56 Min Read

[The AI Show Episode 122]: ChatGPT Search Is Here, McKinsey: AI Worth “Trillions” in Coming Decades & Microsoft AI CEO Calls AI “New Digital Species”

By Claire Prudhomme on November 5, 2024

Podcasts
Featured Image

Wondering how to get started with AI? Take our on-demand Piloting AI for Marketers Series.

Learn More

Twenty years of search habits might be about to change forever. Join Mike and Paul as they explore the latest search features from ChatGPT that could reshape how we find information online. They'll break down McKinsey's eye-opening report on AI's trillion-dollar economic impact, and explore Suleyman's provocative vision of AI as an emerging "digital species." Plus, buckle up for a rapid-fire round covering everything from autonomous AI agent updates to Apple's latest AI moves. 

Listen or watch below—and see below for show notes and the transcript.

Listen Now

Watch the Video

Timestamps

00:04:17 — ChatGPT Search is Here

00:21:27 — McKinsey Report

00:37:02 — Microsoft AI CEO Interview

00:47:06 — Sam Altman’s Latest Interview

00:54:59 — More AI Agent News

00:57:42 — A16z + Microsoft on AI Regulation

01:01:06 —  OpenAI vs. Microsoft

01:03:56 — Microsoft AI Coding

01:05:41 — Apple Intelligence Rollout 

01:08:33 — Runway CEO Statement on AI Companies

Summary

ChatGPT Search is Here

ChatGPT can now search the web far better than before, thanks to this week’s major update of the tool.

According to OpenAI's announcement:

“ChatGPT can now search the web in a much better way than before. You can get fast, timely answers with links to relevant web sources, which you would have previously needed to go to a search engine for. This blends the benefits of a natural language interface with the value of up-to-date sports scores, news, stock quotes, and more.”

The new search integration has several key features, including natural language queries that trigger web searches as needed. It offers direct links to source materials and citations, as well as specialized displays for categories such as weather, stocks, sports, news, and maps.

Users can also access real-time information on current events, scores, and market data, with integration from major news publishers like AP, Reuters, and the Financial Times.

The rollout of ChatGPT search is going to be tiered based on your subscription type. ChatGPT Plus and Team users have immediate access now. In coming weeks, Enterprise and Education users will get access. In the coming months, free users will get access. 

McKinsey’s Report Details AI Impacts on the Global Economy

McKinsey Global Institute recently released “The next big arenas of competition,” a 213-page report that explores 18 future “arenas” that could reshape the global economy and generate a combined $29 trillion to $48 trillion in revenues by 2040.

McKinsey Global Institute defines “arenas” as industries that transform the business landscape and have two core characteristics: high growth and dynamism.

One of these arenas is AI Software and Services—and the implications of McKinsey’s data on growth in this area are pretty staggering.

McKinsey defines AI as “a machine’s ability to perform cognitive functions that we usually associate with humans.” This includes traditional machine learning capabilities that predict outcomes and behaviors, as well as generative AI applications. However, they exclude hardware, such as Nvidia chips, from this arena. 

McKinsey reports a surge in investor interest in advanced AI, particularly generative AI, with equity investments rising from $5 billion in 2022 to $36 billion in 2023. 

The growth of analytical and generative AI is expected to significantly enhance business productivity, projecting industry revenues to reach between $1.5 trillion in a lower range of scenarios and $4.6 trillion in a higher range of scenarios by 2040, with a compound annual growth rate of 17 to 25 percent. 

McKinsey's prediction proposes an estimated range of total economic potential of $15.5 trillion to $22.9 trillion annually by 2040. 

Microsoft AI CEO Interview

The popular Masters of Scale podcast just dropped a new interview with AI leader Mustafa Suleyman that we think is worth a listen.

Masters of Scale is a podcast primarily hosted by Reid Hoffman, the co-founder of LinkedIn and Inflection, and former OpenAI board member and current Microsoft board member. Suleyman is the former co-founder and CEO of Inflection, an AI company that got acquihired by Microsoft.

Now, he is CEO of Microsoft AI, which means he is not only on the bleeding edge of AI development, but also a key player in both Microsoft’s AI strategy and the company’s relationship with OpenAI.

Today’s episode is brought to you by our AI for Agencies Summit, a virtual event taking place from 12pm - 5pm ET on Wednesday, November 20. The AI for Agencies Summit is designed for marketing agency practitioners and leaders who are ready to reinvent what’s possible in their business and embrace smarter technologies to accelerate transformation and value creation.

You can get tickets by going to www.aiforagencies.com and clicking “Register Now.” When you do, use the code POD100 for $100 off your ticket. 

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: We are in the very, very early innings of this, probably the top of the first inning of like adoption and transformation from AI, but it is. It's going to be massive. And so studies like this from McKinsey help put in context how massive, but it's all a guessing game. It's big though. 

[00:00:27] 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. Join us as we accelerate AI literacy for all.

[00:00:55] Paul Roetzer: Welcome to episode 122 of the Artificial Intelligence Show. [00:01:00] I am your host, Paul Roetzer, along with my co host, Mike Kaput. We are recording this November 4th. 10:20 a. m. Eastern time. I have a feeling that might be relevant this week. I don't know. Like there's just, there's so much happening. We talked so much about like all the model updates and releases on the last episode, I think it was, and, it seems like we're probably going to get some more stuff before the holidays here.

[00:01:22] Paul Roetzer: So yeah, these dates are, becoming important and we are now, what about 26 days away from the two year anniversary of ChatGPT, Mike? So that'll be interesting to see what OpenAI does with. With that, I'm expecting search wasn't the final thing we're going to get leading into the two year anniversary. So, lots to talk about today.

[00:01:42] Paul Roetzer: We've got ChatGPT search, as I alluded to. We've got a new report from McKinsey we're going to dive into. We've got some interesting interviews with Mustafa Suleyman and Sam Altman, and then a whole bunch of other updates. So, we're going to get into that, but first this episode is brought to us by our AI for Agencies Summit.[00:02:00] 

[00:02:00] Paul Roetzer: This is our second annual virtual event. It's taking place from 12 p. m. to 5 p. m. Eastern Time on Wednesday, November 20th, so that is coming up fast. I should probably start building my opening keynote for that one. 

[00:02:13] Paul Roetzer: I have notes on it. Like, I generally have a direction, but I'm unsure because I'm talking about AI agents and like the future of agencies and I'm not settled yet on what exactly I think is going to happen with agencies, but I gotta find some mind space in the next week or two here to really sit back and think about that one.

[00:02:32] Paul Roetzer: So AI for Agency Summit is designed for marketing agency practitioners and leaders who are ready to reinvent what's possible in their business and embrace smarter technologies to accelerate transformation and value creation. Important note here for the brand side. If you are a business leader, marketing leader, whatever your role is, if you touch the marketing agency relationship, you'rgoing tona want agencies moving into 2025 that have, are [00:03:00] actively trying to solve for this stuff.

[00:03:02] Paul Roetzer: So, if you have some agencies that are core to your partnerships, maybe pass this along to them to take a look at, could, could be very valuable for them. And valuable for you moving into next year. So during the event, you'll join hundreds of other forward thinking agency professionals to consider ways to recruit AI savvy talent and upskill your team, explore how AI tools can boost creativity, productivity, and operations, hear insider stories from agencies that are piloting and scaling AI successfully, and understand how AI impacts your pricing models and service offerings.

[00:03:32] Paul Roetzer: we're going to have, I think there's six presentations from agencies that are actively doing this. We're going to have kind of case studies. They're going to share inside information about what they're doing and seeing across different areas of AI integration in their companies. You can go to AI four agencies.

[00:03:47] Paul Roetzer: That's FOR ai four agencies.com. Click the register now button. use promo code POD 100, that's POD 100 for 100 off your ticket. Again, check out ai four [00:04:00] agencies.com. There is going to be an on-demand option, so if you can't make it on November 20th, you're in a different time zone and can't be there live.

[00:04:06] Paul Roetzer: No worries. You can, get the on-demand option. So, yeah, AI4Agencies. com, check that out, and Mike, let's get into our main topics. 

[00:04:17] ChatGPT Search

[00:04:17] Mike Kaput: All right, Paul, like you alluded to, first up this week is ChatGPT can now search the web far better than before, thanks to this week's major update with us getting ChatGPT search.

[00:04:31] Mike Kaput: So OpenAI wrote in an announcement this past week, ChatGPT can now search the web in a much better way than before. You can get fast, timely answers with links to relevant web sources, which you would have previously needed to go to a search engine for. This blends the benefits of a natural language interface with the value of up to date sports scores, news, stock quotes, and more.

[00:04:56] Mike Kaput: Some key features of this new search feature include [00:05:00] natural language queries that automatically trigger web searches when needed, direct links to source material and citations, and Specialized displays for categories like weather, stocks, sports news, and maps. Real time information access for current events, scores, and market data.

[00:05:19] Mike Kaput: And integration with major news publishers including AP, Reuters, Financial Times, and others. The rollout of ChatGPT Search is going to be tiered based on your subscription type. ChatGPT Plus and Team users have immediate access right now. In the coming weeks, enterprise and education users will get access to this.

