<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=2006193252832260&amp;ev=PageView&amp;noscript=1">

56 Min Read

[The AI Show Episode 91]: Accenture On Track to Make $2.4 Billion from Generative AI, The AI Industry’s Legal Problems, and How the Public Sees AI

Featured Image

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

Learn More

Despite a solar eclipse, nothing could overshadow the latest AI news! This week on The Artificial Intelligence Show hosts Mike Kaput and Paul Roetzer discuss the implications of Accenture’s earnings from Generative AI, looming legal issues within the AI industry, public perceptions of AI, Amazons future in the AI race and more.

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

Today’s episode is brought to you by rasa.io.

Rasa.io is the ultimate platform for AI-powered newsletters. If you’re looking to transform your email newsletter into a powerful, engaging tool that truly resonates with your audience, rasa.io is the game-changer you need. Join the 500+ organizations already making their newsletters smart.

Visit rasa.io/maii today.

Listen Now

Watch the Video

Timestamps

00:03:04 — Accenture on track to make $2.4B from generative AI

00:13:21 — The AI industry is headed towards a legal iceberg

00:27:47 — Public perception of AI

00:41:20  — Amazon scrambles for its place in the AI race

00:46:54 —  Microsoft and OpenAI are planning a $100-billion AI supercomputer

00:52:33 — Google considers charging for AI search and Perplexity plans to sell ads

00:56:13 — Demis Hassabis’ rising profile within Google

01:00:14 — HubSpot’s new AI-powered content hub

01:04:27 — OpenAI Voice Engine and voice deepfake concerns

01:08:42 —Google Deepmind paper

01:10:50 — You can now edit your images in ChatGPT.

Summary

Accenture on track to make $2.4B from generative AI

One of the companies making the most money from generative AI right now is a consulting firm. Accenture just reported its second quarter fiscal results and a line item jumped out at AI watchers: In those results, the company reported that it had what it calls “generative AI new bookings of over $600 million in the quarter.”

Annualized at the current rate, that means Accenture is making a whopping $2.4B a year from selling generative AI services.

What is Accenture selling exactly? Well, the fiscal results don’t say. But, based on Accenture’s website, they are advertising AI implementation services and business optimization solutions, where they improve productivity with their prebuilt AI solutions.

We also know, from our discussion in Episode 89 of the podcast, that Accenture has teamed up with AWS and Anthropic to bring generative AI to the enterprise.

The AI industry is headed towards a legal iceberg

A report from The Wall Street Journal kicks off by saying: “Legal scholars, lawmakers and at least one Supreme Court justice agree that companies will be liable for the things their AIs say and do—and that the lawsuits are just beginning.”

The report then details the opinions of several legal experts that, if you use AI enough for content and decision-making, you wind up taking on legal liability. That includes liability for the outputs of generative AI, which can extend to the companies using it, not just the ones building it.

Says the article: “Companies that simply use generative AI—say, by using OpenAI’s tech as part of a service delivered to a customer—are likely to be responsible for the outputs of such systems.”

Another report in The New York Times reveals that “OpenAI, Google, and Meta ignored corporate policies, altered their own rules, and discussed skirting copyright law” in order to train their models. The article reports in-depth on efforts at the companies that appear to have used data in murky or problematic ways to feed their models, including data from YouTube.

This mirrors the conversation that occurred last month at our AI for Writers Summit event between CEO Paul Roetzer and IP law expert Christa Laser. In this discussion, Christa outlined top things to understand about generative AI and copyright as it stands right now in the U.S.

These include:

The fact that you do not own anything created with generative AI right now and cannot defend a copyright claim on it.

The fact that your Work for Hire agreements with employees, agencies, and freelancers do not give you ownership over their work if they used AI to create it—since they don’t actually own it either.

And the fact that you, individually as a user or a company, may be violating laws that you don’t realize by using generative AI, because of how these tools are trained.

Public perception of AI

A new segment about AI on The Daily Show is getting a ton of buzz. It’s about the “false promises of AI” and features Jon Stewart skewering AI leaders for saying AI will make everything better. Stewart concludes that the opposite will occur, with issues like displacing labor and reducing the number of jobs.

The segment features select clips from AI leaders like Mustafa Suleyman saying AI is a fundamentally labor replacing tool and a CEO who is publicly bragging about firing 90% of his support staff thanks to AI.

The overall point is that Stewart highlights the contradiction from public AI leaders between optimism for the future productivity gains created by AI, and what that actually means for jobs.

Stewart ends the segment by asking the question, after AI has taken over: “What is left for the rest of us to do?”

Links Referenced in the Show

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: I still believe that it's directionally true that the people who have the best chance to survive and thrive If job loss comes to your industry or your company, we'll be the people who are most AI literate and are the most high performing people, which has always been true.

[00:00:19] 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.

[00:00:50] Paul Roetzer: Join us as we accelerate AI literacy for all.

[00:00:57] Paul Roetzer: Welcome to episode 91 of the [00:01:00] Artificial Intelligence Show. I'm your host, Paul Roetzer, along with my co host as always, Mike Kaput. Hey, Mike.

[00:01:06] Mike Kaput: Hey, Paul. How you

[00:01:07] Paul Roetzer: doing with you? It is Eclipse Day. It is. We're recording this on Monday, April 8th. 10 15 a. m. It was a awesome weekend in cleveland.

[00:01:17] Paul Roetzer: My daughter and I actually went to the women's final Four championship yesterday, which was really cool to get to take her to that. She's 12 and loves basketball, so it it was just such a

[00:01:27] Paul Roetzer: experience to get to be there. And then Cleveland is like the center of the world today.

[00:01:34] Paul Roetzer: Every airbnb is sold out, every hotel is sold out. People are here to watch the eclipse. We're in like prime range in the U. S. to see

[00:01:43] Paul Roetzer: it. it's just waiting for the clouds to go away. So I hope hope everyone's not disappointed today.

[00:01:48] Paul Roetzer: Oh man. all right. So, I am back from spring break. I'm trying to get my feet back under me after a a week away.

[00:01:56] Paul Roetzer: Had a lot of time to think. Didn't work too much, but had a a [00:02:00] lot of time to think about AI while I was just laying on the beach. So, um,

[00:02:04] Paul Roetzer: This is my first thing I'm doing back from spring break, so hopefully my mind, my mind comes along with me today.

[00:02:11] Paul Roetzer: All right, today's episode is brought to us by Rasa. io. Rasa is the ultimate platform for AI powered newsletters.

[00:02:18] Paul Roetzer: If you're looking to transform your email newsletter into a powerful, engaging tool, That truly resonates with your audience. Rasa.io is thr

[00:02:26] Paul Roetzer: game changer you need. Their smart newsletter platform personalizes content for each and every subscriber, ensuring every message you send is highly relevant and incredibly engaging.

[00:02:37] Paul Roetzer: The tool also allows you to automate away the tedious tasks that go into newsletter production, stay ahead of the curve, and make your emails a

[00:02:44] Paul Roetzer: must read.

[00:02:46] Paul Roetzer: Join the 500 plus organizations already making their newsletters smart. Visit Rasa.io/MAII today. Again, that's Rasa.io/MAII. All right, Mike, let's get rolling

[00:03:02] Paul Roetzer: on the main topics.

[00:03:04] Accenture on track to make $2.4B from Generative AI

[00:03:04] Mike Kaput: Sounds good, Paul. So first up, one of the companies that is making the most money from generative AI right now is actually a consulting firm. Accenture just reported its second quarter fiscal results and a line item in these results jumped out at AI Watchers. In the results, the company reported that it had what it calls, quote, generative AI new bookings, and these totaled over 600 million dollars in this past quarter,

[00:03:36] Mike Kaput: If you annualize that at the current rate, that means Accenture is on track to make a whopping 2. 4 billion dollars This is a year from selling generative AI services. Now, what are they selling exactly? Well, the results don't say, and there was some speculation online about what these actually entail. So diving into Accenture's website, they [00:04:00] are advertising generative AI implementation services and business optimization solutions, where they actually improve productivity using their own products.

[00:04:10] Mike Kaput: Pre built AI solutions. And we also know from our discussion in episode 89 of the podcast that Accenture recently teamed up with AWS and Anthropic to bring generative AI to the enterprise. So Paul, I think my first question for you here is like, clearly the market is responding to a major need here. I mean, deploying generative AI in the enterprise, seems be something that is in quite high demand.

[00:04:37] Mike Kaput: How much? help do companies need here and where do firms like Accenture fit in?

[00:04:42] Paul Roetzer: Yeah, I'm sure you hear it all the time. Like when you're out doing, you know, public speaking, especially for me, when, when you do talks in front of like CEOs and business leaders, you just get a

[00:04:53] Paul Roetzer: a line of people coming up saying, can you guys do for us? Like they don't, they don't know where to turn.

[00:04:59] Paul Roetzer: and [00:05:00] everyone trying to figure it out. So if you present to people, you know, the kind of the the steps

[00:05:04] Paul Roetzer: usually recommend about building an AI council, developing your policies and principles, creating an AI roadmap, the AI

[00:05:10] Paul Roetzer: AI roadmap is always the one we're like, who does that? Like, who can we turn to now?

