some announcements from Google in coming days, which we will also cover on next week's episode.
- [78:28]
Paul Raitzer
Head on the Baby Trend. Hey Jen, I think this is just yesterday I saw this Heygen announced in a tweet said baby podcasts are everywhere and our intern just made it a feature. Now you can start a pod even if they're in diapers live. Now on hey Jen comment who's these baby hosts remind you of? Was their tweet. So we'll put that tweet in the link. There is a rumor that Mike and I may be soon turned into a baby version of the artificial intelligence show, so stay tuned for that. I'm sure our team team will have some fun with that. And then just a couple of quick notes, the alpha of all things a big deal. Like we're not going to get into it on this show in depth, but it's something we may come back around to. Like I said before, if you go back to AlphaGo and AlphaZero every Alpha fold, it is the distinct advantage that Google has that has yet to be played out into the models we all have access to. But I think if you go study what they've done there and what they're doing with Alpha evolve, you start to see how they may actually have like a very strong competitive advantage over the other labs and then the a To z challenge from OpenAI. Man, I love that. Like, I wish I had like hours free that I could go, I would 100% go play around with that. That is like, I love that there was a lost city of the Monkey God or something. There was a book I read about like how they use lidar to find these. Oh man, I, I could go all day on that stuff. Like space stuff and like hidden, you know, cultures and stuff. It's just so cool. Lost cities and civilizations. I love that stuff. Okay, listener question then we'll wrap up.
- [80:05]
Mike Kaput
All right, so final segment. We're answering listener questions every week. And this week's question is related to the topic of agents which we talk about quite a bit. This one's specifically about how agencies navigate agents. Do you see agencies in the future being mostly agents with a client facing team that runs the agency agents? So what are agents going to do to the agency model?
- [80:29]
Paul Raitzer
I don't know how far in the future we're talking. I think, you know, we have an AI for Agency summit and this year my keynote. So this was November of 24 was the last one, we do it annually and my keynote for that one was, I think it was like AI Agents in the future of Agencies or something like that. So we'll put a link in the show notes. You can go get this on demand. It's the only place that talk is available I think is on demand. And yeah, I mean I think agencies are going to be fundamentally reimagined. In some cases it might be, you know, totally AI native from the ground up and you just build it with as few employees as possible. And if you focus on distinct services like landing page optimization, media buying, like if it's a very specific repetitive tasks, and that's as, as a, as a whole, that's what your agency does is this collection of tasks that are repetitive and data driven and don't have long time horizon, you know, solutions to them. Like, you know, more like consulting work then. Sure. Like I, I could see these distinct, narrowly focused agencies probably being automated pretty quickly like where most of the work is probably done by the agents and you know, say like a two to three year time period. If you're a bigger consulting firm and you know you're doing more strategic work, then it's going to totally change the way you do it through like deep research products and things like that. But you're not eliminating the humans, you just need fewer humans. So I, I think agencies are just going to look fundamentally different in the next few years. Some will be highly automated and some will just be way, way more efficient, create way more outputs, higher quality outputs through the integration of these tools. But yeah, I mean, they're, they're fundamentally going to transform for sure.
- [82:21]
Mike Kaput
All right, Paul, that's a wrap on another busy week in AI. We've got a lot of updates coming down the line, so looking forward to next week as well. Thanks as always.
- [82:29]
Paul Raitzer
Yeah. And we, you know, some memorial days, the following Monday, we, we'll have a regular episode. Yeah, so we'll do the regular weekly and we'll cover all the model updates and app news and all that good stuff. So thanks, Mike. And again, thanks. As we started the show off like Mike and I are truly grateful to have you all listening each week. We love hearing from you on LinkedIn or in person. And so just thanks for listening and being a part of this growing community. 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.