[00:05:42] Mike Kaput: But if you're a free user, you're going to have to wait for months. In the coming months, eventually, you will get access to this feature. So, Paul. I want to talk a lot about what this means for Google, but first, can you just share your initial impressions [00:06:00] of search within ChatGPT now? Like how useful do you find these new capabilities?

[00:06:05] Paul Roetzer: Yeah, my, my impression, is that they're, they're going to change the way you use ChatGPT honestly. So I think the user experience is really solid, when you choose, you can choose a search. So if you haven't tried it yet, you can actually click search. Or, ChatGPT just intuits that you're conducting something where search would be valuable and it automatically will use search.

[00:06:28] Paul Roetzer: So I personally tested on, like, planning a trip, to Florida, you know, this fall. and so I just said, like, planning a trip to Florida, specifically, it's the city. And, It immediately gave a very logical structure, it pulled in pictures, it had a getting there header, a where to stay, a must see attractions, activities, dining, weather, tips, so all this, like, all the structure that we're used to with ChatGPT where it automatically kind of curates these outlines, but now it had inline citations, I tried one I [00:07:00] think this was like Friday morning or something or Saturday.

[00:07:02] Paul Roetzer: Give me a summary of the Cavs Cleveland Cavaliers game from the last, from last night. And it pulled in Reuters and Talk Sport articles and summarized it, but then linked right to those articles. I tried one just out of curiosity. I said, like, when was the last time the Cleveland Cavaliers started 6 0?

[00:07:16] Paul Roetzer: And it pulled in Wikipedia, NBA, and then it actually pulled in real time stories from Reuters. And the New York Post from the previous night's games. I tried one like, wire tech stock down, tech stocks down, so I was just like experimenting with it. And my overall take is, I really enjoy the no ad experience.

[00:07:35] Paul Roetzer: I went over to Perplexity and I tried some of those same searches again, and honestly, like, Perplexity started to feel a bit obsolete to me. More from the user interface. And it feels very cluttered all of a sudden, which, I mean, if you're used to Google, Google's cluttered as, as is, so Perplexity didn't seem like that big of a difference, I guess.

[00:07:57] Paul Roetzer: But when you have this totally clean interface [00:08:00] with, with no ads or anything, Perplexity just felt different to me all of a sudden, and not as modern and clean. And that's before Perplexity infuses ads, which we know they're working on. 

[00:08:13] Paul Roetzer: that was kind of like an initial reaction. but to go back to, you know, you had mentioned kind of like the bigger picture here about like Google and search.

[00:08:22] Paul Roetzer: I actually was like, I think you and I talked about this in March 2024. I didn't go back and look for the episode where we would have touched on it. But Sam Altman did an interview with Lex Friedman. It was episode 419 of Lex Friedman's podcast. We'll put the link in the show notes. And I remembered, like, as soon as I saw this last week, I was like, wait, what was that Lex interview where he talked about this?

[00:08:43] Paul Roetzer: I remember noting it at the time. So, anyway, I went back to the transcript, and again, we'll put the link in, and Lex asked Sam specifically about search in March 2024. So, I think it's important for people to realize, like, this isn't a new thing. OpenAI has been thinking about search and how they would handle it for [00:09:00] years, and Sam has openly talked about it for a while.

[00:09:03] Paul Roetzer: So, In that interview, Lech said, like, how are people going to access information? You know, people show up to GPT as a starting point. So is OpenAI really going to take on this thing that Google started 20 years ago? And Sam said, I find that boring. I mean, if the question is, will we build a better search engine than Google, then sure, we should go, people should use the better product.

[00:09:25] Paul Roetzer: But I think that would be, to understate what this can be. So again, kind of quoting him, Google shows you 10 blue links, well 13 ads and then 10 blue links, kind of taking a shot at Google. And that's one way to find information. But the thing that's exciting to me is not that we could go build a better copy of Google search, but that maybe there's some better way to help people find and act and synthesize information.

[00:09:48] Paul Roetzer: Actually, I think ChatGPT is that for some use cases and hopefully we'll make it. Be like that for a lot more use cases. he went on to say, but I don't think it's that interesting to say, how do we do a better job of [00:10:00] giving you 10 ranked webpages to look at than Google does? Maybe it's really interesting to go say, how do we help you get the answer, the information you need?

[00:10:08] Paul Roetzer: How can we help create that? In some cases, Synthes synthesize that in others or point you to it and yet others, but a lot of people have tried to just make a better search engine than Google and it is a hard technical problem. It is a hard branding problem. It is a hard ecosystem problem. I don't think the world needs another copy of Google.

[00:10:26] Paul Roetzer: But then he does talk about integrating the ChatGPT, and he says that's cooler. Like if we could build search right into it, as you might guess, he said, we are interested in how to do that. Well, the intersection of large language models plus search, I don't think anyone has cracked the code yet. I would love to go do that.

[00:10:43] Paul Roetzer: And then Lex said, well, what about ads? Have you thought about how to monetize it? And he said, and this is important, this kind of gives a preview of how they may not go the direction Perplexity is going to go and the way Google is built on, says, I kind of hate ads as an aesthetic choice. I think ads needed to happen on [00:11:00] the internet for a bunch of reasons to get it going, but it's a momentary industry.

[00:11:04] Paul Roetzer: The world is richer now. I like that people pay for ChatGPT and know the answers they're getting are not influenced by advertisers. I'm sure there's an ad unit that makes sense for language models. And I'm sure there's a way to participate in a transaction stream in an unbiased way, but basically he's just not a fan of ads at all.

[00:11:23] Paul Roetzer: And he really wants to build their business model based on subscriptions instead. And then the other, you know, a couple points of context here in, so it was in June, I think of this year, June or July, they introduced SearchGPT prototype. So again, we've known this was coming, how exactly they executed it wasn't clear.

[00:11:44] Paul Roetzer: Take care. But even back then, they showed how they were thinking about it, said we're testing SearchGPT, a prototype with new search features designed to combine the strength of AI models with information from the web to give you fast and timely answers. So, [00:12:00] and then, like, one of the key reasons, so getting into, like, why does search matter?

[00:12:04] Paul Roetzer: like, why is this important? You have to remember that, like, large language models have a cutoff date on their knowledge base. So GPT 4. 0's knowledge cutoff, like, when it was trained, is October 2023. So the data in 4. 0, when it's not connected to the internet, it doesn't know anything about the world from the last 13 months, basically.

[00:12:26] Paul Roetzer: And so it was pre trained on the data up to October 2023. So What the benefits of search in the LLM is access to real time information in theory and so far research has proved this out, a reduction of hallucinations. So you can have greater accuracy and reliability if you can vet the outputs with real time data and search information.

[00:12:50] Paul Roetzer: it improves the potential for personalization and localization of results. So, you know, a lot of times that's not being infused into these LLMs where, you know, Mike and I may have [00:13:00] the same experience even though we're different people and maybe in different locales. And it also enables the verticalization of, information of these LLMs to specific industries or domains.

[00:13:11] Paul Roetzer: So imagine in like law, if you don't have the latest legal cases or in healthcare, if you don't have the latest research or reports or in sports, finance, stocks, all of these verticals require up to date information that these models don't have inherently. They're not, they're not traditionally a search engine like Google, where you're used to getting the most updated information.

[00:13:32] Paul Roetzer: And then my final thought here is just like the concerns or weaknesses, this isn't all like sunshine and rainbows, like there's, this does open up, a greater potential for misinformation, you know, once people figure out how to kind of trick these systems, the information they get to be indexed and found, can become an issue.

[00:13:52] Paul Roetzer: So we don't know how reliable they are. There was a tweet I saw from Emily Bender, who's a, University of Washington [00:14:00] professor of linguistics and somebody I follow pretty closely on Twitter. And she was saying, like, one of the biggest challenges to this is people come to believe these things are accurate.

[00:14:10] Paul Roetzer: So one of the beauties of what Google 20 years. is generally, you know, get really good at presenting you accurate information. And you can kind of trust it and then verify sources and things like that. But she was saying, we don't know how reliable these things are, how much they hallucinate. And the problem is that, and this is to quote her, a system that is right 95 percent of the time is arguably more dangerous than one that is right 50 percent of the time.

[00:14:38] Paul Roetzer: People will be more likely to trust the output and likely less able to fact check the 5%. So if people start assuming that search and ChatGPT is right all the time, and they don't even click on the links, and maybe it's error rate is only 5 or 10%, that's a significant error rate. And that can affect, you know, how people use the system and how they trust it.

[00:14:59] Paul Roetzer: Bias and [00:15:00] manipulation become a thing. Again, think about search and how there's the shadow industry of basically just figuring out the algorithms and trying to stay ahead of it. So you already have people trying to figure out how to game these algorithms. It sounds like they're largely using Bing's index, not their own index of the web.

[00:15:16] Paul Roetzer: Yep. So now you have SEOs, you know, that are probably racing to figure out how to game the Bing algorithms to get injected into the search results. Yep. creates an echo chamber because they're going to license, like a lot of the stuff that's going to get surfaced is going to be licensed from specific sources like a Reuters who are willing to do a deal.

[00:15:34] Paul Roetzer: And let's say politically or religiously or whatever, you know, you want to throw in there, they do deals that heavily lean toward one side or the other. Now all of a sudden, or geographically, government wise, now all of a sudden they lean, the results lean in those directions because that's who they did licensing deals with.