[00:05:14] Paul Roetzer: What Accenture is doing is they're at the very top of the market, obviously. I mean, these are like fortune 500, fortune 100 companies. They're working with a hundred million dollar plus deals probably. So

[00:05:23] Paul Roetzer: they're doing massive deals big companies. And I think the 2 billion is just absolutely scratching the surface like the size of this market.

[00:05:33] Paul Roetzer: this is something we've looked at because again, we see this demand all the time. Back in 2016, when marketing AI

[00:05:40] Paul Roetzer: Institute was created underneath my

[00:05:42] Paul Roetzer: PR2020, where Mike and I I work together, um, that was actually actually my vision PR2020. My agency was, we were just going to become like an AI transformation agency.

[00:05:51] Paul Roetzer: We were going going to help companies do this, but the demand just wasn't there. It was too, we were way too early. I mean, we had some pretty large clients at the [00:06:00] time and, you know, They didn't care. Like they weren't even thinking about this stuff.

[00:06:04] Paul Roetzer: So, you know, I think that when I look at this space, the consulting advisory firm agency space, this is the future.

[00:06:12] Paul Roetzer: There's no doubt. Like we did our AI for agencies summit. Was that November, Mike, I think last year. And so I went back and looked at a slide I

[00:06:19] Paul Roetzer: I had, and I of laid out like the the opportunity for agencies and consulting and some of the the areas that I

[00:06:26] Paul Roetzer: were. large language models, strategy and and implementation, AI education and training, AI technology integration, change management, and then a couple more like longer term was

[00:06:37] Paul Roetzer: AI agent support, and then Augmented creativity. So those are some of the areas we looked at. So to try and figure out like where is, The Accenture money coming from, I

[00:06:45] Paul Roetzer: I actually and looked at the earnings call. So it was March 24th, I I think it

[00:06:50] Paul Roetzer: late March, 2024. And so, they talked more detailed about where this money is coming from. So the, let's say Julie [00:07:00] Sweet, the chair and CEO, in the earnings call said, we are also had 600 million in new Gen AI bookings, taking us to 1.

[00:07:08] Paul Roetzer: 1 billion in Gen aI sales in first half of their fiscal year. Expanding our our early lead in Gen AI. which is core to

[00:07:16] Paul Roetzer: to our client's reinvention. We now have over 53, 000 skilled data and AI practitioners

[00:07:22] Paul Roetzer: against our goal of doubling our data and AI workforce from 40 to 80, 000 by the end of fiscal year 2026. 2026.

[00:07:31] Paul Roetzer: So for me, like, as we I'll go through a couple other findings from the call, but basically what, what the way to think about as a listener is it's, we've been looking for hard data to show demand, adoption, and

[00:07:45] Paul Roetzer: and opportunity. And so I this is a really good example of kind of a leading edge firm.

[00:07:52] Paul Roetzer: That is now showing the demand is skyrocketing for this kind of stuff. Julie went went on to say all [00:08:00] clients have to get there, meaning transformation. They need to get the technology transformation. They need to get to reinvention. And that's why you're seeing, even at the constraints, you're seeing that early interest in Gen AI.

[00:08:11] Paul Roetzer: With 1 billion in sales in the first six months of the year, that is the fastest we have built sales in an emerging technology. And what it tells you is that clients understand the importance of

[00:08:23] Paul Roetzer: AI, that they're going to have to reinvent every part of the enterprise.

[00:08:27] Paul Roetzer: So that was quote directly from their CEO.

[00:08:31] Paul Roetzer: so, then then they went on throughout the call to talk about of the examples, say Merck is a client and they said they're doing. launching Generative AI training program for their their employees to create

[00:08:41] Paul Roetzer: class digital leaders. Riyadh Air, they're working with to do more personalized travel experiences for customers. Mondelēz international, has well known brands like Oreo and cadbury.

[00:08:53] Paul Roetzer: They're working to reinvent how they satisfy customers. Um, they talked about an an

[00:08:58] Paul Roetzer: where they're doing like a [00:09:00] complex, new need integration across large language models. So they're basically figuring out how to

[00:09:04] Paul Roetzer: to do the symphony of large language models within companies, and then how to to optimize based on the model that's going to be used.

[00:09:12] Paul Roetzer: So they're building kind of these capabilities internally. A lot of talk in the earnings call about the structure and simplification and and modernization of data systems. In one example they talked about a that had over 50 different enterprise data sources and how they're trying to integrate into a single data source so they can build better generative AI capabilities.

[00:09:32] Paul Roetzer: So, I mean, when you look across what they're doing and where the demand is, it's still. Early in a lot of it is very foundational.

[00:09:39] Paul Roetzer: they're trying to get the data in place. They're trying to educate and train teams. there was one quote they had about their education and training. Oh, here it was. So they acquired Udacity in March, which was a tech training platform.

[00:09:52] Paul Roetzer: and in their newsroom announcement, we'll put in the findings, it said, according to Accenture research, business leaders say their number [00:10:00] one challenge is their inability to upscale their workforces.

[00:10:03] Paul Roetzer: With 51 percent of organizations starting to see negative impacts from worsening iT skills and shortages. This is the one that caught my attention. In addition, 94 percent of workers

[00:10:15] Paul Roetzer: they want to learn new skills to work with generative AI, but only 5 percent of organizations provide gen AI training at scale. That is absolutely aligned with everything we've seen. We've been doing our state of

[00:10:27] Paul Roetzer: of marketing I report for, I think we're in fourth year now to the third or fourth year. And lack of education and training has been the number one obstacle every year. And our data says, I think, I forget what last year was like somewhere around

[00:10:40] Paul Roetzer: like percent or something had any form of internal education and training for AI.

[00:10:44] Paul Roetzer: So, yeah, I I mean, I think again, like

[00:10:46] Paul Roetzer: What we're seeing here is early indications and actual hard data to show that companies are trying to figure this out. I mean, is this aligned with what you're hearing when you're

[00:10:57] Paul Roetzer: out on the speaking circuit, Mike?

[00:10:59] Mike Kaput: yeah, [00:11:00] 100%. And I think it's important to reiterate the point for anyone who's maybe unfamiliar with either the consulting and agency space or just the overall enterprise needs related to AI. I mean, I don't know. What do you think? Like, these companies aren't able to go higher. People, many people internally that have this type of education, you're almost by default forced, if you're not one of the major tech companies who are hoovering up a lot of the technical talent, you're forced to have to work with an outside party.

[00:11:30] Mike Kaput: Because to my knowledge, that's really the biggest avenue for

[00:11:34] Mike Kaput: firms that have people educated in the right way to be able to either deliver implementations or foundational education.

[00:11:41] Paul Roetzer: Yeah, I mean, it's like a two tier approach. You gotta upskill, you know, your own people, provide the literacy

[00:11:47] Paul Roetzer: within your own team and develop the next generation of workers and leaders internally. But then you, you have to lean on people outside. Like, I wouldn't be surprised if is a five to 10 billion project. dollar project. piece of [00:12:00] business for in the next two years.

[00:12:02] Paul Roetzer: And I, know, I I would think there's lots of opportunity. Like they're not even working with the smaller businesses. And this is what we keep stressing to people. It's like, if you run an agency or if you're independent consulting or whatever it

[00:12:12] Paul Roetzer: it is, There is such massive opportunity to help companies figure this out and this is again like we're in the first inning of this aI transformation phase. And as things start evolving,

[00:12:24] Paul Roetzer: going to be all kinds of opportunities open whether you want to go vertical specific and work just with like, law firms or engineering firms or retail or whatever your niche is or

[00:12:34] Paul Roetzer: you want to be more, more broad. So again, if I I was still running an an agency, this what I would have been doing. I mean, I sold that agency in 2021. So it's not a world I've been playing in much,

[00:12:44] Paul Roetzer: but you know, certainly through the Institute, we've done some AI consulting. We try and limit it it's, you know, me and and Mike, capacity is pretty low for us to do the consulting work.

[00:12:54] Paul Roetzer: So we've been pretty limited in it. You know, we may evolve that. There may be some other things we do in the [00:13:00] future, but at the moment, like, it's hard to look at this and not see just. billions of dollars of of opportunity ahead

[00:13:06] Mike Kaput: So, in our second big topic today, uh, if your company is using AI to do anything like Producing content, make decisions, or

[00:13:17] Paul Roetzer: Which if you're listening to this, this, this show, you likely are, you likely

[00:13:21] The AI industry is headed towards a legal iceberg

[00:13:21] Mike Kaput: are actually using, yeah, exactly. A couple of new reports. are things you should be paying attention to, because we're seeing a couple big indications that legal experts are sounding the alarm that the AI industry is, in their words, steaming towards a legal iceberg.

[00:13:42] Mike Kaput: So first, a report in the Wall Street Journal kicked off by saying, quote, Legal scholars, lawmakers. And at least one Supreme Court justice agree that companies will be liable for the things their AIs say and do and that the lawsuits are just beginning. It then details the opinions of several [00:14:00] legal experts that if you are

[00:14:01] Mike Kaput: are using AI for the things I mentioned, you wind up taking on legal liability.

[00:14:07] Mike Kaput: And that includes liability for the outputs of generative AI, which can extend to all the companies using these tools. Not just the ones building them. And the article goes on to say, quote, Companies that simply use generative AI, say by using OpenAI's tech as part of a service delivered to a customer, are likely to be responsible for the outputs of such systems.

