9.7 C
New York

Top Consulting Firms Bet Big on AI, Why AI Isn’t Making Money, and Zoom CEO’s Bizarre Plan for AI Clones in Meetings

Join Paul Roetzer and Mike Kaput as they discuss the surge in demand for AI consultants from major firms like BCG and McKinsey, explore the growing disconnect between AI investments and economic impact, and examine Zoom CEO’s Eric Yuan’s ambitious plans for AI-driven digital twins in Zoom meetings.
Listen Now

Watch the Video

Timestamps
00:03:31 — AI Consultants Success from the AI Boom
00:16:04 — AI Needs to Make Money
00:30:47 — Zoom’s CEO on the Future of Work
00:41:00 — Google’s Gemma 2
00:46:12 — OpenAI’s 2023 Security Breach Raises Concerns
00:49:50 — Apple’s OpenAI Board Seat
00:52:28 — AI and The Olympics
00:56:47 — Nintendo’s Future Plans with GenAI
01:00:00 — Finding GPT-4’s Mistakes with Critic GPT
01:03:41 — Scale AI’s Business Model
01:09:41 — AI Tech Updates: Runway
01:11:57 — AI Tech Updates: Hebia
01:14:01 — AI Tech Updates: Perplexity
01:16:00 — AI Tech Updates: ElevenLabs
Summary
AI Consultants 
The artificial intelligence boom has created surprising revenue for management consulting firms. As businesses grapple with how to integrate and leverage generative AI, they’re turning to consultants for guidance and expertise.
Firms like Boston Consulting Group, McKinsey, IBM, and Accenture are seeing surging demand for their AI-related services. BCG now derives a fifth of its revenue from AI work, up from zero just two years ago. McKinsey expects 40% of its business this year to be AI-related. IBM has secured over $1 billion in AI sales commitments, while Accenture booked $300 million in AI sales last year.
While AI consulting is proving lucrative, it is not without challenges. Consultants and their clients are learning and adjusting as they go, says The New York Times, navigating the rapidly evolving landscape of AI capabilities and limitations.
AI Needs to Make Money
The AI industry is facing a significant disconnect between massive investments and actual economic impact, as evidenced by a few big stories coming out this week.
First, David Cahn of Sequoia Capital has updated his analysis on the AI industry’s revenue expectations, originally published as “AI’s $200B Question” in September 2023.  The new analysis, titled “AI’s $600B Question,” reveals a widening gap between AI infrastructure investments and actual revenue growth in the AI ecosystem.
The revenue gap has grown significantly, from a $125 billion “hole” to a $500 billion hole that needs to be filled by revenue growth annually to justify current capital expenditure levels.
Meanwhile, a report from the US Census Bureau claims that AI adoption rates remain surprisingly low. The Bureau reports only 5.4% of businesses are currently using AI as of February 2024, with slow growth projected.
The Economist also sees a significant disconnect between the hype surrounding AI and its actual economic impact. Their reporting says that many companies are still experimenting with AI or hesitant to invest heavily in it, and that, so far, there’s little evidence of AI boosting productivity or transforming business operations on a large scale.
Zoom CEO wants AI clones in meetings
Eric Yuan, Zoom’s founder and CEO, shared an ambitious vision for the future of work and video conferencing powered by AI in a wide-ranging interview on The Verge’s Decoder podcast.
In the interview, Yuan envisions a world where people can send “digital twins”—AI-powered avatars of themselves—to attend meetings and make decisions on their behalf.
Said Yuan:
“Let’s say the team is waiting for the CEO to make a decision or maybe some meaningful conversation, my digital twin really can represent me and also can be part of the decision making process.”
However, he says this future is “years” away. He also believes that eventually, everyone will have personalized large language models that deeply understand them and can act as their representatives. This AI-driven future, according to Yuan, could even lead to shorter work weeks as artificial intelligence handles more tasks.
He also teased visions of Zoom’s future, more as a comprehensive AI-powered workplace collaboration platform than just a video conferencing tool.
Overall, he paints a picture of AI fundamentally changing how we work and communicate in the coming years, with Zoom aiming to be at the forefront of this transition.
Links Referenced in the Show

AI Consultants:

AI isn’t making money

Zoom CEO wants AI clones in meetings

Gemma 2 is now available

OpenAI Hack

Apple’s OpenAI Board Seat

AI and The Olympics

Nintendo and GenAI

Scale AI’s Business Model

Finding GPT-4’s mistakes

AI Tech Updates

Runway

Hebbia

Perplexity Pro Search

ElevenLabs Iconic Voices

ElevenLabs Voice Isolator

Today’s episode is also brought to you by the Marketing AI Conference (MAICON), presented by Marketing AI Institute. MAICON is an industry-leading event that helps marketers and business leaders at all levels understand, pilot, and scale AI.
This year will be our biggest conference yet—we are expecting 1,500+ AI-forward professionals to attend and we’ve got tons of inspiring keynotes, actionable breakouts, and valuable networking opportunities on the agenda.
Don’t miss this chance to accelerate your AI journey and stay ahead of the curve. Go to www.MAICON.ai to learn more.
Read the Transcription
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
[00:00:00] Paul Roetzer: And time again, we find that people are fearful they’ve fallen behind, that they’re not doing enough with AI in their company. They’re not moving fast enough. And what I can tell you is it’s more likely than not, that’s the norm.
[00:00:14] Paul Roetzer: Welcome to the Artificial Intelligence Show, the podcast that helps your business grow smarter by making AI approachable and actionable. My name is Paul Roetzer. I’m the founder and CEO of Marketing AI Institute, and I’m your host. Each week, I’m joined by my co host. and Marketing AI Institute Chief Content Officer, Mike Kaput, as we break down all the AI news that matters and give you insights and perspectives that you can use to advance your company and your career.
[00:00:44] Paul Roetzer: Join us as we accelerate AI literacy for all.
[00:00:52] Paul Roetzer: Welcome to Episode 104 of the Artificial Intelligence Show. I am your host, Paul Roetzer, along with my co host, Mike Kaput. We are [00:01:00] coming to well, it is Monday, July 8th at 9 a. m. Eastern Time is when we’re recording this. So, It was a, it was an interesting week last week. I feel like we didn’t like have like all the craziness of the two prior weeks while we were away,
[00:01:16] Mike Kaput: Yeah.
[00:01:17] Paul Roetzer: there’s some big macro level stuff we’re going to get into today.
[00:01:20] Paul Roetzer: So we’re going to talk about funding, product updates, the usual stuff. But there’s some, some big topics, I think, related to where is all the money in AI going? We got a lot of money being spent on NVIDIA chips, but, um. what is the value of those at the end user level? What, where are the companies that are making the money?
[00:01:38] Paul Roetzer: Got a fascinating. Look into how the CEO of Zoom thinks about the future of work, which I’m still trying to comprehend. So, lots of good stuff today. Not the breaking news kind of week, but just a ton of big picture things to week. So episode 104 is brought to us by the Marketing AI [00:02:00] Conference or MAICON.
[00:02:01] Paul Roetzer: This is our fifth annual Marketing AI Conference presented by marketing AI Institute. the event is gonna draw probably close to 1500 people to Cleveland this year. So if you haven’t been to it, definitely take a look. It’s MAICON.ai, M-A-I-C-O-N ai. There’s gonna I probably close to 40 sessions.
[00:02:21] Paul Roetzer: so the way we’ve got it broken out, I’m still finalizing the agenda. I have to lock in the cut, the last few, main stage general session presenters, which I’m very excited about. The prospects there. But the rest of the agenda is locked in. So we’re about 90 percent or so done with the agenda.
[00:02:37] Paul Roetzer: There’s going to be your general mainstayed sessions, the feature kind of macro level talks, big picture stuff. And then there’s going to be a strategic leader track and an applied AI track. The strategic leader track is all about. org charts, strategy, resource allocation, change management, technology, like big picture, things for leaders of organizations.
[00:02:58] Paul Roetzer: And then the applied AI is going to be all about [00:03:00] use cases, technologies, demos, case studies, really giving you everything you need to know to Kind of immediately apply everything you’re learning. So really excited about this year’s event. It is September 10th to the 12th in Cleveland. You can still get discounted pricing.
[00:03:16] Paul Roetzer: the rates go up at the end of each month, so you can get in now and get the best pricing that’s available. Again, that’s maicon. ai to learn more. All right, Mike, let’s, dive into the week of some big picture ideas and thinking.
[00:03:31] AI Consultants
[00:03:31] Mike Kaput: All right,
[00:03:32] Mike Kaput: Paul. So first up, we are seeing that the artificial intelligence boom is creating kind of an unexpected windfall for consulting firms.
[00:03:43] Mike Kaput: So, as companies are kind of grappling with how to integrate and leverage, specifically generative AI, they’re turning to consultants quite a bit for guidance and expertise. So we’re seeing firms like Boston Consulting Group, McKinsey, IBM, Accenture, and others [00:04:00] are seeing really like a surging demand for AI related services.
[00:04:04] Mike Kaput: And according to some recent reporting, we’ve learned that BCG, Boston Consulting Group, now derives a fifth of its revenue from up from zero just a couple years ago. McKinsey
[00:04:17] Mike Kaput: expects
[00:04:18] Mike Kaput: 40 its business this be AI related. IBM has secured over 1 AI sales and Accenture, which we have mentioned before, booked 300 sales last
[00:04:33] Mike Kaput: year.
[00:04:33] Mike Kaput: So, there’s not a ton of specific details on exactly what services they’re offering, but some of the ones some recent reporting Say that the work is diverse and kind of totally depends the customer. things like regulatory compliance, AI
[00:04:49] Mike Kaput: for customer
[00:04:50] Mike Kaput: support, establishing guardrails for AI use, and much more.
[00:04:55] Mike Kaput: while this consulting appears to be proving lucrative, it does have some [00:05:00] mean, as we all know, AI is prone to errors and hallucination. It requires a lot of oversight. Constant and tweaking to get it to work right, and consultants and
[00:05:09] Mike Kaput: their clients
[00:05:10] Mike Kaput: are learning and adjusting as they to the New York Times, as they navigate kind of how quickly AI capabilities and limitations are evolving.
[00:05:20] Mike Kaput: So, to kind of kick things off here, Paul, I wanted to maybe get a sense from you of like, opportunity here? It sounds like so many businesses need Some type of consulting or advice or guidance on how to actually use AI.
[00:05:36] Paul Roetzer: Yeah, I mean, we certainly see it ourselves. And so we did talk about, I think you, you referenced this, but so episode 91, if you want to kind of backtrack and go deeper on this, if you’re in the consulting business or you have an agency, or maybe you’re on the brand side looking for consultants. So episode 91, we talked about Accenture and their massive growth from generative AI consulting And then episode, what is this, [00:06:00] 81, we talked about Publis’s group and their kind of vision for where they were going to go with AI consulting.
[00:06:05] Paul Roetzer: So we, again, we, we, see it, we hear it every day. I mean, we talk with big enterprises every day that are looking for guidance. Um. Trying to figure out their roadmaps for generative AI, trying to put policies in place, change management, that comes around to education and training. These are all needs of basically every company we talk to, and, you know, it obviously is a massive opportunity for consultants that figure out to, you know, support in a value based way in these areas, the things that I find so challenging right now.
[00:06:41] Paul Roetzer: So like, I mean, just this week, I have three talks with big enterprises. Every one of them is unique, but they’re all dealing with the same challenges. So different industries, But every time you get on a call and you start talking to these big enterprises, you, you, the things I’m always interested in learning right away is, is there an [00:07:00] AI council already?
[00:07:01] Paul Roetzer: If so, who is on that council? Is it being run by the technologists? Is it the CIO, CTO? Does marketing, sales, service have a voice in this council? So you’re trying to get a general sense of like, who is even involved in AI right now in your You to learn about, do you have generative AI policies?
[00:07:19] Paul Roetzer: Do your employees know the generative AI policies exist? Have they been trained in those generative policies? You want to know, are your employees even allowed to use generative
[00:07:28] Paul Roetzer: AI?
[00:07:29] Paul Roetzer: How, how are they allowed to use it? Is it a use case based? They have to come for permission for every use. Is it platform based?
[00:07:35] Paul Roetzer: Like they have to get permission to use OpenAI, you know, ChatGPT license. Do you have Copilot or Gemini? Are those licenses distributed? Like, No organization is the same right now. And Mike, you’ve done this too. Like you go and you, you have these conversations, we go in and run workshops, we do talks, you meet with the executive team.
[00:07:54] Paul Roetzer: And I just think that there’s this tendency to overestimate how far [00:08:00] along companies are in this, because again, having talked with dozens of these, you know, fortune 100 they’re not that far along. And in many cases, it, I think it’s. It kind of comes down to a bottleneck where the technologists in the company are controlling the AI, what people are allowed and not allowed to do, and so it’s not decentralized where people have the freedom to kind of go and experiment.
[00:08:26] Paul Roetzer: and then I just think there’s just a lack of AI savvy talent within the organizations at the level. So, I’m a big proponent, and I know Mike is. You and I’ve talked about this before. I don’t know that the CTO, CIO should be running the AI strategy in a company. Like they obviously have to be intimately involved.
[00:08:45] Paul Roetzer: They have to have a massive voice in it, but I think that by allowing the technology arms, the information technology arms of the companies to control AI and the roadmap and where it goes. It just [00:09:00] unnecessarily slows things down because their number one priority isn’t finding business use cases for this stuff.
[00:09:05] Paul Roetzer: They’re not proactively going to be seeking the ways to drive productivity and efficiency and innovation where like a marketing or a sales or a customer service may be focused primarily on those Um, the other side has to focus on, yes, those outcomes are important, but not at the risk of, you know, data and privacy and regulations.
[00:09:29] Paul Roetzer: And so I think that you just need all those voices in the same room. So from a consultant perspective, they need help figuring this all out. They need the third party, the objective third party to come in and work with the different stakeholders in the organization, balance the need to. Adhere to guidelines and regulations and protect data and privacy with the opportunity to innovate and re imagine what’s possible in the business.
[00:09:54] Paul Roetzer: And we have to have both sides of those conversations happening simultaneously. So I [00:10:00] think the The consultants are needed many times to come in and figure out what is just our large strategy. What, what platform are we using? What’s, what are the priority use cases that we can get value from right how can we look beyond efficiency and productivity into innovation, new markets, new ideas, new products? what do we do with change management? What do we do with education and training? Like the, it’s very hard for existing. teams within organizations that all have full time jobs to figure all of this out when they themselves likely have no formal training in the deep understanding of AI.
[00:10:38] Paul Roetzer: So, I don’t know, I mean, that’s kind of at a high level how I think about this. Do you see anything different, Mike? I mean, again, you’re having a lot of these same conversations I