[00:15:53] Paul Roetzer: so again, it's, there's all kinds of challenges around this as well. So on the surface, awesome tech, great user [00:16:00] experience. It's clean. I like it. There's a bunch of other macro level stuff going on here though. 

[00:16:06] Mike Kaput: And just as a very small example of what this could mean for marketers, like I just typed in, what are the top AI tools today?

[00:16:14] Mike Kaput: And it says, as of November, 2024, several AI tools gained prominence, blah, blah, blah, lists out 10 different tools from very clear sources, none of which I will probably ever click through to because like, seriously, like I have the name of the tools I'm then going to go search. So like, this is the exact type of top of the funnel query typically.

[00:16:35] Mike Kaput: You might be targeting to really get someone's eyes on your website. Obviously we've talked about for over a year now that kind of strategy is going the way of the dodo, but like this is very clearly giving me the answer I have and enough information to ask follow ups with links that I'm probably not going to be clicking through to nearly as often.

[00:16:54] Paul Roetzer: Yeah. Yeah. And it, you know, it's kind of like the step toward the one, like with [00:17:00] the one answer that you get from voice. So again, like if I say advanced voice, so let's say advanced voice, I don't know if it has search built in or not. I would guess maybe it does. I haven't tried that yet, but let's say that you become reliant on voice interface and like, you know, six, 12, 18 months from now, you're basically just either talking to Apple intelligence or advanced voice or Google assistant, whatever, when you need something.

[00:17:23] Paul Roetzer: Well, now your world is living within whatever bubble that company's algorithm is deciding the right answer is to things. Right. Whether it's what you're going to buy or what you're going to believe or, you know, just real time information. You've now, like the beauty of Google's link strategy is a diversity of places to click through to form your own opinions and points of views and beliefs.

[00:17:49] Paul Roetzer: If we become too reliant on a single interface and assume that whatever it presents to us is right, and in the case of voice, you're only getting one, [00:18:00] then you start to run the risk that we're, we're like really forming people's beliefs and opinions. In, in the light of whoever's controlling the algorithms or the models that present that information, which again, you could get into all kinds of ramifications around that, like who's building the models, what government is supporting those models, things like that.

[00:18:21] Paul Roetzer: So it's infinitely fascinating, like there's so many threads you could pull from a simple, and this is, again, part of our goal with this show is, yeah, chat should be searched awesome. But, like, let's step back for a minute and think about, like, the bigger implications here and the questions that it opens up that we don't necessarily have the answers to, but it's, it's good to not just take the technology on its surface and assume it's, you know, great.

[00:18:44] Paul Roetzer: We gotta, like, think about the other angles here. 

[00:18:46] Mike Kaput: So, to kind of wrap this topic up, like, How bad is this for Google? Like, not only does this come out, but in the past week we got reports that Meta is trying to put its own AI search engine into Meta AI [00:19:00] to explicitly reduce reliance on Google and Microsoft.

[00:19:04] Mike Kaput: Perplexity, despite some legal challenges, obviously, is kind of full steam ahead as an alternate search option, though they may get hurt even more by this. They're going to start charging for ads. Like, how worried are they? Am If I'm the Google search team right now? 

[00:19:20] Paul Roetzer: Yeah, I don't know. It's a, it's a difficult question.

[00:19:23] Paul Roetzer: we have, you know, Bing has certainly seen improvements. So I'm just, like, the current From what I can see on StatCounter, Google owns 90 percent of the search market, Bing's around 4%. but Bing has gained ground. You know, I think we talk a lot about perplexity because we all live, and a lot of our listeners live in sort of this AI bubble, and it's perplexity seems like it's much larger than it probably really is.

[00:19:49] Paul Roetzer: Just because we talk about it a lot, but it doesn't even show up in like the top six search engines, seven search engines, things like that. so their, their market share is, I [00:20:00] think, well below a half a percentage point. It's, you know, now it's growing. Yeah. and I don't know. Like, I think part of this is an interface question.

[00:20:09] Paul Roetzer: Like, what is the interface of the future? Will two, three, four years from now, are, are we largely going to rely on, voice? Are we going to just live within our language models? And they become the primary interface to everything and it's connected to everything. In which case, then, you know, you could see this being a major threat.

[00:20:27] Paul Roetzer: So I would imagine that the teams at Google are, are kind of playing this out as like, what are the possibilities here? What, you know, what's the impact on our ad revenue? you know, how does this change if adoption of ChatGPT continues to skyrocket? And I don't know what they can, what their market share is in language models, but I mean, you're, the study we did where we asked our users, like 1800 people on our state of marketing, I report this year, I think it was like 55 percent said they had ChatGPT licenses at work.

[00:20:56] Paul Roetzer: And that was by far the number one when we compared [00:21:00] to, Copilot and Gemini. Now that may flip in the enterprise market, like the big enterprises, maybe Copilot's higher adoption. But yeah, I think there's just too many unknowns right now about where this goes. Where search goes, where interfaces go.

[00:21:15] Paul Roetzer: Where language model adoption goes, and that'll kind of dictate this, but I'm sure people at Google are paying attention and, you know, playing this out of possibility, possible paths that this could go. 

[00:21:27] McKinsey Report

[00:21:27] Mike Kaput: Alright, so our next big topic this week. The McKinsey Global Institute recently released a report called, quote, The Next Big Arenas of Competition.

[00:21:38] Mike Kaput: This is a whopping 213 page report that explores 18 what they call arenas. These are future areas that could reshape the global economy. And generate a staggering amount of revenues by 2040. Anywhere from 29 trillion to 48 trillion more in revenue combined by 2040 from these [00:22:00] 18 future arenas. They basically define arenas as industries that transform the business landscape and have two core characteristics.

[00:22:08] Mike Kaput: High growth and dynamism. One of these arenas is particularly important to us and our listeners. It is AI software and services. And the implications of some of McKinsey's data on the growth in this area are pretty interesting. So for the purposes of Their report, McKinsey, defines AI as, quote, a machine's ability to perform cognitive functions that we usually associate with humans, and they kind of include in that both traditional machine learning capabilities, where we're using AI to predict outcomes and behaviors, as well as generative AI applications.

[00:22:45] Mike Kaput: Notably, though, they exclude from this kind of arena, this categorization, they exclude hardware, like NVIDIA chips. So, with all that in mind, here's kind of what McKinsey reports about this arena. [00:23:00] First, investors are flocking to companies developing advanced AI, particularly Gen AI. Equity investments in that technology jumped from 5 billion in 2022 to 36 billion in 2023.

[00:23:14] Mike Kaput: Developments in analytical AI and Gen AI are poised to drive the industry's growth by improving business and worker productivity. In certain modeled scenarios, the arena's revenues grow from 85 billion in 2022 to 1.5 trillion in a lower range of scenarios in 2040, and that can go up to as high as 4.6 trillion in some of the more optimistic scenarios.

[00:23:39] Mike Kaput: That's a compound annual growth rate of anywhere from 17 to 25%. Now, past McKinsey research also kind of analyzed more than 500 uses for analytical and gen AI and estimated their potential economic impact. One of them, analytical AI, such as machine learning and deep learning, could amount to an [00:24:00] estimated 9.

[00:24:01] Mike Kaput: 4 to 15 trillion in revenues by 2040. Gen AI could produce 2. 6 to 4. 4 trillion of economic impact through enterprise use cases. McKinsey highlights that three quarters of that value would be in only four areas. Customer operations is number one. Number two, marketing and sales. Number three, software engineering.

[00:24:22] Mike Kaput: And lastly, R& D. Now, All these enterprise use cases do not even account for all the productivity gains of individual knowledge workers who are automating aspects of their occupations. Incorporating all these cases of individual worker productivity enabled by GenAIn addition to the enterprise use cases, could unlock a total of 6.

[00:24:45] Mike Kaput: 1 trillion to 7. 9 trillion of value annually. So if you add all this together. These components together yield an estimated range of total economic potential from AI software and services of 15. [00:25:00] 5 to 22. 9 trillion annually by 2040. Now, that's a lot of numbers. Some of those are, like, eye watering. I don't even believe them half the time looking at them.

[00:25:14] Mike Kaput: Obviously, Paul, we cannot fully predict this economic impact of AI, like, 15 years from now. Bye. As you've noted in some of your posts and writing from the past week, it is, it seems, safe to assume, we are looking at trillions of dollars being invested in and produced by AI in the coming decade. So that certainly seems to me, like reading all this, that despite all the hype in Gen AI and even some of the disillusionment that's already creeping into the market, it sounds like we're just still so, so early.

[00:25:48] Paul Roetzer: Yeah, and it is like the numbers are confusing and I'm listening to you, say them now. It's like hard to wrap your brain around. I read like the key findings like five times just to try and comprehend what they were [00:26:00] saying. Like, cause they, it's weird. They mix AI revenue and economic impact and then they blend like some new data with a report that they did, I think earlier this year that had to do with, you Where they analyzed 500 AI use cases and I was struggling to like interpret what was the new data versus what was the previous thing they were referencing.