[00:14:34] Mike Kaput: Another report in the New York Times that just came out also reveals that there are Significant problems with how, perhaps, these models and tools are being trained and built. They say, quote, OpenAI, Google, and Meta ignored corporate policies, altered their own rules, and discussed skirting copyright law in order to train their models.

[00:14:56] Mike Kaput: The article then goes on to report in depth on [00:15:00] the efforts at those companies that appear to have used data murky or problematic ways to feed their models, including, somewhat controversially and relevant to several AI discussions happening right now, data from YouTube.

[00:15:15] Mike Kaput: So Paul, this starts to mirror basically exactly what you discussed during our AI for Writers Summit with IP law expert Krista Laser last month.

[00:15:28] Mike Kaput: And in that discussion, which I revisited as we talked about running this topic on the podcast, Krista outlined some top things to understand. about generative AI and copyright as it stands right now in the US. So this includes the fact that if you did not know you do not own anything created with generative AI right now and you cannot defend a copyright claim on it.

[00:15:54] Mike Kaput: There's an element to this that your work for hire agreements with employees, agencies, [00:16:00] freelancers do not give you ownership over their work if they used AI to create it, since they don't, in fact, own it either. And then there's this overall problem that individually, as a user or a company, you may be violating laws that you don't realize by using

[00:16:17] Mike Kaput: AI because of how the tools were trained. So, Paul, just kind of like connecting all these dots here. There's a lot to unpack here, but it really just does sound like there is massive uncertainty ahead when it comes to the legal implications of using these tools and of how the tools were trained.

[00:16:36] Mike Kaput: Is that kind of your sense of where this stands today?

[00:16:39] Paul Roetzer: Yeah, and and I I mean, the headlines are coming more frequently now. I think the mainstream media is certainly, like, picking up and running with this story now. There's nothing, like, really new. Per se that has come out, other than like the New York Times article for example, had all kinds of interesting inside source information.

[00:16:59] Paul Roetzer: had [00:17:00] and, uh, like recordings of internal meetings executives. So they were disclosing more about like how this is and how, you know, Blatant the disregard for law was within the big tech companies to do this and how pretty much once OpenAI did it, everybody's like, well, they did it.

[00:17:17] Paul Roetzer: Like, let's just go do it. so we're starting to learn more and more about the internal workings of how decisions were made to skirt These laws in the training of the

[00:17:29] Paul Roetzer: the models and what that means us as end users. So a couple things come to mind. One, you kind of of alluded to mike, but laws vary by country.

[00:17:39] Paul Roetzer: So, you know, oftentimes Mike and I will talk specifically about US law because it's the thing we're the most familiar with. But I do know that, like for our our AI

[00:17:48] Paul Roetzer: for Writers Summit, we had, registrants 93 different countries. So in the with Krista, we wanted to make sure that we talked about other country law around this stuff.

[00:17:57] Paul Roetzer: the U. S. copyright law, [00:18:00] to my knowledge, so they came out with their guidance last March, so over a year ago, on generative AI, where they basically said what you outlined. You don't own it, that if, if generative AI creates it, nobody owns it. It's the prompt is not human authorship, not enough for human authorship.

[00:18:16] Paul Roetzer: And therefore a generated work doesn't, isn't owned by anyone. To my knowledge, despite their listening sessions, they've been going through. I don't believe they have updated any additional guidance or given any additional indications about how. copyright law may evolve in the

[00:18:32] Paul Roetzer: the United States. Now, what does this mean for you, as business of this technology? Two, two things kind of come to mind. One is, you don't own it if you use it to create it. So if you need a copyright on something, you really can't use generative AI to do it. And two is the bigger question around, like, are you at risk? Like, if, so if for some reason, um, Courts decide that [00:19:00] OpenAI and Meta and Google and Anthropic and everybody else trained their models, stole copyrighted data, and therefore the people who used those models to create things are now liable.

[00:19:14] Paul Roetzer: Microsoft and Google and maybe others announced that last year that they would cover the legal costs of that. So, we'll put put link in the show notes, but so I have the September 7th one from. Microsoft, Brad Smith, the

[00:19:30] Paul Roetzer: VP and president, uh, where they said that they were announcing a co pilot copyright commitment where they would assume the legal risks and costs if you were sued for copyright infringement due to to use of co pilot.

[00:19:46] Paul Roetzer: And then they actually updated in January of 2024 to extend it to basically anything done under Azure. So that's how these big. Tech companies are trying to like make it safe for you to use their tools where you don't [00:20:00] have to worry necessarily about being sued

[00:20:02] Paul Roetzer: for, the use the models. I don't know know if going to make your legal department feel much better about it, but that's of where it's at. So, Yeah, you you know, the thing you in the New York Times article, so there was a couple of things jumped out to me on this one. So one was, we've heard about Whisper, like OpenAI's incredible technology.

[00:20:27] Paul Roetzer: I, some of you have probably used Whisper, like if you go in and you, you know, talk to it, You know, know,

[00:20:33] Paul Roetzer: using that technology, but apparently it was, that technology was created to transcribe YouTube videos to train their models. And they knew that that was, certainly ethically questionable and likely legally questionable to do it.

[00:20:51] Paul Roetzer: And that was a prelude to what we talked about about a couple of episodes ago, where they appear to have used YouTube videos train Sora as well. But then, [00:21:00] um Cade Metz and the other New York Times writers said that

[00:21:03] Paul Roetzer: Google employees aware OpenAI had harvested harvested YouTube videos for data, but but they didn't them them because they had done the same themselves.

[00:21:11] Paul Roetzer: And even though Google owns YouTube, they don't own the content. don't have the rights to take that content and train on it.

[00:21:19] Paul Roetzer: and so basically, everyone is violating all of these rights and just sort of going, you know, full steam ahead, figuring out if got to pay some fines or or you know, things catch up to us, we'll figure it out.

[00:21:33] Paul Roetzer: And then they even had a, know, recording of a meta executive where it said OpenAI seemed to have used copyrighted material without permission. And it would take meta meta acknowledge, it would take too long to negotiate licenses with publishers, artists, musicians, the news industry, they said, according to recording.

[00:21:48] Paul Roetzer: So basically they knew what they were doing wasn't. Probably legal, but everyone was doing it. And like, let's, let's go. So the debate becomes like, [00:22:00] people can get, I always laugh. Like I'll put stuff on

[00:22:03] Paul Roetzer: on LinkedIn about this and you get you get people who aren't lawyers, state things as though like they, No law is likethatit was fair use and they should have taken this.

[00:22:15] Paul Roetzer: And you and I I might take the of like, I have no idea. I'm not an an I'm not not a judge. Like, I don't know what end game ruling is, but the debate about whether or not the law is is obsolete a different story. Like you can state your opinion that

[00:22:29] Paul Roetzer: it should have been fair use. Okay, fine. That that's an opinion, but the, we have to wait courts to tell us whether it actually was or not. And so that's, I don't know, that's kind of of like my on on this is like, we. Just know. We're all kind of of observing.

[00:22:44] Paul Roetzer: I think in the end they probably pay some fines and admit that they knowingly

[00:22:49] Paul Roetzer: did this or don't admit it and just pay the fines.

[00:22:52] Paul Roetzer: and we probably move

[00:22:53] Paul Roetzer: on and it doesn't

[00:22:54] Paul Roetzer: down adoption and it doesn't, you know, slow down innovation, but it's certainly worth, you know,

[00:22:59] Paul Roetzer: [00:23:00] Paying attention to, and I wouldn't be shocked, as I said on LinkedIn, if this doesn't lead to a a Senate like, this is the kind of thing that they love to do in Senate.

[00:23:08] Paul Roetzer: Like, bring the company CEOs before and say, how did you train it? you do, you know, it's like, make a show of it. So, I don't know. know, I think it'll continue to gain mainstream media headlines. probably gain the the interest of u. S. S. lawmakers some point here. I don't know know that any of that really affects what any of us are gonna do day to day, though.

[00:23:26] Mike Kaput: right.

[00:23:28] Mike Kaput: One area of this that is

[00:23:32] Mike Kaput: Somewhat relevant to

[00:23:33] Mike Kaput: of the brands listening in is can you talk to me a little bit about how this might affect. Your work with an agency or a freelancer, because we actually have a related story that came out as well from Ad Age that documented how certain brands are cracking down on how their agencies are using AI.

[00:23:55] Mike Kaput: So in some cases they are totally restricting the use of [00:24:00] AI without prior authorization. In others, they're, rightly so, worried about all the ways AI can go wrong, or how it should be used. Can you walk us through, what do I have to start thinking about now that agencies don't own the output of generative AI, so therefore cannot transfer that ownership

[00:24:18] Mike Kaput: to me via a work for hire agreement?

[00:24:21] Paul Roetzer: Yeah, I mean, we say it all the time, like, first thing I do is get legal counsel involved. Like, this isn't something you're going to solve on your own. You've got to have the lawyers in the room.