[00:10:47] Paul Roetzer: am.
[00:10:47] Mike Kaput: No, I was gonna say, I think, you know, I’m not
[00:10:50] Mike Kaput: deeply,
[00:10:51] Mike Kaput: deeply familiar with kind of the big four consulting firms, but I think they sometimes consulting firms of this caliber get some flack because [00:11:00] it’s like, oh, okay, consultants to come in and tell us what we already know about our business.
[00:11:03] Mike Kaput: That’s perfectly valid. I’m sure that happens, but I do really see such a need here for. Experts showing people what’s even possible in the first place, because there really is this huge, awareness gap of not only how to do what you suspect AI you, but even thinking about all the opportunities different roles, the impact on you. Different careers, career paths, jobs, and overall
[00:11:32] Mike Kaput: business transformation.
[00:11:33] Mike Kaput: So I like a rabbit need here for any type of guidance. I mean, we see it every workshop, every talk we do, people are. So desperate for just like some guiding light through some of these processes.
[00:11:46] Paul Roetzer: Yeah, one of the, one of the thoughts I always have is like. When I talk to these big enterprises, I feel like there’s this paralysis where they just don’t move forward because they’re trying to solve The [00:12:00] issues. They’re trying to solve the major platforms. They’re trying to solve all of these sort of macro level things that touch every part of the organization and they lose sight of how much value can be gained by unlocking single use cases within teams or within individuals.
[00:12:16] Paul Roetzer: And so. And that’s why one of our, you know, most popular workshops, the Applied AI Workshop we run, is literally just get in there for three hours and teach people how to identify use cases so they can unlock an hour here, three hours there, five hours there. And so, when you drill into the very specifics and look at actual use cases, and ideally,
[00:12:35] Paul Roetzer: What
[00:12:35] Paul Roetzer: we always tell people is find the ones that the legal team, the IT team aren’t worried about.
[00:12:41] Paul Roetzer: Like very base level things, like the one we always come back to is Descript for the podcast. Everything we do on the podcast is already publicly available. There’s nothing we say here that can’t then be put into an LLM and turned into a summary or cut up into shorts for YouTube and TikTok because it’s all [00:13:00] public data.
[00:13:00] Paul Roetzer: So what
[00:13:01] Paul Roetzer: would possibly be the argument to preventing the use of that kind of technology within any brand that creates webinars and podcasts? None. Like, there’s no argument against it. So, I feel like so many times it just comes, the lack of advancement within enterprises comes from a lack of understanding that there are thousands of cases. Many of them that are totally innocuous. There’s no threat whatsoever on data leakage, privacy concerns, coming up against regulations. Ethical issues. It, you just have to be able to identify those use cases. And if you only did that, and then you just started stacking that, and let’s say each person on a marketing team, as an example, could save 10 percent of their time.
[00:13:47] Paul Roetzer: Well, if you’re in a big enterprise, that’s a massive amount of time savings. And so I feel like that’s, what’s needed. And maybe that’s where consultants come in is just to come in and bring the reality, like we don’t have to do. Everything all at once. We don’t [00:14:00] have to solve AI at a macro level across the whole organization before we do something.
[00:14:05] Paul Roetzer: And if people just had more of a, I guess took more of an initiative to just start identifying single use cases where they can make an impact and then stacking those use cases, we would get way further ahead and we wouldn’t be having these debates about is AI creating value. I mean, it’s a ridiculous thing.
[00:14:22] 
[00:14:22] Paul Roetzer: Um, Because again, like everyone you talk to, if you can explain these very simple things, it’s like, Oh yeah, that would save me five hours a month. Great. Let’s find three other ones like that. And now we just transform your job. Like now you can go do all the things that are on your wishlist that you’re not getting done.
[00:14:37] Paul Roetzer: So. Yeah, I think the more consultants can do that and the more they can verticalize it, because I think the other thing that’s going to be needed is specialization into industries, you know, like law and accounting and HR, and then you start getting into, you know, like retail and e commerce. Like, we need experts in all these areas that can go in and do this kind of education and consulting
[00:14:58] Mike Kaput: Yeah, I [00:15:00] mean, let me know if you agree. I mean, every time we’ve done a workshop recently or been in front of some of these enterprises, it really just does seem like some of have to be rocket science. Like, we have very smart people in the rooms. It’s not like they don’t know their stuff. And yet going through and showing what’s possible with
[00:15:16] Mike Kaput: a few
[00:15:16] Mike Kaput: simple things, you see people’s eyes light and they
[00:15:18] Mike Kaput: get really excited because they start see, oh, this is actually doable.
[00:15:23] Paul Roetzer: Yeah, that’s why our approach is to empower people to do this, like give them the frameworks and the knowledge so they can go do it themselves. That’s why, you know, we created the Scaling AI series. It’s just like here, like once you learn these frameworks, you can go do this yourself. You don’t need to spend a million dollars with consultants.
[00:15:37] Paul Roetzer: You can, if you want to, if that’s what your company requires as a third party to come in and say the same thing you’re going to say, but they’re They’re going to believe them because they’re from an outside consultant. Um, we just want to empower people to have the ability to move faster, to accelerate adoption, responsible use of AI.
[00:15:54] Paul Roetzer: And so that’s the whole premise behind the Scaling AI series, behind what we’re building with SmarterX. [00:16:00] it’s really all about empowering leaders to take the initiative here and start creating more value.
[00:16:04] AI Needs to Make Money
[00:16:04] Mike Kaput: All right. So our second big topic this week is quite complimentary. Um, we’re seeing the AI industry kind of facing what some are calling kind of a significant disconnect between the massive investments in companies and technology and then the actual economic impact.
[00:16:22] Mike Kaput: So we’re seeing a few stories that are Getting a lot of popularity idea. So,
[00:16:28] Mike Kaput: first up, David Kahn at Sequoia Capital, a major VC firm, has actually updated an existing analysis that says, that he does on the AI industry’s revenue expectations. So this was an article that was in September 2023.
[00:16:44] Mike Kaput: Titled AI’s 200 question. The update, the new AI’s
[00:16:51] Mike Kaput: 600 billion question. And it basically reveals that there’s this big between AI infrastructure investments and the actual revenue growth [00:17:00] happening in the AI ecosystem. He is saying that based on his numbers, the revenue gap has grown significantly from essentially a.
[00:17:07] Mike Kaput: 1. 125 billion dollar hole, i. e. what revenue kind of needs to be made up to annually to the current capital expenditures, that is increased to a 500 billion dollar filled by revenue growth. Meanwhile, at the same time, we were reading kind of an in depth report the disconnect between the hype and around and its actual economic impact in The Economist.
[00:17:35] Mike Kaput: And what’s really interesting is their reporting is saying that many companies are still experimenting with AI or invest in it heavily, and that so far there’s little evidence of AI boosting productivity or on a large scale. Now, As part of that reporting, they cite a pretty interesting report from the U.
[00:17:57] Mike Kaput: S. Census Bureau that claims [00:18:00] AI adoption rates based on a bunch of their survey data remain surprisingly low. The Bureau reports that only 5. 4 percent of businesses in their books are currently using AI, and that’s as of February 2024. They do project that number to raise a bit about a percentage point by the fall, pretty low based on kind of what we’ve
[00:18:23] Mike Kaput: been talking about. So Paul, I want to kind of maybe dive in to this from a couple angles. So like, what
[00:18:31] Mike Kaput: do you Disconnect here that people are reporting. I mean, Sequoia is reporting this huge 500 billion hole, you know, even despite companies like OpenAI making up to
[00:18:41] Mike Kaput: now 3 billion annually, it just seems like there’s this huge maybe the census report you could touch on just with this. This seems very low for the adoption rates here. Like, what are you making of These kinds of findings and commentary.
[00:18:55] Paul Roetzer: Yeah, I wasn’t sure on the Sequoia one initially, like I wasn’t sure I was buying the premise when I, when I first [00:19:00] read it, and then I went back and read the original post, so the 200 billion one, where the argument was made a little more clearly, I think, or at least the foundation of the argument was made, and so the basic premise of the Sequoia one, which I generally agree with.
[00:19:12] Paul Roetzer: They’re actually looking at how much money is being spent with NVIDIA, and then how much of that is leading to actual revenue. Now, I think they’re, you know, they’re looking at Microsoft revenue, and they talk about Google revenue. They’re largely looking at product revenue. They’re not really factoring in, like, cost savings, innovation, productivity.
[00:19:30] Paul Roetzer: But that all being aside, in essence, It’s more of an investing thesis. Like they’re definitely writing this from a VC perspective. And what they’re saying is all this money is going into infrastructure, specifically in GPUs to, to create AI, but where is the end customer value? So it specifically says. How much of the capital expense build out is linked to true end customer demand and how much of it is being built in anticipation of future customer demand? This is the 200 billion [00:20:00] question. So that was from the original one. And I generally think like, that’s a, that’s a valid question. And I, cause we do know like Tesla as an example is buying hundreds of thousands of GPUs, not for today. Like it’s not generating revenue today. It’s to build this vision for robo taxis.
[00:20:16] Paul Roetzer: So five years from now. They’re making billions of dollars a year just on the RoboTaxes. So, that’s kind of the premise here. OpenAI and Microsoft, they’re building massive data centers. They’re buying tons of GPUs on the promise of AI. And so,
[00:20:32] Paul Roetzer: it makes a lot of sense, especially under the context of recent conversations we’ve had around ADEPT, which just last week we talked about, like, Not gone, but basically the team’s getting absorbed into Amazon, or AWS.
[00:20:46] Paul Roetzer: it happened with inflection, where they’d spent, you know, how many hundreds of millions of dollars on GPUs to make no money, and the team basically leaves and goes to Microsoft, and now inflection’s left being whatever it is today. Stability, same thing. So, [00:21:00] we are certainly seeing hundreds of millions of dollars, if not billions, spent on GPUs from NVIDIA, that aren’t turning in to near term revenue.
[00:21:09] Paul Roetzer: It’s all being done on this promise of eventual consumer demand. So, basically what they’re saying in the Sequoia One is, we’re following the GPUs. Like, we’re going to find the startups that are building for tomorrow, but are actually solving real world issues, creating real world value. And so they’re on a path to try and find these companies that can match, Their infrastructure investment with a product and a company that is going to actually create end user value.
[00:21:38] Paul Roetzer: So, from that premise, it makes a bunch of sense. Now, the Census Bureau one, this, this makes for really good headlines.