[00:26:20] Paul Roetzer: But I think like to unpack this for people. As you explain, Mike, the key takeaway here is it's huge. It's trillions. And we're at we're just at the start. to try and put it in context, I went and pulled the latest, gross domestic product numbers, GDP numbers for the United States. So the GDP October 30th, 2024 was 29.

[00:26:44] Paul Roetzer: 4 trillion. So again, to put trillions into context, it's huge. The GDP at the moment is 29. 4 trillion, and it's up roughly 5 percent from the previous year. Now, the GDP is the total value of all goods and [00:27:00] services that are produced, in this case, in the United States. within a specific period, which in this case is, you know, an annual basis.

[00:27:07] Paul Roetzer: it is a key indicator of economic health. And it's also when you look to the future and you try and say, well, what's the impact of AI going to be? It is one of the key things you would look at and say, well, what's the impact going to be on GDP? I've had this conversation with a few people who are skeptical of AI having like an outsized impact because on average, you know, the GDP is going to grow Maybe two and a half to three percent in a given year.

[00:27:35] Paul Roetzer: Now we've had a jump in the previous 12 months, but it's pretty common for it to stay in that range. And so if you have someone show up and say, well, it could be 10%, like a year, or in the case of Leopold Aschenbrenner and his situational awareness, he's talking about like, 10, 20, 30 percent per year.

[00:27:55] Paul Roetzer: Yeah. And an annual rate and like compounding over time. And [00:28:00] a lot of people just say, there's just no way. There's no historical context to that, that it's impossible. And so the way that AI could impact GDP, so the economic impact, we'll come back to the revenue, is increased productivity. So we just produce more.

[00:28:16] Paul Roetzer: We can, we output more services. We can output more tangible products. Because we have more time to make more things, basically, because AI is assisting us. So we become more efficient, the efficiency enables us to do more productivity. The other areas, innovation and new product development. The AI is going to help us identify new markets, new product ideas.

[00:28:36] Paul Roetzer: It's going to help us drive innovation that creates growth. as the AI does more and more of people's knowledge work jobs, You have the potential to reallocate the labor force. So we have this kind of finite amount of people that can do work in the United States. I don't know how many people we have in the U.

[00:28:55] Paul Roetzer: S. It's like 300 million or something like 70. A little over 300. And there's about [00:29:00] 136 million full time jobs. So imagine if AI 10 years from now can do 20 to 30 percent of the cognitive labor of those hundred million people that do knowledge work jobs. And you can redistribute those people to do other things.

[00:29:17] Paul Roetzer: So now you have the existing GDP from what we're already doing today. You layer AI's ability to do another 20, 30, 40 percent of that work. And then you redistribute that 20, 30 percent of work to other things. You're increasing the output. So a reallocation of labor into new roles, new markets, new, new businesses.

[00:29:39] Paul Roetzer: you have industry, industry and sector growth. AI enables growth of different industries. They can produce more. And then you potentially have a boost in consumer demand, which drives the growth of outputs because you can personalize experiences and products to them, right? And so these are all like fundamental ways, very tangible ways.

[00:29:58] Paul Roetzer: that you can start to make an argument that AI [00:30:00] will have a massive economic impact on GDP. Now, I mentioned Leopold, like, let's go back real quick to his notes and we'll put this, this was from June of 2024, his situational awareness papers. We'll put the link in the show notes again, but a few key excerpts here.

[00:30:16] Paul Roetzer: He says, As we, and this is one of the things, like, I don't think McKinsey's taking into account AGI and superintelligence. And again, this is one of the big flaws I keep saying we see from researchers and economists is like, they're not considering the future models. They're considering what we know to be true about basically current models.

[00:30:33] Paul Roetzer: But in Leopold's papers, he's saying we're going to get to superintelligence and like when we do, we're going to see massive economic growth. So he says we could see economic growth rates of 30 percent per year beyond, quite possibly doubling a year, each year. This way, and then he does give a caution, this may well be delayed by societal frictions which I 100 percent believe to be true.

[00:30:56] Paul Roetzer: I think it's a large part of why we're not seeing the [00:31:00] massive economic impact right now. And he calls out specifically, arcane regulation might ensure lawyers and doctors still need to be, human, even if an AI system were better at those jobs. Surely, Sam will be thrown into the gears of rapidly expanding robo factories as society resists the pace of change.

[00:31:17] Paul Roetzer: And perhaps we'll want to retain human nannies, all which could slow GDP statistics. Now, a really practical recent example here is, We had the East Coast shut down on a strike, because of the shipping. And it was all over, well, at least in large part, the East Coast. over the future possibility of automating those ports, which is already being done in other countries.

[00:31:44] Paul Roetzer: And so this is, could we ship more? Could we receive more if we were automating the ports? Absolutely. Will we be allowed to automate the ports due to whatever regulations or union or whatever it is that slows that down? That [00:32:00] will be a barrier to GDP growth, even though the AI could do it. And so that's an example where, and I'm not saying right or wrong, either case, I'm just using it as an example of there's going to be friction.

[00:32:12] Paul Roetzer: There will be resistance. And then when you get into the AI revenue, Leopold made the point. Companies will make large AI investments if they expect economic returns to justify it. He gets into trying to project out like OpenAI and Microsoft's revenue. He said one estimate puts Microsoft at 5 billion in incremental AI revenue already.

[00:32:31] Paul Roetzer: He said, every 10x scale up in AI investment seems to yield the necessary return. So we're already seeing this, like the reason they're putting in 500 million, a billion, 5 billion, 10, whatever it is, is because their expectation is a return. They're not doing this just out of like hope that it's going to work.

[00:32:50] Paul Roetzer: They're seeing the results. Leopold then says a key milestone for AI revenue that I like to think about is when will a big tech company, Google, Microsoft, Meta. Hit [00:33:00] a hundred billion revenue run rate from AI specifically. these companies have an on order, on the order of 100 billion to 300 billion of revenue today.

[00:33:10] Paul Roetzer: A 100 billion from AI alone would be a huge, represent, a huge opportunity, and a big fraction of their business. And then you could start to extrapolate out massive growth and he's thinking we might get there by 2026 based on current models. And so he's just pretty much saying like, I like the example Leopold gives about Microsoft.

[00:33:30] Paul Roetzer: He said it may seem like a stretch to like be talking about these crazy numbers, which leads us back to the McKinsey study of like trillions. But he said like how unrealistic really is it if Microsoft has 350 million paid subscribers to Microsoft Office? And as we think about these reasoning engines and these more advanced models, one to two years out, that are basically doing the work of people who are making hundreds of thousand dollars a year, would you as a company be willing to pay a hundred bucks a month for that model?

[00:33:58] Paul Roetzer: Hell yeah. Like, yeah, I mean, [00:34:00] so I haven't, I haven't talked in depth about this yet. We will come back to this, but I built a co CEO GPT. Like I built a strategic partner for myself to run my business. I'm telling you right now, I'd pay a hundred bucks a month to have access to this one custom GPT I built.

[00:34:15] Paul Roetzer: Like it is wild how much value every day I get from this thing as a strategic partner for me to run my companies and like kind of build plans and envision the future. So I think once these tech companies get better at explaining the value of their products in a true value oriented form, not this assumption that I should only pay 20 bucks a month, but what is the value I'm getting out of this as either.

[00:34:41] Paul Roetzer: A replacement or an augmentation to a strategic consultant or advisor, replacement or augmentation to a full time employee that's making 150, 000, 200, 000 a year, a hundred bucks a month is nothing. Like to be able to get that value exchange. And so once these companies get better at demonstrating that and then get way, way better at [00:35:00] onboarding AI literacy, like training and education internally so that the people buying the software are from day one getting value from it, that's where they're all missing the boat right now.

[00:35:09] Paul Roetzer: But as soon as you do that, we are, we are talking about hundreds of billions and trillions of revenue coming from the companies selling these. But then that trickles down to all the other verticals in industry. So I mean, at the end of the day, the thing, as you alluded to, like it's impossible to project this accurately 15 years out, it's possible to project this accurately five years out.

[00:35:30] Paul Roetzer: Like I don't even try and do it five years out, like two, two to three, maybe. but. It's going to have a massive impact, and the thing we keep coming back to, and if you're a listener to this show, a defaulting to taking action every day, saying, how do I get better at this? How do I learn more about AI?

[00:35:53] Paul Roetzer: How do I infuse it more into what I'm doing? How do I help my company move forward? The people who take action [00:36:00] and accelerate their literacy and capabilities have the greatest chance of thriving as we move forward, and the same holds true for businesses. We are, we are in the very, very early innings of this, probably the top of the first inning to use a baseball analogy of like adoption and transformation from AI, but it is It is going to be massive, and so McKinsey studies, like studies like this from McKinsey help put in context how massive, but it's all a guessing game.

[00:36:28] Paul Roetzer: It's big though. 

[00:36:30] Mike Kaput: Yeah, and we talk about that a lot, that it's probably more important to just be directionally correct. Like whether it's 10 trillion or 100 trillion, honestly, doesn't really matter to your average business leader. But do we know today that double digit productivity gains are possible across a wide range of knowledge work?

[00:36:45] Mike Kaput: Yes, we do. What would that be like two years from now? It could be triple digit kings. I mean, so that's the directional trend. Correct. Act accordingly. Yeah. And yes, take, take action. Do [00:37:00] not wait around. Yeah. All right. 