[00:24:32] Paul Roetzer: you've got you've got to get the Genitive Aid policies in for your own team, but also for your providers, for your service providers, like those should extend out to the people you are contracting with to create content, whether they're freelancers or outside video producers, you know, freelance writers or video production, people,

[00:24:50] Paul Roetzer: graphic designers, like whatever it is, anyone who's creating content on your behalf, you need to have clarity of it is you're allowing them to do and then [00:25:00] understanding what it is that they're doing. how, how are

[00:25:03] Paul Roetzer: are they. handling this internally? What are are their trained to do? What are their, how have have their processes evolved? So, yeah, I mean, I think you need to do a legal review of all of your contracts with outside service providers and

[00:25:15] Paul Roetzer: make sure that you're expecting copyright, material, the material that can own a copyright to when it's created from them or, you know, own trademarks to that. That is covered in your agreement with them.

[00:25:27] Paul Roetzer: And, you know, know, our experience has been a lot of agencies and freelancers are like, just trying to figure this stuff out. They don't have generative AI policies for their people. So they may not like their people may be using AI and. The CEOs of those

[00:25:42] Paul Roetzer: agencies might not even know they're using the AI, and they're, they're passing off work the CEO of the agency may actually think is legitimate work. And here they find out that their own people are are using

[00:25:52] Paul Roetzer: these generative AI tools without their knowledge. Like, that's a very real possibility right now.

[00:25:58] Mike Kaput: as we wrap up [00:26:00] here, I'm curious, do you have any advice on elements that are important to have in a generative AI policy? We talk all the time about having this policy, or a set of policies around this, is such a critical first step for Brands

[00:26:14] Paul Roetzer: yeah, I I mean, like, we've gone through this a couple of times where, you know, for me, it's, it can be as simple as. what technologies are they allowed to use? How are they allowed to use

[00:26:29] Paul Roetzer: Like, what is the end product they're allowed to create with it? And then, do they have to disclose the use of technologies?

[00:26:37] Paul Roetzer: So, if you're working in an agency, and your client is assuming copyrighted material is being passed over to them, then That would that, one, you aren't even allowed to use it, like you can only use it for ideation and outline development, but in the actual writing, you're going to have to actually write the thing as the human.

[00:26:57] Paul Roetzer: So, you know, I think you just really need to [00:27:00] go through the process of what are the outputs trying to create? What are are the ways we're using technology? What are the agreements we have? With people, um,

[00:27:08] Paul Roetzer: and then, you know, what does legal us on? So again, it it often comes down simply to me, like, which are you allowed to use?

[00:27:16] Paul Roetzer: How are are you allowed to use them? And what do you have to disclose in the use of them?

[00:27:21] Paul Roetzer: but again, I would always recommend go find a template. Like there's, you can go search for generative AI policy templates. I get asked all the time.

[00:27:28] Paul Roetzer: I originally mentioned this, that we were building, Uh, GPT to to help develop their policies. No, I haven't finished that yet because everybody always asks me if I've created it. hopefully I will, soon. So yeah, maybe we'll just create that.

[00:27:44] Paul Roetzer: It'll help people do it, but there's great templates available online now.

[00:27:47] Public perception of AI

[00:27:48] Mike Kaput: So in third Big Top. topic today, we're to talk a little bit about the public perception of artificial

[00:27:55] Mike Kaput: because a new segment about AI on The Daily Show is [00:28:00] getting a ton of buzz and it's about what they call the false promises of AI and it features Jon Stewart skewering AI leaders for saying AI will magically kind of make everything better when in fact in his opinion There are clear signs that they are actually saying it will displace labor and reduce the number of jobs. So

[00:28:23] Mike Kaput: this segment and true kind of daily show fashion features, you know, cherry pick clips from AI leaders like Mustafa Suleman saying AI is a fundamentally labor replacing tool. and

[00:28:35] Mike Kaput: they interview A CEO who's been publicly bragging for quite some time about firing 90% of his support staff thanks to ai.

[00:28:44] Mike Kaput: And they also. mix in some clips of politicians simply saying over the years, everyone will just have to retrain when technology takes their job. So the overall point here is that Stewart kind of is highlighting the [00:29:00] contradiction from public AI leaders about both optimism from future productivity gains created by AI, and then how they kind of talk around what that actually means.

[00:29:12] Mike Kaput: for jobs. Now, Stuart ends this segment by asking the question, you know, after AI has taken over, what's left for the rest of us to do? So, Paul, this is just one recent example of some public backlash and apprehension around the impact of AI, and I know you had a lot of thoughts on this one. Could you walk us through how you view this?

[00:29:34] Paul Roetzer: Yeah, about an an 11 minute segment. I mean, worth watching. It is funny. it very much gives the

[00:29:41] Paul Roetzer: from the techno optimist slash EAC movement we've talked about. These people who like accelerate at all costs, no matter what, and we'll figure it all out. And we've been transformation before and and it'll work out, and there will be jobs for everybody, be better jobs and and again like Mike and and I

[00:29:59] Paul Roetzer: very [00:30:00] much like are, I would say like, I'm in the, let's keep going. Like, let's keep innovating. Let's keep accelerating this. I think a lot of good is going to come from it, but I'm also of the camp. And and I think Mike probably similar, we should be thinking about the impact this is going to have on people.

[00:30:13] Paul Roetzer: Like, I am not, ignorant. to the disruption this cause. Now, again, like, I always laugh when people these, like, really hardcore stances as though they can see into the future better than the economists in the world who don't seem to know what's going to happen. And so you have all these

[00:30:34] Paul Roetzer: who are are like, no, no, no, it's going to perfect. It's going to work out. Let's just keep going.andthen you have have other people who can,

[00:30:40] Paul Roetzer: You know, convince just going going to be devastation and it's going horrible. And I'm kind of in the of in the middle, like, I actually don't know what's going to happen.

[00:30:47] Paul Roetzer: we shared the AI timeline, I think it was episode 87, where I where I was like trying to give my projection of what I thought might occur, for, for jobs for the economy.

[00:30:59] Paul Roetzer: And [00:31:00] recap that is like, I do think we're going going to job loss. Like, I don't know any other way around this, that there are going to be certain industries and professions, some. It'll happen sooner than others. So like the technology industry, yeah, SaSS is to to

[00:31:13] Paul Roetzer: seeing jobs. Like it's going to feel like like it's happening right now. And it's happening fast in, in the SAS world. Um, customer service, like likely in the next 12 to 18 months, it's going to feel really likethatprofession is going to rapidly change.

[00:31:29] Paul Roetzer: I could see things like BDRs, like, you know, sales. I could see things like junior associates at law firms. There are definitely roles that I think in the next 12 to 18 months, it's going to changing really

[00:31:44] Paul Roetzer: fast for, but when you zoom out at a macro level, I don't we see massive disruption to jobs in the, in the next 12 to 18 months, but that that's just trying to kind of

[00:31:55] Paul Roetzer: of read the market and where we'd see the technology going and what we're hearing [00:32:00] from, you know, and entrepreneurs, like trying to to make our best educated guess at, you at what happens.

[00:32:06] Paul Roetzer: But the thing that I find interesting about this Jon Stewart episode is I think this is an indication of what the, what we can expect as more and more industries and professions start to be affected. Um,

[00:32:20] Paul Roetzer: this sort of, I wouldn't say this is necessarily like societal backlash. This is more commentary, and you know, on a specific segment of of people who have this very techno optimist perspective.

[00:32:33] Paul Roetzer: But I I do think that as more people start to get impacted, you could start to get some backlash, from, from the media and from society. And so I think that all of us who are, know, kind of of more

[00:32:49] Paul Roetzer: living AI day to day and probably have broader context around The depth of of this story, like for some people, again, they'll see a 60 minutes episode and that is all they know about [00:33:00] AI.

[00:33:00] Paul Roetzer: And like, they're convinced whatever they saw in 60 minutes is the the thing.

[00:33:03] Paul Roetzer: Or they watch Jon Stewart and that's the first real interaction they have with AI. And they like, think that that is all the truth. Well, all of us who like, You and I doing the the show, people who to the to the show, we're thinking and hearing way more about this.

[00:33:15] Paul Roetzer: And we have all context to build our our own points of view and on jobs and the the impact on industries and things like that. And so I think you can step back be more realistic about what, what's actually going on. Um,

[00:33:28] Paul Roetzer: but yeah, soIjust think it's interesting from a societal perspective of, how we're going to start dealing with these layers of complexity as more people take an interest in the topic, more people in the media, more people in government.

[00:33:46] Paul Roetzer: it's just gonna, again, it's gonna become a, a, there's gonna be be more depth the story to be considering and to be evaluating against like how different people look at it. I mean, you watched it too, Mike. [00:34:00] Did you have any other thoughts on it or any opinions on it?

[00:34:03] Mike Kaput: No, I really enjoyed the approach and the clip. I tend to agree with you. I think where we're really going to see this perspective continue to rear its head is as we see headlines of possible layoffs or disruptions due to AI. I have no idea what that will look like, but I could see there being Quite a strong narrative eventually as we try to use AI as an explanation for all sorts of possible labor displacement that happens in the next several years.