[00:21:45] Paul Roetzer: Like, I don’t know. like kind of shocked by this one. So, and we didn’t, so this came out in March of 24, but we didn’t, it didn’t surface for us until we saw it referenced in the Economist article last week.
[00:21:58] Paul Roetzer: So their [00:22:00] basic premise is, adoption rates are insanely low. 3, 3. 7%, I think is roughly what they’re saying. They talk about, you know, different use cases for, Marketing automation, virtual assistants. But when I first looked at it, I was like, man, this is a huge sample size. Like maybe it’s Microsoft and LinkedIn and everybody else that’s getting it wrong.
[00:22:20] Paul Roetzer: And the Census Bureau out. So they say that they,
[00:22:24] Paul Roetzer: um, they, Okay,
[00:22:26] Paul Roetzer: So 1. 2 million employer businesses. So massive. So in the United States, just for context, there’s about 24 million businesses, just. So people have a general concept here. so they surveyed 164, 000 businesses and it’s like, wow, that’s the biggest survey size I’ve ever, I’ve ever seen related to AI.
[00:22:43] Paul Roetzer: So that,
[00:22:44] Paul Roetzer: like, this is worth paying attention you digging into the report and I start realizing like, I don’t think this report’s very valid at all. Like,
[00:22:53] 
[00:22:53] Paul Roetzer: so they said there’s an enormous variation in current use by sector from a low of 1. 4 percent [00:23:00] in construction and agriculture to a high of 18 percent in information, information technology. So, you know, I start really questioning The process here, it’s like, okay, I get that you surveyed 164, 000 businesses, which is significant, but what did you ask them to arrive at this? And who, who answered the question? So I want to step back and say. We’re going to talk about this survey for a moment, because I important, and it’s probably been, you know, mainstream media runs with these headlines, but two, I want this to be a lesson in critically analyzing any research data you see, like everyone has Their own reasons for doing research.
[00:23:41] Paul Roetzer: And even when we do our own state of marketing AI, we are very clear up front. This is biased.
[00:23:45] Paul Roetzer: like the people
[00:23:46] Paul Roetzer: answering this are more likely than others to already be adopting AI because they subscribe to the Marketing AI Institute. So it is likely their answers are. the norm. So you have to [00:24:00] always analyze how was the survey collected, who answered it, and what did they ask them?
[00:24:04] Paul Roetzer: And when you go to the, what did they ask them? This report immediately starts to sort of lose me. So they asked two core questions of these 164, 000 companies of which I don’t even know who they were asking. They specifically at one point said, yeah, some of the people responding are in finance, so they might not know AI company.
[00:24:24] Paul Roetzer: It’s like, oh, well, that’s,
[00:24:25] Paul Roetzer: That’s pretty important. How many of them are the finance people and the HR people? but here’s the question between month day to month day. So whatever the time period is, did this business use artificial intelligence in producing goods services? Examples of AI, machine learning, natural language processing, virtual agents, voice recognition, et cetera.
[00:24:46] Paul Roetzer: So the first question is, did you use AI? To produce goods and services. That’s not the same as AI adoption, because that is specifically saying, did you make a product or a service with [00:25:00] AI? It’s not saying, are you using it in accounting or using it in HR? Are you using it in your marketing? It’s saying, are you producing a good or a service with it?
[00:25:08] Paul Roetzer: So I interpret that to is it part of the process to create the good?
[00:25:13] 
[00:25:13] Paul Roetzer: and so I could easily say, well, no, but I’m not sure that you’re asking about my marketing, because it doesn’t say about promoting, marketing, marketing, marketing, planning. Performance management doesn’t say anything. It’s just about production of good.
[00:25:25] Paul Roetzer: So you could already bias this data by people just not being sure what they’re asking. The second is, you’re asking people who don’t know what artificial intelligence So if ask a CEO of a farm and ag business,
[00:25:37] Paul Roetzer: Are you using AI? And maybe they don’t know. Or you’re asking the CFO of a law firm who has heard about it, but doesn’t really know what it is.
[00:25:47] Paul Roetzer: Thanks. So I immediately like, okay, this, this data useless, like, it’s interesting. there’s probably something to be learned from it, but because they didn’t ask [00:26:00] people who necessarily understand what AI is, and they specifically called it production of goods and services. It actually isn’t valid to say that the adoption is 1.
[00:26:09] Paul Roetzer: 7%. They it’s a very narrow, and they even then call out, hey, there’s challenges in how we did this based on who answered. Some people are involved in finance and accounting or unfamiliar with technical plans for the company. most, they actually say the most common challenge presented during cognitive testing surrounded the definition of, and what as AI.
[00:26:32] Paul Roetzer: Now,
[00:26:32] Paul Roetzer: I did not
[00:26:32] Paul Roetzer: click through and see what the Census Bureau considers AI, but I can tell you if they said they were using it, they then asked 13 follow up questions of which they gave them, they said, are you using deep learning, natural language processing, neural networks, again, do you think the CFO knows what that stuff is?
[00:26:53] Paul Roetzer: So, I, again, I’ll, I’ll just say that overall, I wouldn’t spend time reading [00:27:00] this report. I think that it’s important to realize that some of us get caught in reading the Microsoft LinkedIn data, and like, we think that that’s representative of all companies. That is not true. Somewhere between the Census Bureau data and the, you know, the tech industry data is probably the reality.
[00:27:18] Paul Roetzer: But, these numbers are, I true. Extremely artificially low to the probably within the broader industry.
[00:27:28] Mike Kaput: Yeah. the moment I. dug into the questions they asked, I immediately had alarm bells going off just because I know if we had asked these questions, for instance, in workshops, consulting engagements, talks that we do, you would get eight different answers from eight different people, depending on their title, where they are in the room and their understanding.
[00:27:48] Paul Roetzer: Yeah. And I think, you know, this goes back to my frustration with a lot research reports, even from some of the big
[00:27:54] Paul Roetzer: consulting
[00:27:55] Mike Kaput: Yeah.
[00:27:56] Paul Roetzer: where they’re asking. CEOs sometimes or other [00:28:00] C suite executives about AI and those people may not be involved in it at all. And so like, what, what good is a report about what CEOs think is going to happen over the next 18 months with AI and the impact it’s going to have on the workforce if those CEOs don’t their own AI strategy or even more broadly, all the elements of AI.
[00:28:23] Paul Roetzer: So again, it’s just, this is more of just be cautious, be, be, critical in your thinking when you see research data, especially when you see headlines in, in mainstream media or Twitter threads that just seem a little bit off. Like they usually are, like there’s usually a reason your spidey senses go off that this doesn’t seem like research that I should base my business decisions on.
[00:28:48] Mike Kaput: So
[00:28:48] Mike Kaput: to kind of wrap this up, it sounds like, you know, these reports and these research papers are getting at kind of a broad idea of like, okay, there, there is [00:29:00] investment that’s happening that is not yet seeing immediate returns on the consumer side of the business.
[00:29:05] Mike Kaput: But the murkier might think, and nuanced, I would say. Is that
[00:29:11] Paul Roetzer: Definitely. And I think the key takeaway for a lot of listeners here is, time and time again, we find that people are fearful they’ve fallen behind, that they’re not doing enough with AI in their company. They’re not moving fast enough. And what I can tell you is it’s more likely than not, that’s the norm.
[00:29:34] Paul Roetzer: So while the Census Bureau survey, like I think it’s ridiculous, like only 1. 7 percent are doing anything with AI. I think it’s probably within reason that like 5 percent or less of companies have an actual AI Roadmap, change management plan, internal academy, AI council, like, are really doing it, are really scaling AI.
[00:29:59] Paul Roetzer: [00:30:00] Um, that is a very low percentage based on our and Piloting AI different. I think that’s a much higher number where people or teams or departments within organizations are experimenting with it, trying to figure it out, testing a technology, testing a cases. That is prevalent.
[00:30:21] Paul Roetzer: Like, we have definitely entered the stage where piloting of it and experimenting with it is much more common. It’s just often not in some formal structure where the learning is being shared across departments and being pushed up to a council. So, piloting AI think, is way more prevalent. Scaling AI is still Probably much closer to the numbers they’re where it’s truly across all and structured in a formal
[00:30:47] Zoom CEO on Future of Work
[00:30:47] Mike Kaput: All right. And our third big topic today, Eric Yuan, who is Zoom’s founder and CEO, he recently shared, it an ambitious vision for the future of work [00:31:00] and powered
[00:31:03] Mike Kaput: AI. And this came from a wide ranging interview he gave to the Verge’s podcast. So in this interview, Yuan kind of gets a little sci fi, envisioning a world where people quote, digital powered of themselves, to attend make decisions on their behalf.
[00:31:22] Mike Kaput: He said, quote, let’s say the team is waiting for the CEO to make a decision meaningful conversation.
[00:31:29] Mike Kaput: My digital twin really can represent me and decision making process. He does say this kind of future he’s seeing is years away. But he does also believe that eventually, everyone’s going to have kind of personalized
[00:31:44] Mike Kaput: large language models that basically deeply understand them and also act as their representatives at work.
[00:31:51] Mike Kaput: Now, of says that this AI driven, collaborative, AI powered meeting future could shorter [00:32:00] work weeks as handles more tasks. He also talked quite a bit about Zoom’s, future and the vision he has there, seeing the platform more as a comprehensive
[00:32:10] Mike Kaput: AI powered workspace collaboration tool, rather than just video conferencing.
[00:32:17] Mike Kaput: So Paul, you kind of had flagged this interview as one really interesting that you had listened to, and I’d be curious about your thoughts on kind of Yuan’s like core claims here,
[00:32:26] Mike Kaput: like digital twins. Certainly sounds plausible. do you think it’s likely?
[00:32:31] Paul Roetzer: This is one of the more bizarre interviews I have ever listened to. Still, I’ve never listened to this guy talk before, I don’t think. I don’t think I’ve heard an it, it got weird really fast, like, and Nila, I, you know, he did a great job with the interview. Like, he was, he was really pushing him on, like, well, how is this all gonna happen?
[00:32:53] Paul Roetzer: Like, you’re, you’re, talking about, like, this very Not far off world, like three to six [00:33:00] years, where we’re just going to send AI avatars of ourselves to meetings and that, that avatar is going to have the ability to make decisions. Like what, how would that work? Like how, what are the guardrails? and so like, Nela kept pushing him on like all of these, but what about this?
[00:33:16] Paul Roetzer: And how’s this going? And he had no answers to any of it. Like, it was, it was So, the part that I found so bizarre, one, a guy who runs Zoom hates meetings, like, it is very, very clear he despises meetings. doesn’t want to ever be in them, doesn’t like email, doesn’t like, basically, like, he very, you know, specifically he says, like, we should just be with our friends and family, which I don’t disagree with, like, I mean, there’s There’s the spirit of some of what he was saying, that we do too many repetitive tasks, that we should free ourselves up.