[00:37:02] Microsoft AI CEO Interview

[00:37:02] Mike Kaput: Our third big topic this week, the popular podcast Masters of Scale just dropped a new interview with AI leader Mustafa Suleyman that we think is well worth paying attention to.

[00:37:13] Mike Kaput: So Masters of Scale is a podcast primarily hosted by Reid Hoffman, The co founder of LinkedIn, a former OpenAI board member, current Microsoft board member. He is also, with Mustafa, the co founder of the AI company Inflection. So, Mustafa is the former co founder and CEO of that company. That company basically got acquihired by Microsoft.

[00:37:35] Mike Kaput: Now, Mustafa is the CEO of Microsoft AI, which means he's not only on the bleeding edge of AI development, he's also a key player in both Microsoft's AI strategy and something we'll talk about a little bit more this episode, the company's relationship with OpenAI. Mustafa is a key player in governing how that works.

[00:37:55] Mike Kaput: So, Paul, you kind of found some things that jumped out here in this [00:38:00] episode that were worth noting. Can you walk us through them? 

[00:38:03] Paul Roetzer: Yeah, so I do think it's worth listening to, and as you mentioned, in the context of Suleyman's role with Microsoft, it is very instructive of where they're going. And so as one of the key players, it's definitely worth paying attention to.

[00:38:19] Paul Roetzer: So I'm going to highlight a few of the key points, but one he talked about up front is recursive self improvement. Because the key thing with this interview is He was kind of time stamping these key elements and when he thought they might be occurring. So recursive self improvement is basically the ability for the models to, to identify their own weaknesses and flaws and hallucinations and fix themselves.

[00:38:41] Paul Roetzer: This sounds awesome on the surface, it also sounds terrifying, because this is one of the things that the Doomer side worries about, is these things develop the ability to recursively self improve, Which could enable, like, a fast takeoff where they [00:39:00] just become really, really smart, really, really fast when they can fix themselves, basically, and find these slots.

[00:39:06] Paul Roetzer: So, he said he sees that coming into view in 2025, that teams will start experimenting with that. We have heard about this from, I think, Claude, they've talked a little bit about it. I think we heard a little bit about this with O1 preview from OpenAI, as you develop the ability to do reasoning and chain of thought, part of that process is to identify when the chain of thought breaks, when it's no longer true, when like a falsehood has been found or misinformation is found within it, to be able to go back and fix that.

[00:39:41] Paul Roetzer: And so that is something to watch. It is, you know, we talk a lot about that Andrej Karpathy. Intro to LLM's YouTube video from January 2024, recursive self improvement is one of the things he got into. one of the other things I was not surprised to hear them talk about was EQ versus IQ. So, [00:40:00] intelligence, versus, you know, emotional quotient versus, intelligence quotient.

[00:40:05] Paul Roetzer: And, or emotional intelligence. And the key here is this is what he was trying to build an inflection. That's why I wasn't surprised at all. And they're that they're bringing that to Microsoft. So IQ is what these models are really good at. It's often how they measure them is their cognitive abilities, intellectual potential, logic, problem solving, math skills, analytical skills, reasoning, language, comprehension.

[00:40:27] Paul Roetzer: Those are all the natural things. But emotional intelligence is more the ability to recognize, understand, manage, and influence one's own emotions and those of others. Self awareness, self regulation, motivation, empathy, social skills, not what you would traditionally expect from an AI. And so in his case, he said, it turns out that actually tone, style, emotional intelligence of these models.

[00:40:51] Paul Roetzer: the extent to which they will ask you questions, the extent to which they reflect back on the type of language that you might use and so on, that delivery vehicle for the [00:41:00] substance is perhaps more important to the majority of consumers than just an objective regurgitation of Wikipedia, he said. He went on to say, I think that's going to be one of the key capabilities.

[00:41:08] Paul Roetzer: I think everyone's starting to wrestle with that now, as we look at this agentic future. is what is the personality of these models and the good and bad. I mean, we talked about, you know, the challenges of when these things become too human like when they have a high emotional intelligence, that if that is not properly protected or contained, then you have humans developing unhealthy, deep relationships with AIs that seem very human like.

[00:41:38] Paul Roetzer: So this is a very slippery slope. I mean, right now we've just talked about emotional intelligence and Recursive self improvement. Now, Mustafa talks about these things as inevitabilities and like that they're actively looking to build this in. Now, keep in mind, Microsoft may do both of those things more ethically than others will, or more ethically than maybe an open source model would [00:42:00] allow third parties to build.

[00:42:02] Paul Roetzer: He got into AI agents. He said, The first step for the agentic future is that your co pilot has to, have the ability to see. This becomes extremely important. We've talked multiple times about Microsoft's efforts with this. We have Project Astra from Gemini, OpenAI is working on this. They want these things to be able to see the screens.

[00:42:21] Paul Roetzer: Claude, we, you know, computer use we talked about. So, he says the AI companion has to be able to see, and having an aide or an assistant or a companion that is really seeing the pixels that you see on the screen in your browser, your desktop, your phone. Microsoft means there's a new level of sort of awareness about your sensory input that enables the companion to observe what you're seeing and be able to do things.

[00:42:42] Paul Roetzer: Now, ironically, last week Microsoft's co pilot Twitter account tweeted, If only your browser could see what I see. Oh wait, co pilot Vision will be able to very soon. So, this is aligned with what Microsoft is saying they're going to do. [00:43:00] so that was an important one. Memory was another one. We've talked a lot about memory being one of the next key on locks.

[00:43:07] Paul Roetzer: So, memory is the ability to go from Conversation to conversation and inflection or ChatGPT or Microsoft Copilot and have the AI remember everything and be able to personalize everything it's done based on your history. Now OpenAI has been working on this. They've talked a lot about memory. But he said, we're going to nail memory.

[00:43:27] Paul Roetzer: I mean, I'm really confident 2025 memory is done. Permanent memory. If you think about it, we already have memory on the web. Copilot has really good citations. It's updated about 15 minutes ago. Knows what happened in the news on the web, so on. And we're just compressing that to do it for your personal knowledge graph.

[00:43:47] Paul Roetzer: When you add in your own documents, emails, calendars, stuff like that, memory is going to completely transform experiences. because it's frustrating to have a meaningful conversation with like a Gemini ChatGPT inflection. And [00:44:00] then go on an interesting exploration around some creative idea and then come back three or four or five sessions later and you have to start again.

[00:44:06] Paul Roetzer: Like it doesn't remember anything. So memory is a really key thing. You also talked about models. So the good news is the models are getting bigger and smaller, which we've talked many times on this show about, that they're going in both directions. But he does think that the biggest models have a lot of room to go, that there's plenty more data that they can infuse into these things.

[00:44:25] Paul Roetzer: They're not going to see a slowdown in the frontier models for at least the next few years. And so the frontier models are going to kind of have an outsized impact, but the smaller models are going to be critical to the future. I like the example he, he used here, which we've shared before, like why small models make sense.

[00:44:43] Paul Roetzer: He said small is definitely going to be the future because if you think about it, the very large model, when you ask a query of these frontier models, it's lighting up the neural representations of billions of pathways, which are not relevant to the query at that hand. It's like if someone asks me something and my entire [00:45:00] brain fires to answer it, that's not what happens in our brains.

[00:45:03] Paul Roetzer: There's very small pieces of our brain tied to specific cognitive tasks. And so our brains are highly efficient because we don't use every neuron for each cognitive thing. Right now, the large language models are basically firing every neuron when you say, What was the score of the Cavs game last night?

[00:45:19] Paul Roetzer: Like, every neuron is kind of And so they're trying to build these small models that allow for these things. And then I'll kind of end with his final thoughts, and this sort of aligns with what we just talked about with the potential of, you know, growth and economic impact. So, I think that this is a moment to found companies, scale companies.

[00:45:37] Paul Roetzer: It's a moment to really pivot careers, even if you're not an entrepreneur, even if you're an activist or an organizer or an academic. This is the moment to really pay attention, because by 2050, the train will have left the station, it'll be quite different. And this is a moment where we really do have a chance collectively to shape and influence things, and nothing is predetermined.

[00:45:59] Paul Roetzer: [00:46:00] It's really, it really is within our reach to shape it for the best of humanity, and I think that's quite, we're very lucky to be alive at this moment. It feels incredibly empowering, and it's a great responsibility. Oh, I don't agree with everything Mustafa says, and there's elements of what they're working on that I don't, I'm not that excited about, as a, as a human.

[00:46:20] Paul Roetzer: I like his thoughts at the end, and I kind of echo those, and I've, I've said it many times, like, the only way we do what we do, and I think as much as I do about this, is because I believe we have a possibility of an incredibly abundant future, and I choose to be optimistic about it, despite the concerns and fears and the uncertainty I don't find worrying constantly about that stuff, does me any good.

[00:46:45] Paul Roetzer: And so I choose to try and take actions to ensure the greatest possible outcome for myself and my family and everybody else. And that's kind of how I keep going each day with this stuff. 

[00:46:56] Mike Kaput: Yeah, I feel like we need a regular segment on how to stay [00:47:00] optimistic and keep our sanity. Keep your sanity while covering this space.