[00:34:37] Paul Roetzer: Yeah, yeah, and I'll, I'll, I'll trust one other thing. so over the the weekend, it came up this, uh, year, year and a half. I don't remember when I first used it, but I included in a Intro to AI for Marketers class I was was teaching, I think was the first time I said it, that AI

[00:34:54] Paul Roetzer: AI won't replace marketers, marketers who use AI will replace marketers who don't. Actually, I think it was in the piloting AI class. [00:35:00] course series the end of 2022. So I may have have actually said this like

[00:35:03] Paul Roetzer: a year and and a half, two years ago. and I see people like, who don't like that who, who, who like, think it's just a load of, of crap. And,Inever really

[00:35:16] Paul Roetzer: understood the reaction people have to but for context, like if you've heard this quote, um. and it does, you know, cause you to have an unfriendly reaction, I want to to explain where it came so in 2017, so back in 2015 2015

[00:35:32] Paul Roetzer: 16, there was this belief that radiologists were done, that AI was going to become superhuman at reading CAT scans, MRI scans, and that we just

[00:35:41] Paul Roetzer: wouldn't need radiologists do what they do, the machines would just take over. And so that was kind of like a growing and then in 2017, Curtis langlotz, I think that's say his name, Professor of Radiology and Biomedical

[00:35:56] Paul Roetzer: Informatics at Stanford University School of Medicine, [00:36:00] said at an annual meeting of the Radiological Society of North America, radiologists who use AI aI will replace radiologists who don't. So his whole premise the AI aI wasn't going to take over. But the tools were going to to be so

[00:36:14] Paul Roetzer: and the radiologists who used them were going to be so much more effective, so much more more productive than the the radiologists who didn't,

[00:36:22] Paul Roetzer: that over time you either used AI or you didn't. But if you weren't using it, you weren't going to have a job in it was going to become that

[00:36:29] Paul Roetzer: to what people did. And so I looked down and I out and I thought, wow, that's going to be true for every knowledge work profession. You know, writers, designers, engineers, product developers, whatever it was, salespeople, That the people who used AI would become so much

[00:36:46] Paul Roetzer: more efficient at their jobs, that when the came for jobs to be impacted by AI, You are going to look around at at your staff, you can staff of 10 or a staff of 10, 000, and you are are going to say, [00:37:00] who are the most productive people on

[00:37:02] Paul Roetzer: this who are the are the most efficient on this team? Who create the most value in the the company? And my belief at at the time was, well, it's obviously going to be the people who are better at at AI. Like,

[00:37:12] Paul Roetzer: if you become adept at using these tools and you get through kind of the fear and anxiety of AI and everything, and you

[00:37:18] Paul Roetzer: you say like, I'm just going to figure out how to do them. I'm going to become 10 percent more efficient or 30 percent more efficient or 70 percent more efficient, whatever it it is. That when the time came, were like, wow, we don't need need as humans doing accounting or law or marketing or graphic design or whatever it it is. We don't

[00:37:33] Paul Roetzer: don't need as many of these people around. These are the 10 best, like they're outperforming everybody by 30%. That's who we're keeping. That

[00:37:41] Paul Roetzer: was the premise of quote,thatAI wouldn't replace marketers or writers or designers. The people using AI would replace the the ones who don't. I don't know how that's not still true. Like, I see all these

[00:37:54] Paul Roetzer: like complaining about this quote and saying it's ridiculous.

[00:37:56] Paul Roetzer: Like, how is it ridiculous? Like, I I don't, I don't I don't [00:38:00] understand the logic of how it's not, not true. But the, what I I have evolved do is like, I don't use it anymore in my talks and in in my

[00:38:07] Paul Roetzer: And the reason is, I don't know that it holds true over time. Like, I do think that there comes a point with AI agents and like this evolution of intelligence explosion, like we

[00:38:18] Paul Roetzer: about in 87, episode 87. Where, maybe it actually does replace people, too. Like, maybe we just don't need that many

[00:38:26] Paul Roetzer: and even the the ones that are using it, there isn't gonna be roles for them. I don't know. Like, that is certainly a possible future. But at the moment,

[00:38:34] Paul Roetzer: I still kind of believe that it's directionally true that the people who have the best chance to survive and thrive If job loss comes to your industry or your company, we'll be the people who are most AI literate and are the most high performing people, which has always been true.

[00:38:55] Paul Roetzer: So if you, if you assume AI improves your performance and [00:39:00] you can per outperform your peers, then it is just a

[00:39:03] Paul Roetzer: to make you more valuable in the company, which makes you safer when job loss happens. Like, and so again, if someone has a perspective that. It makes that false, I'd love to hear it, but just, I've thought logically about this for like two years

[00:39:17] Paul Roetzer: and I've yet to find the fault in it other than the who just assume all jobs are going away and they're making like, A leap five years out and saying, well, the AI is just going to take everybody's job.

[00:39:27] Paul Roetzer: What are you talking about? It's like, okay, maybe, but I don't know that to be true yet. I don't know, like, am I crazy? Like, do you have you have another perspective on this? You've heard me say this like a million times,

[00:39:37] Mike Kaput: You know, I've always struggled with

[00:39:39] Mike Kaput: some of the backlash against that type of quote too, because it seems pretty obvious if you. Change the analogy to other technology. If I said to you, hey, computers probably aren't going to replace people, but people using computers versus those who don't will probably be more economically valuable.

[00:39:59] Mike Kaput: That seems [00:40:00] obvious to me.

[00:40:01] Paul Roetzer: right? Like if it's if you, yeah, if you stay on the the tool talk, It's like, okay, if you're a sales leader, but you refuse to use a CRM and you still want to in your Excel, which Mike, you andImean, going back to when we were at DHC, we were still in 2020.

[00:40:15] Paul Roetzer: dealing with clients who had sales teams who refused to enter data into Salesforce, and they still managed it through CRM.

[00:40:21] Paul Roetzer: or a a notepad. notepad. And that's what I feel like this is. It's like, okay, there's no way those people keep their job. It's kind of the similar analogy of like, you have a new tool that makes

[00:40:33] Paul Roetzer: so much better at at your job and you refuse to use it. I'm sorry, I'm going to you with someone who's using the tools. That's basically all the quote is saying.

[00:40:43] Paul Roetzer: So, yeah, I'm sorry, like, it's not, and it's not even like like a hot issue. like it bothers me that much, like I kind of laugh at it. but again,Ithink pretty sound logic. Like AI tools help you become a higher performer. And if a company is going to [00:41:00] reduce job force because technology makes their workforce more efficient, it's pretty logical that that the who use the the tools,

[00:41:07] Paul Roetzer: To be better performers, keep their jobs over the the people who don't. Yep. So

[00:41:13] Mike Kaput: seems pretty obvious to me. All right, so let's dive into some rapid fire topics

[00:41:20] Mike Kaput: here.

[00:41:20] Amazon scrambles for its place in the AI race

[00:41:20] Mike Kaput: So up, Amazon is making some big moves in AI, and of them are maybe a little uncomfortable, because the company just announced a, 2. 75 billion investment into Anthropic, which we reported on in the past. It's the largest investment that Amazon has made into a company and obviously a clear way to compete with Microsoft slash OpenAI.

[00:41:45] Mike Kaput: But Amazon's relationship with Anthropic is far less cozy than it appears, says reporter Alex Heath in The Verge. Heath learned that the company's AGI team, led by SVP Rohit Prasad at [00:42:00] Amazon, is trying to compete directly with Anthropic by building their own flagship model. Amazon's flagship model they're building is codenamed Olympus, and the goal, it sounds like, is for

[00:42:12] Mike Kaput: Olympus

[00:42:13] Mike Kaput: to outperform the newest Claude models from Anthropic by the middle of this year.

[00:42:19] Mike Kaput: When Olympus

[00:42:20] Mike Kaput: released, The Verge says, quote, it'll be plugged into nearly every part of Amazon and made available to other businesses. So Paul, in the reporting from The Verge, they kind of dived into how some

[00:42:34] Mike Kaput: unnamed Amazon employees mentioned that there is this interesting dynamic where Amazon was competing with Claude, even though they are relying on Claude for many of their foundational model services and capabilities, and that everyone's kind of waiting, holding their breath until Olympus comes out.

[00:42:55] Mike Kaput: Where does Amazon stand right now in the [00:43:00] AI race?

[00:43:01] Paul Roetzer: A couple of thoughts come to mind here. First, I think it's important to have the context that AWS is the the largest cloud I don't remember exact market share, but I would say it's like low 40 of the market. So if you think about. AWS, Google Cloud, and

[00:43:16] Paul Roetzer: and Microsoft Azure primary cloud providers, AWS dominates that market.

[00:43:21] Paul Roetzer: A couple episodes ago, we talked about early researches showing that one of the sticky factors selecting a foundation model or the generative AI you work with is where is your data already at and who do you trust with that data?

[00:43:38] Paul Roetzer: So it it obviously makes a ton of that AWS and, and, And

[00:43:42] Paul Roetzer: Amazon are going continue to try and benefit from the fact they house data from many of the largest companies in the world. Their initial strategy, going back to 2023, 2023,

[00:43:55] Paul Roetzer: appeared to be this, the same as as they do for everything the everything store for aI. You [00:44:00] want an an AI? Come to Amazon. Great. Come on we got Anthropic. We got Cohere. We got our own, you know, Titan model, whatever you need. Like we, we got it

[00:44:08] Paul Roetzer: it you. We just want to process the data. We want to run the compute and you know, know, increasing your AWS bill month. so that made made a lot of sense.

[00:44:18] Paul Roetzer: But we also know that Amazon has much broader interest aI. So we talked on a recent episode about Rufus, which is their generative AI that's going to to be infused amazon. com. So you're going to have a shopping assistant, that, you know, know, you can have a conversation basically.