[00:33:50] Paul Roetzer: We shouldn’t have to go to all these meetings. We, maybe we don’t even need to work five day weeks. Maybe it’s three day weeks and we shouldn’t have to do emails and we shouldn’t have to do all these tasks. Like, I get that. Like there is [00:34:00] a spirit of that, but it just comes across as someone who really doesn’t like work at all, specifically meetings, which his platform enables.
[00:34:08] Paul Roetzer: And so he has this vision for the future that seems quite disconnected from. Reality where we’re just not going to do any of it. Like the AI is, I think at one point you’ve said like, like 90 percent of what we do, the AI is just going to do. And Neil was like, well, but that we have a lot of project managers that listen to our podcasts.
[00:34:26] Paul Roetzer: You’re saying they’re not going to have any jobs. He’s like, yeah, yeah, basically. So. I think that I don’t want to live in the future he envisions, which made me actually quite worried because we’re huge we use the platform for webinars, for meetings, for chat, for everything. and I don’t buy into the future this guy’s seeing.
[00:34:47] Paul Roetzer: Like, I don’t, I don’t think it’s, A good future. Like I don’t want my AI showing up and making decisions and I don’t wanna meet with Mike’s AI and have Mike’s AI meeting, making
[00:34:56] Paul Roetzer: decisions.
[00:34:58] Paul Roetzer: and so it was kind of [00:35:00] like, well, when do you actually show up? Like when do people get the privilege of your human form coming to meetings?
[00:35:07] Paul Roetzer: And it was like, eh, not much. Like my AI me. So I just felt like. It was a very uncomfortable vision future from a company that plays a key role in the future of work. And the thing I found most striking is as the CEO, He apparently had zero plan for how this was actually all going to happen.
[00:35:32] Paul Roetzer: And they don’t plan to necessarily build any of it themselves. It was almost like, I think AI is going to be able to do this stuff and we’ll enable it within our platform. And now if you want to know how we’ll do it, or like how many versions of the AI avatar there will be, or what models it will be built on, or how we’ll make it secure, how we’ll know to trust its decision making ability, like, eh, We’ll, we’ll worry about that in like three, four years when this starts becoming reality.
[00:35:58] Paul Roetzer: It’s like, no, you’re [00:36:00] presenting a future that is quite different from the present. You should have some form of an idea of how that’s going to happen and what the impact is going to be on your own employees. Like, could you imagine, Mike, if I sat here
[00:36:12] Paul Roetzer: and said, yeah,
[00:36:13] Paul Roetzer: yeah. So in like three years, roughly, this is what SmarterX and Marketing Institute are going to look like.
[00:36:19] Paul Roetzer: And 90 percent of what all of our employees do is going to be done by their AI digital Mike may has
[00:36:25] Paul Roetzer: some questions for me about what that means for
[00:36:27] Paul Roetzer: his job. It’s like, yeah, Mike, it’s nice that you can go do talks, but I think your digital those talks And we’re just going to like,
[00:36:35] Paul Roetzer: you know, you’re going to license your persona to us because it’s part of your employment agreement and you’re gonna be like, wait, what?
[00:36:41] Paul Roetzer: And so that’s, I could just imagine like being a zoom employee and like listening to this interview and being like, I have no idea what this company is. It was weird. And I think like. just as you know, the future of work is a, is a very important area of research to me. It’s one of the things I spent a lot [00:37:00] of my time thinking about.
[00:37:01] Paul Roetzer: And this was one of the more abstract versions of the future of work I have heard with no detail as to how to occur or what the
[00:37:13] Mike Kaput: Yeah, my first question as a Zoom investor or employee or, you know, Executive would be, hey, if we’re getting rid of all these people, or if we are sending AI avatars to meetings, does our pricing need to
[00:37:26] Mike Kaput: change? Because it’s all based on user seats, like, are we
[00:37:30] Paul Roetzer: Minor
[00:37:31] Mike Kaput: details,
[00:37:31] Paul Roetzer: you know? It’s weird. Like, you know, for as much, um, as I, challenge Elon Musk sometimes, and, you know, he has wild visions for the future. At least there’s like master plans behind how he’s going to get there. Like he’s building a massive gigafactory for, you know, as a data center for GPUs.
[00:37:55] Paul Roetzer: He’s, you know, got plans for the robots. Like, at least there’s a plan [00:38:00] and like, you could logically play it out and say, these are the five things that need to happen for this future he envisions. Great. I just feel like that’s, That’s what a CEO does. Lay out a vision, but then like lay out a plan of the milestones to get to that.
[00:38:16] Paul Roetzer: And maybe, I don’t know, maybe just decided like to start talking about this wild idea he had in his head that morning and hadn’t actually like planned to say it. And now he’s got to like re engineer, but it sure sounded like something he’s been thinking about for a long time when he was saying it in this interview.
[00:38:32] Mike Kaput: Yeah. So just to kind of wrap this up though, you know, we are seeing,
[00:38:37] Mike Kaput: I don’t know if we are seeing a fundamental shift in behavior just yet, but we are seeing, you know, people have, for better or for worse, AI note takers now showing up
[00:38:46] Mike Kaput: to meetings. Like, how are you seeing, you know, we’ve got meeting summaries, which are Zoom.How are you seeing some of the way we do work change a little bit due to AI right now? Yeah.
[00:38:57] Paul Roetzer: I do think that this idea of AI assistance is just going to [00:39:00] become more prevalent. So you know, having your note taker there with you, having that, that assistant summarize, analyze the information, make recommendations to you, you know, create Zoom’s doing. Like, again, I’m, I’m a fan of. The initial features that Zoom has put into the product and the fact that they’ve made them part of the package, you don’t have to pay more for them.
[00:39:19] Paul Roetzer: Like they’d made some really good decisions and that’s why until I heard this longer vision, I was like, I actually was really impressed by what Zoom was doing with AI. I thought it was very thoughtful. so I do think that, you know, in the coming years, it’s going to be extremely common to have AI assistants and whether it’s a single general AI model, like a GPT five or six that can just do everything for you, or a collection of AI assistants, almost like GPTs or co pilots that are built to do specific tasks, I think that’s more likely in the near term, we’re just going to have a collection of AI assistants and eventually maybe they become so generally capable, just a single assistant, but I do think that the professionals [00:40:00] and leaders who are that find ways to think about that.
[00:40:03] Paul Roetzer: I think like Ethan Mollick would call it co intelligence, his, his book, that you just have this intelligence on demand for everything you’re doing in work. That’s possible now, and I think that the leaders and practitioners who figure out how to do that well can have a massive competitive advantage in the near term, while most other people are still trying to comprehend it and find some basic use cases.
[00:40:28] Paul Roetzer: So, like, I think of this, everything I do as a CEO, as a podcast host, as an author, as a, you know, a researcher, Every single workflow I go through, I now think about, can I build for this? Like, this is something I do the same thing every time. can I just build a thread in Gemini, or can I build a GPT that does this?
[00:40:49] Paul Roetzer: So, that’s, it’s already flipped my mentality to where I’m constantly thinking about the way to create an assistant that can work with me on these tasks. And that keeps getting [00:41:00] smarter
[00:41:00] Google’s Gemma 2
[00:41:00] Mike Kaput: All right, let’s dive into a bunch of rapid fire for this
[00:41:03] Mike Kaput: week. So
[00:41:04] Mike Kaput: first up, in the last few weeks, Google has officially released GEMMA 2, which is the latest iteration of its open source family. Now, this is available right now to researchers and developers. GEMMA 2 is the second entry in GEMMA.
[00:41:19] Mike Kaput: The open source model family, which Google has built and released based on the of models that are
[00:41:26] Mike Kaput: available commercially.
[00:41:28] Mike Kaput: So Gemma 2 is available in 9 billion and 27 billion parameter sizes and offers significant improvements in performance and efficiency compared to the previous version.
[00:41:39] Mike Kaput: Google actually says the 27 billion parameter model competes are twice its size, while the 9 billion parameter others in its class, including Meta’s Llama Llama3 8 billion parameter model.
[00:41:53] Mike Kaput: Google has
[00:41:54] Mike Kaput: also, uh, emphasized that they have done responsible AI development with Gemma2, [00:42:00] using rigorous safety processes during training.
[00:42:02] Mike Kaput: They also say they’re working on open sourcing their text watermarking technology, SynthID, before for GEMMA models. And GEMMA 2 is going to actually start being available next month for customers who can deploy and it on Vertex AI. So Paul, Can you give us a little context?
[00:42:23] Mike Kaput: Like, there’s so many new of open source options. Like, why does Gemma 2 matter here?
[00:42:29] Paul Roetzer: Well, I think for the non technical audience, for the business leaders, practitioners, knowledge workers, um, it’s not that huge of a deal. Like you’re not going to probably go, you know, build something on Gemma 2 tomorrow. It is smaller, it’s faster, it’s open for developers. So what it does is it accelerates possibilities of what can be built and the accessibility of those tools.
[00:42:50] Paul Roetzer: but kind of simultaneous to this. So Logan Kilpatrick, who we talked about before, he was at formerly OpenAI, he’s now at Google, in their AI studio working on [00:43:00] Gemini API and AGI. He tweeted on July 6th, And this is kind of like, he’s one of the guys notifications
[00:43:08] Paul Roetzer: for. he tweeted, Everyone should be getting ready for the cost of intelligence to go to zero.
[00:43:13] Paul Roetzer: It’s coming sooner than you would expect.
[00:43:16] Paul Roetzer: He
[00:43:16] Paul Roetzer: then filed that up and said, Intelligence is a spectrum. For 80 percent of use cases, with minor optimizations and a little focus, current models are sufficient. 20 percent of cases require yet another jump. No one wants to train a model that is more expensive than current versions.
[00:43:35] Paul Roetzer: Costs will keep falling. So when I saw that tweet, unrelated to, you know, when I was first thinking about the Gemma 2 conversation today, it’s just this idea of intelligence becoming insanely cheap, if not free. And so I shared his tweet and I said, this was like, what, Saturday morning or something. So this makes me think.
[00:43:58] Paul Roetzer: What happens to OpenAnthropic, et [00:44:00] cetera, if Google gets to AGI, however you want to define it first, and gives it away? What is the large language model frontier company business model if you can’t charge for intelligence on demand like they do now?
[00:44:13] Paul Roetzer: And
[00:44:13] Paul Roetzer: so I think, like, at a high level,
[00:44:16] Paul Roetzer: The importance of something like GEMMA 2 and even like LLAMA 3, you know, from, from META and right now it’s just kind of this game to like keep one upping each other with smaller, faster, more efficient models, is that it’s making the accessibility of intelligence, these AI assistants that we were just talking about, available At almost cost.