[00:47:06] Sam Altman’s 20VC Interview

[00:47:06] Mike Kaput: All right, let's dive into some rapid fire for this week. So first up is something that actually hit the docket like right before we started recording, because a new interview with Sam Altman just dropped. It was an interview on the 20VC podcast, which we've talked about before with a host who's an investor named Harry Stebbings.

[00:47:25] Mike Kaput: And basically over 35, 40 minutes, they covered a ton of ground. They covered things like the trajectory of model improvements. They talked about whether or not scaling laws will continue. And also, Altman had some really first draft predictions about where we're headed, among many other topics, so, Paul, I guess I was going to kinda maybe Pull out a few things that jumped out at me and, you know, if you have any thoughts on some of these topics, we'd love to kind of hear you putting them in context.

[00:47:51] Paul Roetzer: Yeah, it sounds good. I'd see, so I had seen some clips of this last week because it was done live and then they published it on Monday morning [00:48:00] as a, a full podcast. So I have not had a chance to listen to the full podcast. I just saw some clips of it last week. So yeah, I'm kind of like following along with the audience on this one and seeing what you pull out of it.

[00:48:11] Mike Kaput: So, and again, I would highly encourage you to listen to the whole thing. Like we've said before, listening to what these key players are saying is really important. This is not a very long podcast, relative to some of the longer form stuff out there. So, I'm just going to cover a few things that I found to be important.

[00:48:28] Mike Kaput: So, first up is, it's very clear OpenAI is betting really heavily on O1 slash reasoning. Sam said, we want to make things better across the board, but this direction of reasoning models is of particular importance to us. He says it's going to unlock all sorts of possible things that can drive things forward.

[00:48:46] Mike Kaput: He said, quote, so you should expect rapid improvement in the O series of models, and it's of great strategic importance to us. Interestingly, this is a great question. I loved Harry asking. He said, How do you think [00:49:00] about the definition of AI agents today? What is an AI agent to you? And he's basically getting at a point that we've talked about at length, which is the semantics around agents.

[00:49:08] Mike Kaput: The, how we talk about them is getting very muddy and unclear. So Altman said, This is like my off the cuff answer. It's not well considered. But, something that I can give an agent, he's saying, is something that I can give a long duration task to and provide minimal supervision during execution for. Now, Stebbings then asked as a follow up, what do you think people think about agents that actually they get wrong?

[00:49:36] Mike Kaput: Altman goes on at length to kind of describe, look, normally you hear people talking about agents as like this thing that'll go do stuff for me on the internet, you know, go like order from a restaurant or something, or book me a flight, and he's like, the category I think though is more interesting is not the one that people normally talk about, where you have this thing calling restaurants for you, but something that's more like a smart senior coworker.

[00:49:57] Mike Kaput: where you can collaborate on a project [00:50:00] and the agent can go like do a two day task or two week task really well and ping you when it has questions but come back to you with like a great work product. Now, Paul, that's kind of his first draft thinking about it. I found that to be much more helpful in kind of formulating how I think about a definition of agents.

[00:50:18] Paul Roetzer: Yeah, and it's, I mean, part of it plays into this further confusion of what an agent is and how we're going to define these things. He's definitely aligning his view of it more with their reasoning models, their O1 models. You know, it's interesting. I haven't really thought about this until now, but when we go back to the previous conversation on what would you be willing to pay per month, especially if you're talking about like senior level support.

[00:50:42] Paul Roetzer: I almost wonder if there isn't a value based model kind of more similar to like a Fiverr or one of those task oriented sites where there's a marketplace where you say, I'm willing to pay, you know, 500 for a logo. I wonder if there isn't a way for ChatGPT and Gemini and others to have this value based model where you say, [00:51:00] here's the problem I'm trying to solve.

[00:51:02] Paul Roetzer: I'm willing to pay 15, 000 to solve this. And then you have your agent in their world, your reasoning model that maybe goes off for a day or a week or a month to work on that problem. Because the cost of inference is going to be higher as it has to go through dozens or hundreds of steps. And so I just wonder if there isn't a marketplace for agents that are value based where you You just let the market decide what they're willing to spend to solve something.

[00:51:30] Paul Roetzer: And I don't know, it's kind of a fascinating idea. I don't know if they've considered that approach or if people are thinking about that. But I think as these tech companies, like I said earlier, go back to a better approach to educating people on the value these models create and get away from this expectation that's 20 or 30 bucks a month for all the value.

[00:51:47] Paul Roetzer: I wonder if there isn't a way to do that where you just present a problem and he's willing to pay to solve because otherwise I got to hire an advisor or a consultant or, you know, a staff member to do this thing. 

[00:51:59] Mike Kaput: Yeah, that's [00:52:00] really interesting. and it's interesting too how much of this conversation somewhat parallels kind of what Mustafa was saying because as part of this as well, Sam Altman was asked about like, hey, where do you think we're going in the next year or two?

[00:52:12] Mike Kaput: And he said, among other things, quote, without spoiling anything. I would expect rapid progress in image based models, which was kind of what Mustafa was also getting at. he was asked, when we think about scaling models, like scaling these AI models, how many more model iterations do you think scaling laws will hold true for?

[00:52:31] Mike Kaput: He said, without going into detail about how it's going to happen, the trajectory of model capability improvement is going to keep going like it has been going, and I believe that it will be doing that for a very long time. So that's also very interesting that he's committing to 

[00:52:45] Paul Roetzer: that. Yeah, it was funny.

[00:52:47] Paul Roetzer: He tweeted, I think this was on Sunday or something, November 2nd. I heard O2 gets 105 percent on GPQA. So, I wasn't really sure what GPQA was. a [00:53:00] graduate level Google proof Q& A benchmark. So, there was a research paper in November 2023 about a challenging data set with 448 multiple choice questions written by domain experts in biology, physics, and chemistry.

[00:53:11] Paul Roetzer: and then Sam tweeted, damn, wrong account, and he was kinda, so I think he's being, like, trying to be funny, like, he has Burner accounts, and he usually tweets the stuff from Burner accounts. But, you know, I think, again, there's always some element of truth to what Sam does with this stuff. He's always joking around where, you know, these models are going to get smarter and they're pretty confident in their ability to make them significantly smarter.

[00:53:36] Mike Kaput: And then I'll end with this final, kind of comment he had and they, you know, Harriet asked him like, Hey, what is the five, 10 year horizon look like for open AI, for AI in general? I won't read his whole answer, but he kind of made this really important point that he's like, You know, I think in five years it looks like we have an unbelievably rapid rate of improvement in technology itself.

[00:53:59] Mike Kaput: He then [00:54:00] goes on to say the pace of progress is totally crazy and we're discovering all this new stuff both about AI research and also about the rest of science if we hit kind of this AGI moment. And that feels like if we could sit here now and look at it would seem like it should be very crazy.

[00:54:15] Mike Kaput: But then the second part of the prediction is that society itself actually changes surprisingly little. An example of this would be that if I think if you ask people five years ago, if computers were going to pass the Turing test, they would say no. And in fact, we kind of roughly speaking have already passed the Turing test and society didn't change that much.

[00:54:34] Mike Kaput: He said, it just sort of went whooshing by. That's the kind, kind of example of what I expect to keep happening. Which is progress, scientific progress keeps going, outperforming all expectations, and society in a way that I think is good and healthy. So. He is sticking to his plot points of things are going to move very, very quickly.

[00:54:55] Mike Kaput: Yep. Yep. 

[00:54:56] Paul Roetzer: As 

[00:54:57] Paul Roetzer: I expected the interview would say. 

[00:54:59] More AI Agent News

[00:54:59] Mike Kaput: All right. [00:55:00] So next up here in our rapid fire session is we've got a bunch more news this week about AI agents, which are always a hot topic. So first up, Salesforce has announced the general availability of AgentForce. This is its new AI layer that allows companies to build.

[00:55:15] Mike Kaput: And deploy autonomous agents that can take action across business functions. And while Salesforce is gathering steam, Google has revealed a bit more of a measured timeline for its AI agent ambitions. At the same time, CEO Sundar Pichai has announced that Project Astro, their AI agent initiative, won't launch until 2025 at the earliest.

[00:55:37] Mike Kaput: So this is aiming to kind of create AI assistants that can both understand the world through smartphone cameras and then perform complex tasks. That was first demoed by Google at their I O developer conference in May of this year. LinkedIn is getting into the AI agent game. They are releasing their first autonomous tool, Hiring Assistant.

[00:55:56] Mike Kaput: This is an AI recruiting agent that can handle everything from [00:56:00] writing job descriptions to sourcing candidates and engaging with them. And finally, in the startup world, A company called Sierra, which is led by the former Salesforce co CEO Brett Taylor, is making waves with a possible funding round that could value the company at over 4.

[00:56:17] Mike Kaput: 5 billion. They have AI agent technology that focuses on automating customer service tasks. So kind of bringing it back full circle to that McKinsey report, which showed that customer service slash support was one of these big, big areas of opportunity. So, Paul, it certainly seems like we can soon expect AI agent capabilities, however people are defining them, to be in a bunch of these platforms.

[00:56:42] Mike Kaput: Did any of these like jump out at you as particularly noteworthy? 