[00:44:35] Paul Roetzer: I'm going on a trip to Italy. what do do I need? it can talk to you and it can actually recommend products based on like where you're going, like rather than you searching by filters and keywords. So they have that. can't forget the fact that they have hundreds of thousands of robots. So actually in a news release from October 2023 on their site in their newsroom, we now have over 750, 000 [00:45:00] robots working collaboratively with our our employees, on highly repetitive tasks and freeing employees up to better deliver.

[00:45:06] Paul Roetzer: for our customers. They're testing drones that deliver packages. Like they, they have a lot of use for it's dramatically impactful to their business. So it makes a ton of sense

[00:45:19] Paul Roetzer: that they wouldn't be sitting back And waiting for everyone else's models. Plus they have a whole bunch of proprietary data. Oh, and they have have Alexa.

[00:45:28] Paul Roetzer: So they have all of these things would make up the recipe for a very powerful foundation model, data and distribution being the two ones we always talk about.

[00:45:38] Paul Roetzer: they have both of those things that maybe only. I mean, at Google, certainly, like I'm trying to think who else could even compete meta, maybe different types data, but, there's only probably three to five companies in in the world that have this level of data distribution.

[00:45:55] Paul Roetzer: So it would make a lot of sense to build a a competing foundation model. [00:46:00] the only other thing, at it's like, man, is this a space? Like you can understand how Accenture's making two and a half billion dollars consulting on this, because if you're a big enterprise and you're trying to figure out what to do, and you're like,

[00:46:13] Paul Roetzer: Oh, should I go cohere? Oh

[00:46:14] Paul Roetzer: Oh, wait, I can go with Cohere, but right within AWS. Oh, but AWS is building Olympus. Maybe I should like, wait for,

[00:46:20] Paul Roetzer: Oh my gosh, this is why I say like, we're in inning one of figuring out what this means to enterprises, how to adapt it once you do adapt it, the change management needed behind it. So, yeah, it's fascinating.

[00:46:33] Paul Roetzer: But this is the first I'd heard of Olympus. Like I don't, had you heard anything about it prior to this? Yeah, I hadn't either. so yeah, just kind of another intriguing thing. And like we, we've been talking a lot about Apple. building the the next version of Siri. I got to to imagine

[00:46:49] Paul Roetzer: that Amazon is working feverishly on the next generation of Alexa as well

[00:46:54] Microsoft and OpenAI are planning a $100-billion AI supercomputer

[00:46:55] Mike Kaput: So in some other microsoft and OpenAI appear to be planning to build a 100 [00:47:00] 100 billion dollar AI supercomputer called Stargate.

[00:47:04] Mike Kaput: I love that name. That's

[00:47:05] Mike Kaput: a great name. You gotta always have a slightly, epic and sinister sci fi name for these, I think. So, this coming from some reporting from the information.

[00:47:17] Mike Kaput: They spoke to several sources familiar with this project, and according to the sources, the supercomputer would contain millions of specialized chips to power OpenAI's artificial intelligence and could cost upward of 100 billion. Now,

[00:47:34] Mike Kaput: for some perspective here, this is

[00:47:36] Mike Kaput: a hundred times the cost of some of today's largest data centers.

[00:47:41] Mike Kaput: And while this project has not yet been formally greenlit by Microsoft, it appears to track with recent commentary from both Microsoft and OpenAI and many others about the need for massive investment in core infrastructure like the chips that power [00:48:00] AI. For instance, in episode 83, we had reported on how Sam Altman was possibly seeking up to trillions of dollars to build AI chip infrastructure.

[00:48:10] Mike Kaput: Now, with this particular project, there is one big catch. The information says that, quote, Microsoft's willingness to go ahead with the Stargate plan depends in part on OpenAI's ability to meaningfully improve the capabilities of its AI. Which seems to indicate that gPT 5 better seriously impress Satya Nadella.

[00:48:31] Mike Kaput: And if it does, executives at Microsoft have, so far, discussed launching Stargate as soon as 2028 and expanding it through 2030. Now, Paul, let's talk about the kind of big picture implications here. This is 100x bigger than anything they've been doing before in this space, so it's pretty clear they're thinking very, very big and seriously planning for the future.

[00:48:57] Mike Kaput: Like what type of AI future do they [00:49:00] see where they need this level of infrastructure?

[00:49:03] Paul Roetzer: So I keep mentioning episode 87 today, but if you go to episode 87, the AI timeline, we kind of address why in that episode, but the why here is openAI in particular, and I assume Microsoft,andmany other leading AI

[00:49:19] Paul Roetzer: AI researchers and companies. Right now, continue to believe that the advancements in intelligence, um, will follow scaling that if you give more computing power, more, which is powered by by more chips, more data,

[00:49:37] Paul Roetzer: means more data centers. So if you keep building these, this infrastructure capability, that the models will continue to evolve. To get smarter. and they're, you know, NVIDIA, obviously this is why they're, making so much money, um,

[00:49:54] Paul Roetzer: cause they, they power a lot of the training runs. So these training runs are going going to get into the billions of like it's [00:50:00] rumored, I I think GPT 4 took. I don't know, hundreds of millions maybe to train, but we're going to,

[00:50:06] Paul Roetzer: the next couple of years, we're going to to start multi billion dollar training runs for these major models, and that's just the training.

[00:50:15] Paul Roetzer: Then you get into the inference, which is the moment you or I ask it to do something. So you go give it a prompt, there's chips powering the compute, That enables it to give an answer or to generate an an image, or generate a video. So if Sora wants to be able able to generate five minute videos, That a lot of compute power at inference to do that. And so

[00:50:38] Paul Roetzer: so they're building all of models. So they're looking out out ahead five years, 10 years out and saying, okay, if scaling laws continue to go, if we keep giving it it more chips, we keep giving it more data, we are are going to need way more data

[00:50:52] Paul Roetzer: We're going to need way more energy capacity, like nuclear fission, nuclear fusion, it's why Sam invests in a nuclear fusion company.[00:51:00]

[00:51:00] Paul Roetzer: We're going to need stronger electrical grids. We're going to need all this stuff. So they're preparing for a world where this

[00:51:07] Paul Roetzer: this intelligent explosion occurs, and we need way more infrastructure to do the training of the models and the inference for the the use the the models. and that's basically what happens.

[00:51:18] Paul Roetzer: I mean, just this morning. TSMC, so Taiwan Semiconductor Manufacturing Company, which, is the linchpin to the generation of these chips. it's obviously in Taiwan. they announced this morning, the Biden administration said on Monday, a plan to send up to 6. 6 billion billion in federal grants to TSMC.

[00:51:42] Paul Roetzer: and I think another 5 million in loan guarantees, uh, for a 25 billion Arizona expansion bring a third TSMC fabrication plant to that state, fabrication

[00:51:54] Paul Roetzer: is where they, fabrication of the chips. so so yeah, this is a major [00:52:00] thing and, I think we're going to keep hearing more and and more about. Investments being made in this space. And again, from a not investing advice, but when you think about where the money is going go, and people always ask about like, what do you invest in an AI?

[00:52:15] Paul Roetzer: Think about the think about all the companies that are going to make all of this possible. and that's a good starting point to think about where, where's the money and the value going to be generated over the next decade. A lot of of it is going to be in the companies that make aI possible.

[00:52:33] Google considers charging for AI search and Perplexity plans to sell ads

[00:52:33] Mike Kaput: So in some other news, we have two big AI announcements that are shaking up search potentially. So first, the Financial Times reports that Google is considering charging for its AI powered search features. One option being discussed by Google executives is adding AI powered search to its premium subscription services.

[00:52:55] Mike Kaput: So in this proposed plan, Google's traditional search engine would remain [00:53:00] free while ads would still appear on basically everything. I mean according to the Financial Times, quote, Google's traditional search engine would remain free of charge while ads would continue to appear alongside search results even for subscribers.

[00:53:15] Mike Kaput: Now, that's just one pathway being considered. It is not yet official, but interesting to note. Now, while Google is exploring new revenue models, The AI search startup and Google challenger, Perplexity, which we talk about all the time, is embracing a tried and true revenue model, which is advertising.

[00:53:37] Mike Kaput: Adweek has reported that Perplexity is going to start selling ads. Perplexity It doesn't sound these are going to be the traditional kind of search ads that we're used to, at least to start. The company plans to start introducing native ads in the related questions section that you see at the bottom of perplexity results.

[00:53:58] Mike Kaput: So, You'll see if [00:54:00] you use Perplexity to search for anything or ask any type of question. A section pops up titled Related that has all these follow up questions and suggested topics you can kind of inquire about next. And soon brands will be able to actually influence what shows up in those questions.

[00:54:19] Mike Kaput: Um, Perplexity has not confirmed exactly when these ads will roll out, but they did say they will launch in upcoming quarters. So Paul, what did you make of these two announcements and how they relate to Surge? Yeah,

[00:54:34] Paul Roetzer: of this? I mean, we've talked about it before, like Google's dominant, you know, know, moneymaker, profit engine is, ads. and people, like the way the generative search

[00:54:47] Paul Roetzer: works right now, the ads aren't, it's not obvious how you're going to get as many throughs on the ads. So I think Google's looking and and saying like,

[00:54:55] Paul Roetzer: how, how are how are we going to evolve? So we're not as dependent upon the 10 blue [00:55:00] links or or ads that appear above the 10 blue blue links. The 10 ads that appear above the 10 links. and Perplexity's is realizing like, wow, you really need an ad model make this viable. So, yeah,

[00:55:13] Paul Roetzer: it's just like the, know, the ongoing issue of Genitive AI seems to be what consumers are going to prefer. How do do we make doing it? I have been a big advocate. You know, I really don't want ads and perplexity.