[00:44:35] Paul Roetzer: And Google has an entire other business, out there, all these other revenue streams, where if they chose to, if they were able to build smaller, better, faster models, and even bigger, more generally models, if they chose to just give it all away, Then what is the Anthropic and OpenAI business model?
[00:44:53] Paul Roetzer: Like all their revenue right now comes from charging for access to intelligence. So I don’t like, [00:45:00] this is something we’re going to go deep on right now. It’s not like main topic at this point, but it just has me thinking like, what really happens. If we do arrive at a point where the models are only incrementally better.
[00:45:10] Paul Roetzer: um, but for 80%, as, as Logan’s saying, if
[00:45:13] Paul Roetzer: 80 percent of the
[00:45:14] Paul Roetzer: use cases don’t even in models, they just need them to keep getting smaller and faster and smarter and more efficient, then 80 percent of what we do in business could leverage these models, which Google could basically give away, they it away.
[00:45:29] Paul Roetzer: and, then not only does it accelerate adoption, but it also And value creation within enterprises. It throws a crazy wrench into how this whole frontier model, you know, business is working. And maybe there are only a few winners in the end because it’s hard for everybody else to make money. Now you could make money as a service company or doing vertical integrations of these models, things like that.
[00:45:53] Paul Roetzer: But I don’t know. I mean, it’s just, it has me thinking like what the next six to 12 months is going to look like as these models [00:46:00] become smaller and as Apple rolls them out onto the iPhone and it’s like, If 80 percent of all use cases can be done on your device with these small models,
[00:46:08] Paul Roetzer: I don’t know, it’s an
[00:46:09] Paul Roetzer: interesting world to to start thinking about.
[00:46:12] OpenAI Hack
[00:46:12] Mike Kaput: So
[00:46:13] Mike Kaput: in some other news, we are hearing finally a report, from early 2023 that OpenAI, was actually hacked and a hacker gained company’s internal messaging system.
[00:46:26] Mike Kaput: So this incident was previously until now when it was broken by the New York Times, and according to that report, the hacker managed to steal about OpenAI’s technology from an online forum discussed the company’s latest developments. However, the core systems, housing, and building OpenAI’s technology were not compromised.
[00:46:49] Mike Kaput: Apparently, OpenAI executives disclosed this breach to employees and the board 2023, but chose not to make it public information [00:47:00] or customers in their minds had been stolen. So that was for the breach. for staying it. So, this has, really sparked internal debate at OpenAI.
[00:47:11] Mike Kaput: And especially because after the incident, we’ve talked about Leopold Aschenbrenner on a previous episode. He be an OpenAI technical program He raised concerns about the company’s ability to protect its secrets from foreign adversaries like China. He was later fired for allegedly confidential information.
[00:47:32] Mike Kaput: disputes that. He did anything of the sort. He actually had sent a memo to the board around that time arguing that OpenAI was not doing enough from a security perspective. So Paul, on episode 102, we talked about the big claims in Ashenbrenner’s situational awareness essay, more about kind of great power competition, national security, and how this artificial intelligence.
[00:47:55] Mike Kaput: Like, what’s your kind of perspective on, OpenAI’s [00:48:00] security and safety, like, is it as secure as it needs to be as a company given the importance of what it’s building?
[00:48:06] Paul Roetzer: No, it’s not, I would assume, and that’s part of the reason why last week we talked about the appointment of, the NSA leader, former NSA leader to the board. I uh, think we keep getting more validation for what Leopold’s situational awareness report says.
[00:48:24] Paul Roetzer: Um
[00:48:25] Paul Roetzer: They, you know, they keep kind of divulging things that are verifying his concerns.
[00:48:32] Paul Roetzer: We’ve heard similar warnings from Dario Amodei, Vanthropic, where he’s, you know, in podcast interviews said that any state actor that wants the secrets is going to get them. It’s just a matter of, you know, how much they have to spend to acquire them. But at Anthropic, I think he said there’s like two people, maybe three that know the model weights, cause they try and isolate that key information.
[00:48:53] Paul Roetzer: Cause that’s their fear is they would get that in the training sets and things like that. So it’s not like the frontier model [00:49:00] companies don’t know this is happening through, through cyber warfare, through hacking, through espionage, like. They’re fully aware, it’s going on, and so it’s a, it’s a major issue.
[00:49:12] Paul Roetzer: It seems like OpenAI is taking it seriously, maybe, maybe they were before this, but they certainly are now, and they’re trying to put steps in place better everything. But you know, it’s, it’s interesting. I finally watched Oppenheimer over the
[00:49:27] Paul Roetzer: weekend.
[00:49:28] Paul Roetzer: And so like, it’s, yeah, this stuff’s like, um, it’s like reliving the think, with, you know, these very dangerous, potentially dangerous secrets that other governments are going to want access to.
[00:49:42] Paul Roetzer: yeah, some great movies are going to be made about this stuff in the coming years. And I’m probably, the movie’s going to
[00:49:47] Paul Roetzer: be very close
[00:49:48] Paul Roetzer: to reality.
[00:49:50] Apple’s OpenAI Board Seat
[00:49:50] Mike Kaput: Alright, So in some other news, Apple an OpenAI board seat. So Phil Schiller, who’s Apple’s app store a former marketing [00:50:00] chief at the company has been selected for an observer role on open AI’s board. As a board observer, Schiller attends board meetings, but doesn’t have voting rights or other director powers.
[00:50:11] Mike Kaput: Um, this position takes effect later this year. And interestingly, it kind of puts par with Microsoft because as OpenAI’s largest backer, Microsoft also has an Observer seat on the board. Interestingly, this comes, you know, on the heels of Apple’s June announcement that they’re integrating ChatGPT into their devices as part of their upcoming AI So
[00:50:36] Mike Kaput: Schiller actually seems like an okay fit for the role because, you know, he doesn’t lead Apple’s AI initiatives, but he’s had this very long standing role
[00:50:45] Mike Kaput: managing Apple’s
[00:50:46] Mike Kaput: brand and launches. So, Paul, this kind of question, like, what is going on here? Because Microsoft has a board position that, essentially came with several billion dollars of investment.
[00:50:59] Mike Kaput: Apple, [00:51:00] which is a rival, just got one for free.
[00:51:02] Paul Roetzer: Yeah, I mean, we always got to keep in mind, Microsoft owns like 49 percent of OpenAI, and I think the first hundred billion in profits or something like that goes to Microsoft. So yeah, the dynamics of the Apple Microsoft relationship, they’re just fascinating. I mean, they’re frenemies, they, they have business deals together, they have battled each other for decades.
[00:51:22] Paul Roetzer: so it’s just, I don’t know, it’s just intriguing and I don’t think we know too much else. But the one thing that comes. at it, it sure seems like there’s going to be more to the Apple relationship than was announced that, you know, integration of ChatGPT is fine. It’s a nice feature.
[00:51:40] Paul Roetzer: It could be turned off whenever. I don’t know that you’re giving a board seat for, for that. There just has to be more to the product roadmap and the partnership roadmap between these two companies. And then it’s just interesting to know that Microsoft’s going to, To have a front row seat to know that, but again, like [00:52:00] Microsoft profits from it.
[00:52:01] Paul Roetzer: So in a weird way, it’s kind of like Microsoft wins no matter what. If they do massive deals with Apple and OpenAI ends up making, you know, billions of dollars in revenue, from that relationship, Microsoft is getting all that money. So they’re probably like, yeah, come on in Apple. It’s cool. It’s, it’s all, we’re profiting from all of it.
[00:52:21] Paul Roetzer: So it might be as simple as that as Microsoft doesn’t care because any money OpenAI is making is their money.
[00:52:28] AI and the Olympics
[00:52:28] Mike Kaput: So, in another story, we’ve heard that NBC set to introduce an innovative powered feature for the upcoming Paris This is into place on its streaming platform, Peacock.
[00:52:42] Mike Kaput: It’s called Your Olympic Recap. And it’s going to use an AI generated voice of legendary sportscaster Al Michaels to deliver personalized daily Olympic events. This system works by allowing users to customize their experience [00:53:00] in the Peacock app. They can input their name, select up to three sports of interest, and choose two types of highlights that they want to see. Each
[00:53:09] Mike Kaput: morning, they’ll then get a tailored recap narrated by the AI version of Al Michaels. Apparently, NBC trained the
[00:53:16] Mike Kaput: AI voice using previous appearances on the network. It was developed team at Peacock of Engineers. Data scientists, product managers, and to maintain the quality and accuracy of these recaps, NBC says that a team of editors is going to including the clips, before making a recap available.
[00:53:39] Mike Kaput: They’ll start being available, from July 27th on supported browsers and iOS devices. the first personalized out that following day. So, Paul, this seems like kind of a novel, cool use of AI. can you walk us through kind of your thoughts on this? Cause you had posted about this, quote, [00:54:00] floodgates are opening.
[00:54:01] Mike Kaput: AI voice. Video and voice is going to explode into the mainstream.
[00:54:05] Paul Roetzer: I haven’t had time process this one beyond that initial instinct, but as I started thinking about it a little bit in preparation for today, I think what it’s going to do is normalize the experience of interacting with AI
[00:54:22] Paul Roetzer: content. So when you take something like the Olympics with with such
[00:54:26] Paul Roetzer: broad reach, millions people you assume that would AI content, it just is a step towards normalization of the I could imagine it sounds like they’re not doing this with Peacock, but you could also see a choice where you’d say choose language and that Al Michaels could actually come to you in your chosen language. We know Eleven Labs and others would enable that kind of technology. So I think what it’s going to do if it wasn’t happening already is celebrities and media will see the to rapidly [00:55:00] multiply their reach, impact.
[00:55:02] Paul Roetzer: revenue, sports leagues, media companies, brands will see the opportunity to expand their reach and impact. And I think it just, like I said, it just opens the floodgates. Like everyone is going to start doing it because it was normalized in a mainstream way. So it’s, it’s probably something that by, you know, the end of 2024 and going into 2025, it just becomes commonplace interact with the AI avatar. Maybe the Zoom CEO is not that far off. Maybe we’re okay
[00:55:34] Paul Roetzer: with AI avatars,
[00:55:36] Paul Roetzer: you know, showing world. Um, it does probably accelerate the issue of people’s inability to identify deep fakes versus authorized AI content. So if it becomes commonplace for celebrities to authorize AI avatars for different experiences, then how do I know if it’s a faked?
[00:55:56] Paul Roetzer: or authorized use of their persona. [00:56:00] we don’t have the tech, the tech hasn’t caught up to so yeah, I mean, this is a fascinating one, but I do think that it’s going to be one of those sort of pivotal moments we’ll look back on and say, yep, that’s when AI avatars went mainstream when people became, you know, experienced.
[00:56:15] Paul Roetzer: And you could imagine it being applied to all kinds of things again. Like I started immediately thinking about like sports and,