[00:56:45] Paul Roetzer: Yeah, it's interesting. LinkedIn's starting to kind of get in the game and infuse, I mean, obviously they're owned by Microsoft, so they're, you know, they have access to a lot of the technology and I think you're going to start seeing more and more of that stuff infused into LinkedIn.

[00:56:57] Paul Roetzer: The Sierra one, back on episode [00:57:00] 116, we talked about like an interview with Brett Taylor that might be worth revisiting for people, but I actually laughed, someone had a, one of those AI agent landscapes already, and I was like, oh man, we already hit the hundreds or thousands of agents in a landscape moment, so.

[00:57:19] Paul Roetzer: there's no turning back now, so I don't, I don't know how in the world you'd keep that, that landscape accurate and updated, but yeah, agents, whatever they actually are and however we end up defining them are going to become a key part of your life and the future. sometime between now and November 20th, I'm going to figure out what I have to say about them because that's what my opening keynote is for the AI for Agencies Summit.

[00:57:42] A16z + Microsoft on AI Regulation

[00:57:42] Mike Kaput: So in some other news, Microsoft and Andreessen Horowitz have joined forces to come out against burdensome AI regulations. So Microsoft CEO Satya Nadella and Microsoft President slash Chief Legal Officer Brad Smith Joined A16Z, as Andreessen Horowitz is [00:58:00] colloquially known, they joined Mark Andreessen and Ben Horowitz, the two head people at this firm, in publishing a joint policy statement that advocates for less AI regulation.

[00:58:13] Mike Kaput: So key aspects of this position, which was published online, include pushing for market based approaches over government regulation. Advocates for unrestricted access to data for AI training. Argues that machines should have the, quote, right to learn, similar to humans. Supporting open source AI development is another piece of this, while opposing regulatory frameworks that might restrict it.

[00:58:36] Mike Kaput: And they also call for regulation only when benefits outweigh costs, with the industry determining those calculations. This statement also notably reframes copyright concerns around AI training, suggesting that copyright law shouldn't prevent AI systems from using data to learn. So the companies are framing their position as protecting innovation and startups, but [00:59:00] critics point out that their stance primarily serves to prevent meaningful oversight of AI development.

[00:59:06] Mike Kaput: So Paul, this stance that they have on these regulations doesn't particularly surprise me. but why are we getting a joint formal statement from these two companies now? 

[00:59:16] Paul Roetzer: I, the only thing I could come up with is related to the election. I don't know. Like, I think they're, timing wise, obviously in the United States, the election is November 5th.

[00:59:26] Paul Roetzer: So, it seems like they're just getting out ahead of this from, whichever administration's going to be in office standpoint, sort of stating their claims. I don't know. Not sure what else is happening that would time this. I, there was nothing new in it, per se. I mean, we kind of knew their points of view.

[00:59:40] Paul Roetzer: the big one, just to reiterate, is. They want regulations to focus on the application layer, not the model layer, meaning, I think the, you know, let, let's, I don't like using the guns example, but, um. electricity, nuclear, whatever, that like, [01:00:00] it's, it can be good or bad, it's how you use it. And they want AI models to be treated that way.

[01:00:06] Paul Roetzer: That the models can do good or bad, and it is the application of the models that should be regulated and penalized, where they're used. And then the copyright one I thought was interesting to just blatantly come out and say, like, we don't think copyright should be a thing. Like, we don't think that should prevent at all that the AI models have the right to learn just like humans.

[01:00:26] Paul Roetzer: I don't necessarily agree with that, standpoint, but there's a lot of things A16Z would say that I don't necessarily agree with, but there's also things they say that make a lot of sense. And then the one that I most aligned with, I would say, at the end, is they said how people thrive in an AI enabled world.

[01:00:44] Paul Roetzer: This very much aligns with how we talk about it, said, policy should fund digital literacy programs that help people understand how to use AI tools to create and access information. It should also support workforce skill development and workforce retraining programs to help people secure jobs in an AI driven [01:01:00] world.

[01:01:00] Paul Roetzer: I'm 100 percent on board. That final point. Nice. 

[01:01:06] OpenAI vs. Microsoft

[01:01:06] Mike Kaput: Alright, so next up we also got some more comments from Sam Altman in a Reddit a MA, and he revealed that the company's AI projects are facing significant delays because they simply don't have enough computing power. So. Apparently, a computing bottleneck is affecting several high profile OpenAI projects.

[01:01:24] Mike Kaput: that includes Sora. reports suggest the current version of Sora takes over 10 minutes to generate just one minute of video. And the computing constraints are also impacting things like the vision capabilities for advanced voice mode in ChatGPT, which have been postponed indefinitely, it sounds like.

[01:01:42] Mike Kaput: And it turns out that April demo of VoiceMode was rushed to compete with Google's I. O. conference. Despite internal concerns, the technology was not necessarily ready, which leads to this postponement and also the fact that VoiceMode did not come out on time. So, [01:02:00] OpenAI is apparently working to address these limitations.

[01:02:02] Mike Kaput: Reuters has reported the company has been collaborating with Broadcom to develop its own AI chip. Altman has said the company, in the meantime, is focused on improving its O1 series of reasoning models. And he also said, quote, we have some good, very good releases coming later this year. Nothing that we are going to call GPT 5, though.

[01:02:22] Mike Kaput: So, Paul, let's connect some dots here. Like, why is this a problem for OpenAI right now, and how much of this is related to what we've talked about in past episodes about the relationship with Microsoft, who's supposed to be helping them keep access to capital and compute? 

[01:02:40] Paul Roetzer: Yeah, I would say, I mean, one, one part you could read into this that Microsoft is purposely throttling access for some reason.

[01:02:49] Paul Roetzer: Yeah. I don't know that that's true. I think the more likely scenario is that Microsoft has their own vision and AI ambitions now with Mustafa at the head. [01:03:00] And they have far greater need for access to compute themselves to be doing what they're doing. And so there's only so many NVIDIA chips to go around and so many data centers to build on.

[01:03:11] Paul Roetzer: So, I think it's just that the demand for this stuff is massive. And so Microsoft, I believe we talked about like, allowing OpenAI to do deals with like Oracle and others to get more compute. And obviously, Sam's trying to, you know, Rangle, their own compute moving forward. So I think that'll kind of continue on as a story.

[01:03:32] Paul Roetzer: I think, you know, they've said before they probably weren't going to announce GPT 5, but I don't know if like 01 when they get out of just preview mode, I'm not so convinced that 01 isn't sort of the next model. They just aren't going to call it GPT 5. So I think we're probably seeing the elements of GPT 5 coming to light, that just not under that name.

[01:03:56] Microsoft AI Coding

[01:03:56] Mike Kaput: So some other Microsoft related news, [01:04:00] GitHub, which is owned by Microsoft, has announced that it is expanding beyond its exclusive relationship with OpenAI by bringing multiple AI models to their popular Copilot coding assistant. So GitHub Copilot will now integrate three major AI providers, Claude 3. 5 Sonnet, Gemini 1.

[01:04:19] Mike Kaput: 5 Pro, and OpenAI's latest O1 Preview and O1 Mini models. So, this timing is kind of notable because it comes amidst all these reports that OpenAI is worried about Anthropic taking the lead over them in code writing capabilities. Now, GitHub CEOs has framed this as a response to developer demand for choice.

[01:04:43] Mike Kaput: He said, we're not saying one model is better than the other. We believe it's for developers to decide. So Paul, we know that like the major AI companies are pretty seriously interested in creating models and products to aid developers. So totally natural, there's going to be a [01:05:00] bunch of competition in this space.

[01:05:03] Mike Kaput: But is this another sign still that there's trouble in this relationship between Open AI and Microsoft? m

[01:05:10] Paul Roetzer: it was, it seemed like this was unexpected. I would imagine OpenAI was aware this was happening, but it was pretty big news in the AI developer world when this occurred. And so again, I'm not sure that that world was expecting it to happen, but I don't, think that this is an insignificant move by Microsoft and GitHub to enable this for the developer community.

[01:05:35] Paul Roetzer: Something to keep an eye on in the ongoing OpenAI Microsoft relationship. 

[01:05:41] Apple Intelligence Rollout 

[01:05:41] Mike Kaput: Alright, so next up, Apple has an AI rollout that is facing some early growing pains. The company has kind of taken this staggered approach to releasing Apple Intelligence features. The first wave, which is arriving in iOS 18. 1, brings kind of modest improvements that many users might find [01:06:00] underwhelming compared to all the promises we heard at WWDC.

[01:06:04] Mike Kaput: The initial release of Apple Intelligence includes basic features like writing tools for text editing, improved Siri interactions, smart replies, and messages. Nothing that's really kind of a wow moment here. Now, more transformative features are being held for iOS 18. 2 in early December. This includes things like ChatGPT integration, an image playground for AI image generation, and visual intelligence for real world object recognition.

[01:06:31] Mike Kaput: Now, the company is kind of making this bet that even if they're late to market, their implementation will be more secure, more reliable than competitors. So Paul, you are kind of an Apple power user. Like what do you make of the Apple intelligence rollout so far? 

[01:06:47] Paul Roetzer: It was wild. So October 28th is when it came out.