[00:55:26] Paul Roetzer: I'm paying my 20 a month happily, but I also know that 20 a month is not going to be sufficient for them to build a scalable business model

[00:55:34] Paul Roetzer: that's going to keep getting, you know, the hundreds of millions or billions of dollars in funding that they're going to need to actually compete and be, anything of significance to a to Google's market share. Cause as much as people talk about perplexity and much you and I like it, I mean, it

[00:55:49] Paul Roetzer: it is a. A very, very minuscule threat at all to Google's dominance in search right now. It's a, I don't know, it's probably not even like a a [00:56:00] tenth of a percentage point, I wouldn't imagine.

[00:56:03] Paul Roetzer: so now that can change, like obviously disruption can happen fast, but they get a a lot of love, it's not like it's having any measurable impact on Google's business at the moment.

[00:56:13] Demis Hassabis’ rising profile within Google

[00:56:13] Mike Kaput: So related to some more Google news, there's two new stories that came out that kind of have us wondering, you know, we've talked about if Google is kind of trying to raise the profile of Demis Hassabis, who's the head of its AI efforts at Google, at Google DeepMind. Now we've spoken about Demis since literally the beginning of Marketing AI Institute as one of the most important people.

[00:56:37] Mike Kaput: in AI. He's been integral to Google DeepMind, which resulted from Google acquiring his company in 2014. He's been personally involved in major AI breakthroughs, including AI beating a human champion at the game of Go, and AlphaFold, which is predicting the 3D structure of basically all known proteins. And these two stories that [00:57:00] came out are all about Demis.

[00:57:02] Mike Kaput: and how he is essentially trying to save Google from itself. one is called, quote, Can Demis Hassabis Save Google from Big Technology? And the other is called, The Chess Master Trying to Propel Google's AI Push from the Wall Street Journal. Both are worth a read to kind of understand how did Google get to the current crossroads it's at with AI, and how did Demis influence that, and what could he influence going forward.

[00:57:30] Mike Kaput: But I think they're also really important for a different reason, which is, there seem to be efforts within Google to raise Demis profile, and one of these articles even blatantly speculates he may be tapped to be the next CEO, though he denies that. Now, Paul, can you connect the dots for us here? Um, why does this matter?

[00:57:51] Paul Roetzer: It just seems Strategic on Google's part or on Demis part. Like, [00:58:00] someone is intentionally positioning him in this way. it would would really,

[00:58:09] Paul Roetzer: The more I thought of it, it would be really weird if it was Google. Google. Because you would think Sundar would would have to be signing off on this strategy.

[00:58:16] Mike Kaput: Right.

[00:58:17] Paul Roetzer: I don't, see, I

[00:58:18] Paul Roetzer: I don't know how Google works though internally their spokespersons and like what people are allowed to say. Like I I know Amazon is very rigid with who says what publicly. like I think like I think at Amazon you have to go speaker training before you're even allowed to go represent Amazon in an event.

[00:58:36] Paul Roetzer: I don't know how Google works from that perspective, but I mean, I mean, Demis is an is an extremely. Influential and important person in AI. He's obviously extremely influential and important within Google.

[00:58:52] Paul Roetzer: There's certainly been questions around sundar's leadership and if he's the right CEO [00:59:00] for where they're going. I don't know that Demis would want the job.

[00:59:05] Paul Roetzer: Like, I mean, he didn't even, he sold to Google so he could be a researcher. Like he didn't want to be a commercial, you know, CEO of a commercialized company. Like he's a, he's a researcher through and through, so I don't

[00:59:19] Paul Roetzer: I just interesting to keep an eye on. I listen to every interview Demis. Ever does, like, if there's ever a podcast that comes out, I listen to it, like, I just get enough. Like, I think he's so smart. think he's a genuine person, um,

[00:59:34] Paul Roetzer: and I think he's going to be very important to where humanity goes. Not, not just within AI and, know, marketing and business, but like just humanity in general. So yeah, I don't know. Give it a listen. We'll drop a, there was a podcast he did recently with Dwarkesh, in February that

[00:59:49] Paul Roetzer: that actually, was part of the motivation for the aI timeline episode we did in March. that one of the, one of the things I listened to on my trip that I, when I created that timeline.[01:00:00]

[01:00:00] Paul Roetzer: So it's a really good interview. If you've never listened to an interview with Demis, there's all kinds of great podcast interviews he's done. He's been on Lex Fridman. go listen to some of them.

[01:00:09] Paul Roetzer: He's amazing guy, really fascinating to hear what he has to say.

[01:00:14] HubSpot’s new AI-powered content hub

[01:00:14] Mike Kaput: So another development this past week, HubSpot has added a host of AI features to its existing CMS Hub product and rebranded this into kind

[01:00:24] Mike Kaput: a new AI powered offering called Content Hub. And Content Hub now has features like AI content repurposing capabilities, something called brand voice, which creates on brand content at scale that's written in your unique brand voice.

[01:00:41] Mike Kaput: And things like AI translations, AI image generation, blog generation, blog post narration, and more. So, Paul, we've HubSpot having both, you know, worked at and you owned HubSpot's first ever partner agency. As a longtime HubSpot watcher, what [01:01:00] does this mean for HubSpot customers?

[01:01:03] Paul Roetzer: I mean, they've they've been pretty methodical. About generative AI. and I mean by that that is they haven't really been like first to market on any of this stuff. Like they're, kind of of like taking their time, plotting right plays that'll

[01:01:21] Paul Roetzer: add value to their customers. Uh, these seem like really good. I I don't, we haven't did you test any of these yet within our our platform? Have we had a to look at any of these?

[01:01:30] Mike Kaput: I've tested a couple of the content generation features, but not yet the remixing or the

[01:01:36] Mike Kaput: or the brand voice features.

[01:01:37] Paul Roetzer: Yes, so my, my main take, think is like, What is the value of it embedded within here versus we have Claude, we we have

[01:01:49] Paul Roetzer: we have Gemini, like we're already paying for all these other things that do similar things.

[01:01:54] Paul Roetzer: Uh, I don't know if they're using, I assume they're using GPT to do this. Like they haven't [01:02:00] built their own model. And I know that. so then

[01:02:02] Paul Roetzer: then the question becomes like, Okay, so if I can go in and I can take a landing page and and I give it to this remix thing and it then writes my emails and social shares and whatever,

[01:02:12] Paul Roetzer: if I've I've already trained a GPT to do the exact same thing and that GPT is using GPT 4 or 4. 5 or whatever comes next, am I I just better off just continuing to use GPT? ChatGPT, like, I don't know.

[01:02:26] Paul Roetzer: And so that that would be my question as a user, these capabilities of your your HubSpot customer, compare them to other tools you have and say, Hey, actually, this is great. Like it's integrated with everything got in HubSpot.

[01:02:37] Paul Roetzer: It saves me 20 minutes when I'm doing this. Like, I'm just going to go ahead and use this. but like we're

[01:02:43] Paul Roetzer: we're always saying, like, like it's all about keep testing. You got these use cases, go play around with it, see if it's better than than what you're using, and keep up with it. And I would, the would, the other thing I would say is

[01:02:53] Paul Roetzer: so last week, story was released that Google was considering an an [01:03:00] acquisition offer for so it's probably worth at least mentioning this. at this point, all it was, was someone leaked it. I

[01:03:07] Paul Roetzer: I don't know would leak this or why they would leak it other than maybe somebody who wanted to push this forward a little bit is that Google was talking to their, customers.

[01:03:14] Paul Roetzer: investment bankers about the idea of putting an offer together to buy HubSpot. That was, that was the story. Not that an offer was pending, not that you know, it had been made. yep. I have to remember like back in pre IPO days, like 2010 or 11, I think the last, round of funding HubSpot raised, I'm not not mistaken.

[01:03:34] Paul Roetzer: It was 32 million, maybe, Google Ventures was one of of the back then. Like, it's not like they don't know each other. So it's kind kind of logical. There's a great, segment on the All In podcast. I don't know if you had a chance to listen to that, Mike, but where they talked about why this deal might or might make sense. Yeah. so it it kind kind of worth following. it would by far Google's biggest acquisition ever. So their, I just, I [01:04:00] pulled this up. Their biggest acquisition ever

[01:04:01] Paul Roetzer: was 12. 5 billion for Motorola. Mm. Other than that, it it was 3. 2 billion for Nest and YouTube was like 1. 65 billion. So hubSpot would have be 40 to 50 billion. I mean, HubSpot's like a $35 billion market cap company right now. You gotta pay a premium on top of that.

[01:04:17] Paul Roetzer: It's probably gonna be in in the 40 to 50 billion range. so would it even make sense google? I don't but it's out there, so I just wanted to make sure we we at least acknowledge it.

[01:04:27] OpenAI Voice Engine and voice deepfake concerns

[01:04:27] Mike Kaput: So OpenAI has shared the results of a small scale preview using a model they have called VoiceEngine, which can accurately simulate voices using just 15 seconds of audio. So this model can create what OpenAI calls emotive and realistic voices using just 15 seconds of an audio sample, which you can then use to create a voice that would say anything you like.