[00:56:23] Paul Roetzer: you know, having Jim Nance, you know, being able to do the same thing with the masters next year and. The NFL being able to, you know, expand their reach. So yeah, I think this is going to be a huge area and I would be fascinated to see someone do like a deep dive on all the the ways applied transform entertainment and media and brand communications.
[00:56:47] Nintendo’s Future Plans with GenAI
[00:56:47] Mike Kaput: So Nintendo this past week has actually taken a pretty distinctive stance on the use of AI in game development. So Nintendo’s president recently addressed the company’s position [00:57:00] in game development during an investor
[00:57:02] Mike Kaput: Q& A. He acknowledged that, look, game development and AI Technology been intertwined. however, he drew kind of a clear line when it came to how Nintendo is going to use generative AI.
[00:57:14] Mike Kaput: He says, while we are flexible in technological developments, we hope to continue
[00:57:19] Mike Kaput: to deliver value to us and cannot be technology alone. So, he said, despite recognizing the creative potential of Gen AI. Nintendo has no plans to incorporate its first party games, and the primary concern here, especially for them, is kind of intellectual property rights issues that arise when they’re using AI other people’s work.
[00:57:43] Mike Kaput: This is Kind of in sharp contrast with other industry issues. players, I mean, like Electronic Arts, for instance, is kind of fully embracing generative AI their CEO actually says that over half their is going to be positively impacted by it moving forward. So, [00:58:00] Paul, obviously it’s about video game industry,
[00:58:02] Mike Kaput: but
[00:58:02] Mike Kaput: it’s just genuinely an interesting take on generative AI.
[00:58:06] Mike Kaput: where it seems like using the technology is almost unavoidable. Like, what do you make of this approach? Do you expect some other companies to maybe take a strong stance on this?
[00:58:18] Paul Roetzer: The one I have tremendous respect for it, I mean, I think it’s a, an honorable decision. I don’t know if it’s going to end up being the right business decision, but it’s, you know, I respect that they’re making the decision. I think it’s interesting that it’s a Japanese company
[00:58:33] Paul Roetzer: doing it,
[00:58:33] Paul Roetzer: because in Japan, They’re not enforcing copyright law on material used to train generative AI models.
[00:58:40] Paul Roetzer: So it’s actually, it’s not like in the U S where there’s uncertainty and you have to make ethical decisions. it’s not a legal decision per se yet. In Japan, you don’t have to make this decision because it’s not. a legal decision. They’ve, they’ve models. So it makes it even a more honorable decision because it is an [00:59:00] ethical decision and it’s a human based decision where they’re saying, no, we respect the creators and we’re going to allow them to do what humans do.
[00:59:08] Paul Roetzer: So yeah, I mean, fascinating to see how it plays out. I’ll be interested to see if other companies take this kind of approach and come right out and say, we in our creative process. I do think, I don’t know if it’s going to be preferred. Like, if users will prefer it, but I think there will certainly be, at minim large niches of users who want human based creative.
[00:59:35] Paul Roetzer: Like, they’re gonna want to buy from companies that they know made it. It’s almost like what Etsy was supposed to be, I think what Etsy started as, where it’s like, This is humans making this stuff. It’s not manufactured in is people creating things. I almost wonder if there isn’t that kind of market for content and games and movies and things like that, where it’s human creativity on display and not [01:00:00] AI
[01:00:00] Finding GPT-4’s Mistakes
[01:00:00] Mike Kaput: So some more OpenAI news. They have actually just introduced something called Critic GPT, which is a new on GPT
[01:00:10] Mike Kaput: 4, and it’s actually designed to identify and critique errors in ChatGPT’s outputs, particularly in code. OpenAI
[01:00:18] Mike Kaput: wrote about this release, quote, as we make advances in reasoning and model behavior.
[01:00:23] Mike Kaput: ChatGPT becomes more accurate and its mistakes become more subtle. This can make it hard for AI trainers to spot inaccuracies when they do occur, making the comparison task that powers reinforcement learning from human feedback, RLHF, much harder. This is a fundamental limitation of RLHF, and it may make it increasingly difficult to align models as they gradually become more knowledgeable than any person that can provide feedback.
[01:00:51] Mike Kaput: To help with this challenge, we trained CriticGPT to write critiques that highlight inaccuracies in ChatGPT answers.
[01:00:59] Mike Kaput: Now, OpenAI admits always get it right, but it does say that critiques from a human
[01:01:06] Mike Kaput: working with CriticGPT were actually preferred by the people training AI 60 percent of the time compared to the outputs alone were producing.
[01:01:17] Mike Kaput: So, Paul, this is It’s something that kind of seems to indicate we’re progressing towards a world where AI trained and evaluated, at least in part. By like, is that accurate to say?
[01:01:31] Paul Roetzer: Yeah, I mean, they’ve known that the RLHF has limitations. It’s difficult to scale because it does require humans and those humans have to then be trained. So, you know, in the article where they talked about this, they say the GPT 4 series of models, which powers ChatGPT, is a line to be helped with.
[01:01:49] Paul Roetzer: and interactive through RLHF.
[01:01:52] Paul Roetzer: It’s a key part of how they’ve trained the models, but it is difficult to scale. And so they have proven it [01:02:00] can be applied to code, but this is just the beginning. Yann LeCun, who we’ve talked about on the before, on the podcast before. Was it OpenAI working on this? And he’s now at Anthropic. He actually tweeted and said, this is just the beginning.
[01:02:14] Paul Roetzer: I expect we’ll see more benefits from scalable oversight techniques on real world tasks. So what that means to us, the average user, is that the AI will be able to start checking its own work, checking its own reasoning, chain of thought, the accuracy of its based you know, what it’s doing.
[01:02:34] Paul Roetzer: of truth in some manner. but I think more importantly, this is the basis for real world agents that improve and self improvement is key to unlocking the real potential of AI and also one of the greatest fears of   are the Doomers, that once these things have the ability to self improve and check themselves, [01:03:00] that they can, intelligence explosion happens that they can then take off.
[01:03:05] Paul Roetzer: And so I think this, study could be overlooked real easily, but I can tell you, the other labs are also working on it. I’m sure Anthropic is working on it. Google’s working on it. Everybody’s working on the same concept. And once you get to 90%, 95%, 98%, accuracy, we’re entering the realm where these things now have the ability to self improve at a rapid rate.
[01:03:31] Paul Roetzer: And And we’re not set up as a society to handle So I think this is something we will be talking about again in the coming months.
[01:03:41] Scale AI’s Business Model
[01:03:41] Mike Kaput: So we’re also hearing an interesting story about scale AI, which is a major AI startup 14 billion. And there’s a report that’s come out saying they are shifting their focus. to
[01:03:55] Mike Kaput: hiring highly professionals in the US to train [01:04:00] models. Now, this is a big change from their previous employing low cost in countries like Africa, India, and the Philippines for all these AI data labeling and training tasks that they. Build their business on.
[01:04:16] Mike Kaput: So Scale. ai now has about 300, 000 US based contractors, including PhDs, doctors, and lawyers. And these contractors basically play this crucial role in the fine tuning process AI models. They’re teaching them to generate accurate and human like models. responses. So this is part of a larger pivot on scale’s end towards higher value AI training tasks, particularly those that go into tuning LLMs.
[01:04:44] Mike Kaput: It sounds like this approach, though, is not without its drawbacks. It’s definitely more cutting into scale AI’s gross profit margins. Contractors have also complained about inconsistent work availability, payments, and some technical issues with how scale AI is doing this.[01:05:00] 
[01:05:00] Mike Kaput: So Paul,  I just wanted to kind of get your thoughts on like, what are the implications here on how we’re training and fine tuning whole? Because this kind of pulls back on how some of this stuff gets done.
[01:05:13] Paul Roetzer: Yeah, so without getting into scale scale and their practices, I listened to an interview with Mustafa Suleyman last week, and he was actually talking about this idea of post training and fine tuning these models, and it aligns really well here. So I’ll just read a quick excerpt. He said h, so how do you go about collecting high quality data?
[01:05:36] Paul Roetzer: Because obviously in pre training, it’s about volume of tokens, and they’re, Are the hyperscalers, they’ll have longstanding advantage because they already own search engines or YouTube or whatever, right? So he’s saying to build the front frontier models, you just need a ton of data and Google’s got all the data and you know, other people that are training these big models have access Whereas post training, you need a small number of very high quality tokens to [01:06:00] align the model to the behavior that you want for your product. And you can collect that from scratch. So he then talks about his experience building Pi at Inflection. Like when we built Pi, we created, and to this day, the most high quality human like conversational AI.
[01:06:16] Paul Roetzer: With the best EQ even today in the market. And we didn’t use any data from big providers. We collected all of it ourselves from scratch by training paid teachers. We called them teachers. Some people call them Raiders, but the crucial thing for a startup is you have to really, really, really pay attention to training teachers.
[01:06:35] Paul Roetzer: You have to pay them a lot of money. Like, I’ll just tell you from our perspective, we selected people who had an undergraduate education, nothing less, that largely spoke English as a first language, with some exceptions, that had domain expertise that we thought was valuable. Like, maybe they said they were very passionate about history, or they had a good kind of cultural knowledge, or they were movie buffs, or whatever it is.
[01:06:57] Paul Roetzer: They had to pass 20 hours of training and testing by [01:07:00] us. We would give them multiple choices. They would have to do sentence completion. They would have to do spot differences. They would have to do really quite hard analytical tasks. So he basically is talking about this idea of not just having. cheap labor that can look at A and B outputs and say A is better, but that actually can do extensive analysis because they’re bringing domain expertise.
[01:07:22] Paul Roetzer: And so you can imagine now how you can take a foundation model like Gemini or, GPT 4 or 5, and you can use the same foundation model. And then you can do this post training. with PhD level experts in their fields, um, or, attorneys or CPAs. So imagine now going into verticals and saying, okay, we’re going to take the same foundation models, everybody else, but our training, our post training with experts, we’re going to pay a bunch of money is what’s going to differentiate the model from others and the applications that we build.
[01:07:57] Paul Roetzer: And then Mustafa actually hints at Microsoft [01:08:00] here. So then he says, I’m a great believer in building and owning your own product, and as much as possible, controlling the key bit of the value there, which in my opinion is the large language model, and everything around that is secondary. The words that come out of the large language model are what you have to focus on, and that means that I think it’s reasonable to break off the pre trained model and get that from somebody else.