[01:06:51] Paul Roetzer: So I did, I think I said in the last episode, I finally went and bought the new iPhone. And knowing August or October 28th is when they were going to release 18. [01:07:00] 1, I get the new phone, the phone's the same phone as the previous phone, basically. I had a 14, there was like no noticeable difference initially, so I download the features and I'm like, or I download 18.

[01:07:11] Paul Roetzer: 1 and I'm like, where is it? Like, I thought this thing was supposed to glow when you talk to SirIt's not glowing. And so I even went to Surrey, I was like, what, is this the new Surrey? Like, is this the new features? Like, where is it? And I was so confused. Like, I had no idea why I didn't have these features.

[01:07:27] Paul Roetzer: So I finally go into like my settings and I find this Apple intelligence feature in there and I click on it and then it says, join the wait list for Apple intelligence. I was like, join the wait list? What are you talking about? Like, why do I have to join a wait list? It is, it's so bizarre. Like the whole experience.

[01:07:44] Paul Roetzer: And I just kept thinking, I was like, but I like, I talk about this stuff and follow it for a living. And I didn't know I had to go in and join a wait list to then get like about 36 hours later, I think it showed up and then you get in there. You start playing with, it's like, Oh my gosh, this is [01:08:00] it. Like, this is what running all these TV ads for and having Tim Cook personally tweeting about the Apple intelligence age, like, Oh, it's, it's just bad.

[01:08:10] Paul Roetzer: Like, it's so. It's so disappointing. So, I don't know, I guess we'll wait till December 2nd and see what they come out with then, and then they've announced like 18. 4 will be in April of next year, and that's when maybe Surrey actually gets better. I don't know, but it looks like this is going to be a long game for sure, and it is highly disappointing, I would say, thus far.

[01:08:33] Runway CEO Statement on AI Companies

[01:08:33] Mike Kaput: Alright, our final piece of news this week, we got a pretty thought provoking post from the CEO of Runway, the AI video generation company, his name is Cristobal Valenzuela. And in this post, he argues that we have reached the end of what he calls the, quote, AI company era. Not because AI has failed, but because it has become fundamental infrastructure, much like electricity or the internet.

[01:08:58] Mike Kaput: He argues this [01:09:00] transformation marks a crucial shift in how we should think about AI and its role in business. So, as a result, he's kind of announced that Runway, which is originally known as an AI company, It has basically reframed itself as a media and entertainment company with a singular vision to use AI as a fundamental tool for storytelling.

[01:09:20] Mike Kaput: So Valenzuela here kind of draws a historical parallel comparing their work to the invention of the camera. It's not just a device for capturing images. But it's a catalyst that spawned entire industries in cinema, television, social media, etc. So basically they have this new vision where they see AI as infrastructure rather than an end goal.

[01:09:41] Mike Kaput: Similar to how companies kind of stopped being internet companies once the web became universal. They want to focus on creating new forms of expression and storytelling rather than just simply advancing AI technology. He talked a bit about the concept of what he calls quote universal simulation and world [01:10:00] building where content can dynamically generate itself in response to viewers.

[01:10:04] Mike Kaput: And they're looking to break down the traditional one way media consumption models in favor of interactive and generative experiences. So basically they're saying the next wave of innovation here won't just come from improving AI models, but it'll come instead from companies that kind of understand how to use AI To create new forms of media and experiences.

[01:10:27] Mike Kaput: So Paul, for a pretty short post, there's like a lot to unpack here. Like, I want to kind of take this in two parts. Like, first, what do you make of his claim that we've reached the end of AI companies? And then, what does this actually mean for like, Runway's direction and focus? 

[01:10:44] Paul Roetzer: I'll actually go in reverse here.

[01:10:45] Paul Roetzer: I don't think it means anything to their direction and focus. It's just how he wants to describe it. The reality is he's living in a bubble. Like, so I honestly, so when I started Marketing Institute in 2016, I thought by 2020, I wasn't sure we were going to need the name, like AI in the name. Like I thought [01:11:00] Marketing Institute might just be redundant.

[01:11:02] Paul Roetzer: And like, everything's just going to be an AI company and we won't need to differentiate. I was very wrong on that. It was like Elon Musk, you know, projecting full self driving by 2020. Like, we were just off. Like, and so I'm off by five years at this point. Like, we're still nowhere near companies not needing to differentiate whether they're AI forward, AI first, whatever you want to call it.

[01:11:23] Paul Roetzer: Um So, I don't agree with him that we don't. I mean, in his world, fine. Like, I get that he wants to, you know, just be known as a media and entertainment company, but the reality is the fact that they're an AI first, AI native, whatever you want to call it, media entertainment company is what makes them different.

[01:11:39] Paul Roetzer: So, he doesn't need to position it that way if he chooses not to, but there's the vast universe of media entertainment companies aren't AI literate yet. Like, they're not really doing this and infusing it in. So, The fact that somebody is or is not using AIs is actually different. And you can do that for healthcare, law firms, marketing [01:12:00] agencies, SaaS companies.

[01:12:01] Paul Roetzer: Like, there's plenty of software companies I talk to who I wouldn't even consider AI forward yet. Like, and so it's a differentiator. Like when we're, when we're considering our own tech stack at the Institute, I want to know whether, I don't want to know it's a software company. I want to know it's an AI first or AI native or.

[01:12:18] Paul Roetzer: AI4, whatever you want to call it, that they're infusing AI and intentional about, and they have a vision for building smarter software, like, that makes them different. And it will for the foreseeable future. So I don't agree with him that we've arrived at a point where companies are just companies again.

[01:12:32] Paul Roetzer: That is, we're nowhere near that yet in most industries. But again, for their positioning, fine. Like you can say whatever you want. To me, Runway is an AI company, like, I don't, it's going to continue to be. It's, One of the first ones we started following back in 2018, 19, and it's still what makes them unique is their integration of these models to do amazing things and move their industry forward.

[01:12:53] Paul Roetzer: But getting rid of AI and that just, it's just a personal choice. So yeah, I mean, [01:13:00] it is worth following. I think they do amazing stuff. It's a company we've been paying attention to for a really long time. I don't, I don't agree that that is a relevant positioning for most people that would listen to our.

[01:13:15] Mike Kaput: Alright, Paul, that is a wrap on this week. We've got a big week ahead of us, probably some AI news, some non AI news and stuff tomorrow as well. Yeah. It's going to be making, dominating the airwaves, 

[01:13:28] Paul Roetzer: but Just saying, everybody, whatever, whatever, whatever direction you're going with the election, whether it goes your way or not, remember, we're all in this together and like, we all gotta pick up the pieces the next day and do our thing, whether you're candidate one or not, so just be kind to each other and, um You know, I, we don't want this show to ever be political, but I'm just going to be from a human perspective saying like, let's, let's move forward together in a positive direction, whether you're in the United States or globally, whatever, but yeah, it's, uh.

[01:13:56] Paul Roetzer: It's 

[01:13:57] Mike Kaput: going to be an interesting week when we're the third. No kidding. And if [01:14:00] you want a distraction from anything going on this week, go take a second to leave us a review. If you have not, we really appreciate all feedback we get. It helps us make the show better. So if you haven't done that and you have the ability to do it, please do that.

[01:14:15] Mike Kaput: And last but not least, go check out the Marketing AI Institute newsletter. It has all of this week's news in AIncluding a bunch of stuff we did not get to on today's podcast. So go to marketingaiinstitute. com forward slash newsletter. Paul, thanks again. 

[01:14:32] Paul Roetzer: Thank you, Mike, and thanks, everyone, for listening.

[01:14:33] Paul Roetzer: We will be back next week with our regular weekly episode. 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.[01:15:00] 

[01:15:00] Paul Roetzer: Until next time, stay curious and excited. Explore AI.
Claire Prudhomme

Claire Prudhomme is the Marketing Manager of Media and Content at the Marketing AI Institute. With a background in content marketing, video production and a deep interest in AI public policy, Claire brings a broad skill set to her role. Claire combines her skills, passion for storytelling, and dedication to lifelong learning to drive the Marketing AI Institute's mission forward.

Related Posts

[The AI Show Episode 123]: Trump AI Policies, Problems with OpenAI’s New Model, GenAI for Business Strategy & Visa’s 500 AI Use Cases

Claire Prudhomme | November 12, 2024

From Trump's AI policy shakeup to OpenAI's Orion challenges and practical GenAI planning breakthroughs. Mike and Paul decode this week's AI transformation in Ep. 123 of The AI Show.

Podcasts

[The AI Show Episode 117]: OpenAI’s Wild Week, Sam Altman’s Prophetic Post, Meta’s AR Breakthrough & California’s AI Bill Veto

Claire Prudhomme | October 1, 2024

OpenAI's turbulent week, Sam Altman's predictions, Meta Connect 2024 highlights, SB-1047 updates, FTC's AI crackdown, Anthropic's valuation, and more industry news dissected.

Podcasts

[The AI Show Episode 121]: New Claude 3.5 Sonnet and Computer Use, Wild OpenAI "Orion" Rumors, Dark Side of AI Companions & Ex-OpenAI Researcher Sounds Alarm on AGI

Claire Prudhomme | October 29, 2024

Episode 121 of The Artificial Intelligence Show talks about the new Claude 3.5 Sonnet, Claude's computer use, OpenAI "Orion" rumors, and much more.

Podcasts