[01:04:53] Mike Kaput: Simply based on whatever text you give it. Now, OpenAI recognizes this has the potential to [01:05:00] go very, very wrong. they released a statement about this saying, We recognize that generating speech that resembles people's voices has serious risks, which are especially top of mind in an election year. However, they also shared a bunch of positive uses of the technology, including helping patients recover their voices. So this all comes amid heightened security, concerns over the security of elections and also the potential for fraud here in a lot of different industries, including financial services, which somewhat related.

[01:05:34] Mike Kaput: The Wall Street Journal also mentioned this. Just documented a report that financial services firms are now racing to counter AI generated calls that use deepfaked audio to game voice authentication software. So, Paul, you have wrote a bit on LinkedIn about this, and specifically that OpenAI has had this type of technology.

[01:05:56] Mike Kaput: Since late 2022, and didn't release it due [01:06:00] to concerns around misuse. You wrote that other AI companies won't show this restraint. I mean, are we ready for voice deepfakes and synthetic audio?

[01:06:11] Mike Kaput: Obviously we are

[01:06:12] Paul Roetzer: are not. It's a society.

[01:06:15] Paul Roetzer: Um, yeah, my, what I was I was implying on the others won't have restraint is this is the, this is the argument of like, the EAC movement, the techno optimists, is like, just create it all, release it all. It's like, really? Like, are we really better off if we give everyone this kind of capability? This is where I

[01:06:31] Paul Roetzer: I like, sometimes with that movement And a lot of times I'm like, but there has to be some middle ground here. And this is where I think a middle ground makes a lot of sense that maybe we should

[01:06:41] Paul Roetzer: not put all this stuff out in the world. But the way these things work is like, if you're at one lab and you are are very on this technology and the company you're at doesn't want to release it, then you're just going to go somewhere else and release it.

[01:06:53] Paul Roetzer: Somewhere else under an open source model. So it's going to get out. Like if we don't already have this ability from another [01:07:00] company, these research labs are all working on the same problems. The moves between the the research labs all the time.

[01:07:08] Paul Roetzer: as an an example, Noam Brown, who's an open AI researcher who who we've talked previously when he was at Meta and solved the game of diplomacy.

[01:07:17] Paul Roetzer: And prior to that, Texas Hold'em and and things like that. So he's now at at OpenAI and he said in tweet, haven't disabled voice authentication for your bank account and had a conversation with your family about AI voice impersonation yet,

[01:07:32] Paul Roetzer: Now would be a good time because he knows that just because OpenAI hasn't released, it doesn't mean somebody else isn't going to read

[01:07:38] Paul Roetzer: And so that would be be my takeaway for you again, like personal perspective, likeIwas in my Schwab account yesterday and it has an option to turn on voice authentication. Absolutely not. Like it's, soIdon't know

[01:07:52] Paul Roetzer: know why that. And in the banking industry and other industries, like that's a thing right because that seems like a really dangerous [01:08:00] capability to offer, because this is pretty obviously technology that's two years old already.

[01:08:06] Paul Roetzer: and so I think it's a good reminder too, that when see that the AI we see today, and we're blown away by it, just know that you're not

[01:08:15] Paul Roetzer: even seeing the most advanced stuff. And like, in some ways there's going to be things that are created that you can't even comprehend.andsometimes

[01:08:25] Paul Roetzer: text like a year or old already before we see it. So just a good reminder that these research labs and the governments have AI technology we don't know about yet, that is far more advanced than what we think we know. The most advanced technology in in the world is.

[01:08:42] Google Deepmind paper

[01:08:42] Mike Kaput: That is a good reminder. Now, As we wrap up here I am going to touch on two final stories.

[01:08:50] Mike Kaput: Um,

[01:08:50] Mike Kaput: the first is that we got a new research paper from Google DeepMind that appears to debunk some of the major concerns that we all have about large [01:09:00] language models. Specifically, it's a big concern that they don't get their facts straight, that they get factual information wrong or make stuff up, and it's a huge barrier to LLM adoption.

[01:09:12] Mike Kaput: but according to this research, the paper finds that certain LLM agents may be able to achieve superior performance on certain fact checking tasks when compared to crowdsourced human annotators using a method called Search Augmented Factuality Evaluators, SAFE. Basically, it breaks down responses into individual facts, then fact checks each of them.

[01:09:35] Mike Kaput: And the researchers using this approach found that compared to human fact checkers, LLMs using this process were better at using Google search when they were given access to the internet to accurately fact check information. And it also found that LLMs are 20 times cheaper than human fact checkers for these types of tasks.

[01:09:57] Mike Kaput: So really what we're looking at here, Paul, it [01:10:00] sounds like, is that. There may be pathways forward where some of the big concerns around LLMs are solved for in the relatively near future. Was that kind of your read on this?

[01:10:11] Paul Roetzer: Yeah, I mean, I think I talked about again in episode 87, like just assume any current weakness or limitation of these models is going to solved for in the not too distant future. And so the takeaway there's don't build a business, like don't

[01:10:27] Paul Roetzer: create a startup around a limitation of these models. and don't pick a career path built on a current limitation of these models. You just kind of have to assume all this stuff.

[01:10:37] Paul Roetzer: we talk about memory as another example, it's to reason, it's ability to get facts correct. All of of that stuff is being solved for right now in every major research lab.

[01:10:50] You can now edit your images in ChatGPT

[01:10:50] Mike Kaput: All right. last but not least, you can now edit the images that you create in ChatGPT. Now after you use DALL E3 right in [01:11:00] ChatGPT to generate an image, you can now edit that image using text prompts. To do that, you just select an area of the image that you want to edit and then describe your changes.

[01:11:11] Mike Kaput: using a natural language prompt, or you can just tell chatGPT what edit you

[01:11:17] Mike Kaput: to make and where to make it. So Paul, this is, it sounds like similar functionality to things like Adobe's Generative Fill and Mid Journey's certain capabilities in some of the new versions. Like, does this in your mind make ChatGPT a serious image generation competitor?

[01:11:36] Paul Roetzer: Uh, I think it, it's more that, they have distribution, so you can be a,

[01:11:42] Paul Roetzer: know, you can follow on later with the same features like Apple's going to do with a bunch of their generative AI stuff. And as long as you got a bunch of customers who are using

[01:11:50] Paul Roetzer: product over the other ones, like I I don't pay for products, so I I don't have that capability.

[01:11:56] Paul Roetzer: I played around with with it weekend. It's pretty slick. Like it's, [01:12:00] it works right in the app. I don't have have to log into computer to do it. So yeah, I think it's just an example of you can follow on with stuff. As long as you have a customer base that, we'll use use it. And

[01:12:10] Paul Roetzer: it it seems like a really nice feature add on. And I think we get Sora, we're going to to get even more capabilities. think Sora is going to be,

[01:12:17] Paul Roetzer: I don't know if it's going to replace DALL E, but it seems like they're kind of positioning it as both and video generation and editing.

[01:12:24] Mike Kaput: All right, Paul, we through a ton of topics today. As a quick reminder to everyone, if you are not subscribed yet to our newsletter, we cover not only the topics. This week in AI that we covered on the podcast, but we also go more in depth into them in the newsletter and we cover all the other topics that we couldn't get to.

[01:12:44] Mike Kaput: So go to marketingainstitute. com forward newsletter to get that weekly brief on what is happening that's most important in artificial intelligence. Last but

[01:12:55] Mike Kaput: least, I would also say if you have not. rated the [01:13:00] podcast yet and you have found value in it, please, please, please take a moment to just leave us a quick review on your podcasting platform of choice.

[01:13:08] Mike Kaput: It helps us get the episodes into the hands of even more people. So we very much appreciate that. Paul, thanks so much for breaking down this week in aI.

[01:13:18] Paul Roetzer: Thank you,

[01:13:18] Paul Roetzer: Mike, and thanks everyone

[01:13:20] Paul Roetzer: listening. As always, we'll talk to you next

[01:13:22] Paul Roetzer: week.

[01:13:23] 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.

[01:13:46] Until next time, stay curious and explore AI.

Related Posts

[The AI Show Episode 89]: A New In-Depth Sam Altman Interview, The “Inflection Point” for Enterprise Generative AI Adoption, and Inflection AI’s Big Shakeup

Claire Prudhomme | March 26, 2024

The AI Show analyzes tough news for generative AI companies, including Sam Altman's latest interview, a16z's enterprise research, changes at Inflection AI, and more.

[The AI Show Episode 84]: OpenAI Releases Sora, Google’s Surprise Launch of Gemini 1.5, and AI Rivals Band Together to Fight Deepfakes

Claire Prudhomme | February 20, 2024

Episode 84 provides insights on OpenAI's Sora for video generation, Google's Gemini 1.5, and tech giants' aim to regulate deepfakes with the C2PA standard.

[The Marketing AI Show Episode 57]: Recap of 2023’s Marketing AI Conference (MAICON), Does Sam Altman Know What He’s Creating? and Generative AI’s Impact on Jobs

Cathy McPhillips | August 1, 2023

This week's episode of The Marketing AI Show talks about MAICON 2023, a mind-blowing article on Sam Altman, and generative AI's impact on jobs.