[01:08:23] Paul Roetzer: So again, what does Microsoft do? They have licensing deal with OpenAI. So they have the foundation model, but then on top of that, they can train it themselves. So that’s a good approach, but I think you need to own your fine tuning stack. And I would not give the fine tuning stuff to somebody else. You have to train your teachers because that’s not going to go away anytime soon.
[01:08:45] Paul Roetzer: So I think what it does is it shows where, where Microsoft’s going. It’s a prelude to what a lot of other companies are going to do. A lot of other SaaS businesses are going to take these like perplexity. They’re going to take these base models and they’re going to. fine tune or post train the models with their experts to do specific [01:09:00] things.
[01:09:00] Paul Roetzer: That’s the defensible position. That expert training that goes on top of the frontier models is what creates the moats for the companies. But if you want to think about verticalization strategies is also where that comes in. If you want to think about career opportunities, you can now envision a day where my, marketing experts, lawyers, finance VCs, like it’s your expertise.
[01:09:24] Paul Roetzer: that can help post train these models that is extremely valuable. And I think we’ll have an entire ecosystem of people whose jobs is to do this kind of post
[01:09:35] Mike Kaput: Yeah,
[01:09:36] Mike Kaput: that seems like there’s some huge implications there. I’m sure we’re going to be talking about more.
[01:09:41] AI Tech Updates – Runway
[01:09:41] Mike Kaput: All right, so in our last segment of today’s episode. this is kind of becoming a in our rapid section is we’ve got a handful of updates.
[01:09:53] Mike Kaput: We’ve got funding announcements, product more. So Paul, I’m going to go through these now. very quickly for each [01:10:00] just going to kind of get, you know, hopefully you’re kind of lightning round take on what’s going on here. So first, Runway, which is a leading AI video generation startup. is in discussions to raise a round of funding.
[01:10:14] Mike Kaput: This they are looking to reportedly raise about 450 million at a a valuation of approximately 4 billion. So this also comes at the same has recently released Gen 3 Alpha, their generation model. and that claims to offer user control and speed. So,
[01:10:38] Mike Kaput: Paul,
[01:10:38] Mike Kaput: you this past week hands Gen 3.
[01:10:44] Mike Kaput: what did you make in the funding announcement? What were impressions using Gen 3?
[01:10:49] Paul Roetzer: We’ve been long time fans of Runway, I’ve been following them and using it for five plus you. My initial reaction to Gen 3 is that it’s not obvious [01:11:00] the improvement. You have to be really good at prompting it. So this is a product where prompting is essential and they have a whole guide of how to do it, like how to instruct the kind of shots you want.
[01:11:10] Paul Roetzer: So this. Out of the box, this goes to the premise that if you’re an expert in video production, you’re going to get way more value out of Gen 3 than a non me. So you got to put in the work to make Gen functional.
[01:11:25] 
[01:11:25] Paul Roetzer: Um, Costs about a dollar for 10 seconds of video, so you can create up to 10 seconds.
[01:11:30] Paul Roetzer: And then it’s about a like a hundred credits, which So. It’s an alpha though. So more to come, the technology and keep getting more impressive. People who know how to prompt it are getting incredible results from it. It’s definitely worth checking out, but it requires some expertise get value from.
[01:11:49] Paul Roetzer: I have not, I’ve been unimpressed with the five or six videos but also
[01:11:57] AI Tech Updates: Hebia
[01:11:57] Mike Kaput: So we also have a [01:12:00] big AI fundraising announcement in another area of domain expertise. Hebia, which is a startup in generative AI search. For document search and analysis, they have actually secured a Series B funding round of million Now that round was led by Andreessen Horowitz and valued the company between 700 and 800 million. million. dollars. So basically what Hebea does is its AI can analyze billions of documents simultaneously, they claim, including different file formats like PDF, PowerPoint, spreadsheets, and then give you specific answers about them.
[01:12:37] Mike Kaput: The company right now primarily targets financial services firms that could benefit from that type including hedge funds and investment banks, but the product also could be applied to law firms and other professional domains. So, Paul, this seems like a pretty significant chunk of change for a startup that does AI document search
[01:12:56] Mike Kaput: and analysis.
[01:12:57] Mike Kaput: Like, how opportunity here [01:13:00] for industries like financial services, accounting, et et cetera?
[01:13:04] Paul Roetzer: It’s huge. I think the verticalization is going to be key to unlocking value. The part I can’t figure out about this company is what their tech stack is. there’s nothing in TechCrunch about it. Allie Miller shared on Twitter that she was a proud investor. Allie closely. she’s a proud investor, an AI startup helping knowledge workers across legal, real estate, consulting, investment banking, pharma, government, and more.
[01:13:30] Paul Roetzer: And so I had asked Allie, I don’t know, she wasn’t sure. It just didn’t reply. but I did searches, like I was trying to find this perplexity. So I said, what, what language model are they using to power the reasoning and language? Like, I’m really intrigued to learn more. I’m definitely bullish on this kind of company.
[01:13:47] Paul Roetzer: I just can’t figure out what their tech stack is. Makes them claim they can do reasoning so well. So I don’t know. I’m intrigued by it. definitely a product worth following. Just, I would love to understand a little bit more [01:14:00] how
[01:14:01] AI Tech Updates: Perplexity
[01:14:01] Mike Kaput: Alright, next up, we have some product updates from Perplexity. They have announced significant upgrades to their ProSearch feature, which makes it better at advanced problem solving and research.
[01:14:12] Mike Kaput: It now has multi step reasoning, advanced math and programming capabilities, and comprehensive programming. capabilities. So this basically allows ProSearch to tackle complex queries, breaking them down into steps, planning out goals and answers more efficiently. So what this does and kind of how perplexity sounds like they’re looking at it is that this mimics the work of basically an expert Research Assistant.
[01:14:40] Mike Kaput: So, ProSearch is now to all users with five free uses every four hours. If you’re a Perplexity Pro subscriber, unlimited access daily to the feature. So, Paul, this seems like some interesting updates to a tool we use and talk about often. Like, we’re kind of getting into this area where this might [01:15:00] Be able to function as an AI research assistant in the true sense of the word.
[01:15:04] Mike Kaput: Like what do you think?
[01:15:06] Paul Roetzer: I, noticed it. So if you use the app, it does start to show you more about its process, like that it’s going through so you can see it happening. Aravind did talk about this on the Lex Freeman podcast on June 19th that I know on, I think it was episode 103 or 102. So he goes into more detail about how this stuff works, but it’s definitely a direction that they’re going and is trying to know, a research assistant that has, or a knowledge assistant, they would probably call it, you know, is more reliable and goes through different steps and, you know, reasoning processes. So, yeah, I mean, again, under the context of, I have some questions around Perplexity’s business practices and ethics, but the product itself is wildly impressive if set that stuff
[01:15:54] Mike Kaput: all
[01:15:54] Mike Kaput: right, our final updates here, to end here, [01:16:00] both of them come from
[01:16:00] AI Tech Updates: ElevenLabs
[01:16:00] Mike Kaput: Eleven Labs. this is a we talk about pretty often.
[01:16:05] Mike Kaput: First up, they announced a big expansion of their Reader app. is what we talked about last will actually read. documents and content for in professional sounding voices. And what they’ve done here is they’ve actually added four very well known voices from legendary performers from Judy Garland, James Dean, Burt Reynolds, and Sir Lawrence Olivier.
[01:16:28] Mike Kaput: And this basically comes from partnerships with each of these stars. So you
[01:16:33] Mike Kaput: can have
[01:16:34] Mike Kaput: them read to you, any type of give the app. Second, 11Labs has introduced a new tool called Isolator. Basically, this extracts clear any audio source.
[01:16:48] Mike Kaput: It strips
[01:16:49] Mike Kaput: away background noise and leaves only the and you
[01:16:53] Mike Kaput: can upload
[01:16:53] Mike Kaput: an audio file or record to the app.
[01:16:55] Mike Kaput: through the So
[01:16:58] Mike Kaput: Paul, like, let’s kind of [01:17:00] really quickly walk through what 11 Labs is up to. You’re a big fan of the Reader app. Like, What do you make of the celebrity voices? Definitely seems like Voice could be really and useful as well.
[01:17:11] Paul Roetzer: think licensing from the estates is a really smart play and Again, probably like the Al Michaels thing, you’re just going to see an explosion of this stuff. There’s so much money to be made. People are going to love the familiarity of celebrity and actors, actresses. So I think this is going to, you know, a year from now, we’ll look back and it’s just be like, Oh, we didn’t have that before.
[01:17:31] Paul Roetzer: I couldn’t on demand have these different voices. So that’s going to probably be everywhere. It’ll probably be built into, you know, and Google. yeah. It makes a ton of sense for everybody. And then, in terms of the voice isolator, I think the key here is like, if you do any kind of production work, whether it’s through Descript or 11 Labs, like it’s hard to keep up with the features that are being released by these companies and how valuable they are and how, like, they’re [01:18:00] Simple they are to integrate.
[01:18:02] Paul Roetzer: And this goes back to that whole idea at the beginning about finding value in these like specific use there’s just thousands of this, of like great tech that’s been built to do specific things. And you can find so much value in enterprises if you know where to look. And if you have frameworks to be assessing the technology regularly and finding ways to integrate it into your workflows.
[01:18:25] Paul Roetzer: So. Yeah, stuff like this is, is cool to see and definitely something I flagged for our team and say, we should probably take a look at
[01:18:31] Paul Roetzer: at
[01:18:31] Mike Kaput: For sure.
[01:18:33] Mike Kaput: All
[01:18:33] Mike Kaput: right, Paul, that’s a wrap on this week in AI. A couple final quick announcements here. If you have not yet subscribed to our newsletter, go to marketingainstitute.
[01:18:43] Mike Kaput: com forward slash newsletter. We cover all the news that didn’t it into this. episode everything we talked it’s a very helpful to start your week and catch up on artificial intelligence. Last but not least, please, if you have not already, leave us a review. Your [01:19:00] feedback is directly incorporated intothe show to make it better.
[01:19:03] Mike Kaput: And also, reviews help us get into the earbuds or AirPods of more listeners weekly. So, if you haven’t done that, we’d so, so appreciate any and all feedback you have. So, please leave us a review, if you can, on your favorite platform. All right, Paul, thanks so much for breaking it down for us this week.
[01:19:23] Paul Roetzer: Thanks everybody for joining us. We’ll talk to you again next week.
[01:19:26] Paul Roetzer: Thanks for listening to The AI Show. Visit MarketingAIInstitute. com to continue your AI learning journey and join more than 60, 000 professionals and business leaders who have subscribed to the weekly newsletter, downloaded the AI blueprints, attended virtual and in person events, taken our online AI courses, and engaged in the Slack community.
[01:19:49] Paul Roetzer: Until next time, stay curious and explore AI.

Related articles

Recent articles