AI takes center stage at Google Cloud Next, WPP inks major ad deal, and hot AI startups make waves! Join Paul, and Mike as they dive into Google's AI-powered cloud offerings, WPP's collaboration with Google on AI-generated ads, and the latest developments from buzzworthy startups like Humane AI Pin, Udio AI, and Captions.
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.
00:02:25 — Google Cloud Next
00:18:43 — Google’s Blockbuster Ad Deal with WPP
00:30:39 — Hot AI Startups Getting Attention - Humane AI Pin’s Brutal Early Feedback
00:37:05 — Hot AI Startups Getting Attention - Udio AI Music Generator
00:41:40 — Hot AI Startups Getting Attention - Captions AI-Powered Creative Studio
00:47:16 — Brands Adding AI Restrictions to Agency Contracts
00:51:35 — AI Legislation Updates
00:55:09 — Google’s Demis Hassabis Chafes Under New AI Push
00:58:27 — Gemini for Google Workspace Prompt Guide
01:02:23 — TikTok Plots Using Virtual Influencers for Advertising
01:04:50 — ChatGPT Is Now “More Direct, Less Verbose”
01:07:40 — Adobe’s ‘Ethical’ Firefly AI Was Trained on Midjourney Images
AI Announcements at Google Cloud Next
Google just wrapped up its Google Cloud Next event, a major conference that showcases the tech giant’s cloud infrastructure arm and AI played a starring role at the event.
Some of the big AI updates announced by Google include the availability of Gemini for Google Cloud, which integrates the latest Gemini models across a range of Google Cloud capabilities, including:
An AI Hypercomputer that the company now offers for generative AI infrastructure.
We also saw more AI-powered features announced for Google Workspace, including AI meetings and messaging add-ons that will take notes for you. And we got an announcement of Google Vids, a new AI-powered video creation app.
Google is also leaning heavily into AI agents.
Google Cloud’s CEO Thomas Kurian kicked off the conference with his extensive talk about how AI agents would be used by businesses to automate many different types of tasks, including analyzing information and helping shoppers find the right items to buy.
The primary vehicle for this discussed by the company at Next is the Vertex AI Agent Builder, which allows people to build conversational agents. This is a no-code product that allows you to build conversational agents on top of Gemini, and have the answers those agents provided by “grounded,” or tied to something considered a reliable source.
Google Cloud also announced Creative Agent, which is an image and text generation AI tool that can help marketers and businesses develop storyboards and personalized content for ads and social media.
Google’s Blockbuster Ad Deal with WPP
Google and the world’s biggest advertising group are teaming up on generative AI. WPP has announced a major collaboration with Google that will see the advertising group use AI to produce ads.
The partnership will focus on integrating Google’s Gemini models with WPP’s existing AI-powered marketing operating system, called WPP Open. Which means WPP clients like Coca-Cola, L’Oréal, and Nestlé who use that system will get access to AI-powered ad generation capabilities.
The first phase of the partnership focuses on four use cases:
First is enhanced creativity. WPP Open Studio will integrate Gemini 1.5 Pro to help with basic creative tasks like writing headlines or turning sketches into images by pairing brand content and guidelines with Gemini 1.5 Pro’s 1 million token context window.
Second is content optimization. Gemini will provide an upgrade to WPP’s AI system that predicts how well marketing content will perform.
Third is AI narration. WPP is going to use Gemini to automatically create video narration scripts, then send these scripts to ElevenLabs to generate synthetic voices for the video.
Fourth is product representation, where WPP will use generative AI to create product images.
Hot AI Startups Getting Attention This Week
Several hot AI startups are getting attention this week—and not all of it is good...
The Humane AI Pin is an ambitious wearable device that aims to replace your smartphone, but it comes with a steep price of $699 and a $24 monthly subscription.
The concept is intriguing: a screenless gadget that relies on an AI assistant to handle everything from phone calls and text messages to answering questions and snapping photos. However, the AI Pin falls short of delivering on its promises.
The device has faced criticism for it’s unfinished features, inconsistent performance, and a host of bugs that make it difficult to recommend in its current state.
Udio AI’s launch is a formidable competitor to Suno’s v3 model. Suno, released to the public just weeks ago, was a remarkable breakthrough, particularly in realistic, human-sounding vocals.
Unlike Suno, which is focused on putting music-making tools into the hands of average consumers, Udio also sees itself as a tool for musicians, and the founders say its creative abilities along those lines should ease creators’ concerns about their use of training data.
“We have been guided by musical people from the outset,” says Sanchez, “and what that means is that we’re making a product which is going to enable those folks to create great music and, to be clear, to make money off of that music in the future.”
Captions, an AI-powered creative studio that started as a popular iOS app, has now expanded its capabilities to web and desktop users. This innovative tool offers a range of features designed to simplify video editing using AI.
The tool has impressed early adopters, including our team, who are exploring its potential for creating podcasts, videos, and event footage.
Backed by $37.5 million in funding from leading VC firms such as Kleiner Perkins, Andreessen Horowitz, and Sequoia Capital, Captions is poised to make a significant impact in the world of video editing.
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
[00:00:00] Paul Roetzer: I mean, we're talking about. Lots of different expertise is brought together in different roles across creative teams that you have single foundation models that can do all of it now. again, I am not saying better than humans or in place of humans.
[00:00:16] Paul Roetzer: I'm just asking questions out loud here of like things that I don't have the answers to.
[00:00:20] 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.
[00:00:58] Paul Roetzer: Welcome to episode 92 [00:01:00] of the Artificial Intelligence Show. I'm your host, Paul Roetzer, along with my co host, Mike Kaput. Good morning, Mike.
[00:01:06] Mike Kaput: Morning, Paul.
[00:01:07] Paul Roetzer: We are recording this April
[00:01:10] Paul Roetzer: 15th. 1025 AM Eastern time. In case, as always, something breaking happens today. I feel like last Monday, there was like three new models introduced.
[00:01:19] Paul Roetzer: So who knows what Monday is going to bring us this week. All right. Today's episode is brought to us by Rasa. io. 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.
[00:01:39] Paul Roetzer: io is the 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. 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 [00:02:00] must read.
[00:02:01] 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, we have a lot of Google talk today. This is like the Google episode, I think. So let's get started with everything
[00:02:24] Paul Roetzer: Google.
[00:02:25] Mike Kaput: Yeah, it really is, and the first big thing related to Google this week is they just wrapped up their Google Cloud Next event, which is a major conference that showcases the tech giant's cloud infrastructure arm.
[00:02:39] Mike Kaput: Now, at this year's event, AI played a very significant role. You could even say it played a starring role, because many of the big updates that were announced were related. to artificial intelligence. So, for instance, we now have Gemini for Google Cloud, which integrates the latest [00:03:00] Gemini models across a range of Google Cloud capabilities.
[00:03:04] Mike Kaput: This includes things like Gemini Code Assist, which offers AI powered assistance to developers. It includes Gemini Cloud Assist, which gives AI guidance on how to manage applications in the cloud. in Google Cloud. And they have things like Gemini in security, which boosts security team productivity using Google Cloud.
[00:03:23] Mike Kaput: And there are many other Gemini applications being baked right into the platform. Among some other significant updates, they announced that they now have something called AI Hypercomputer, which the company now offers for generative AI infrastructure, which includes a number of. different advanced chips to help you run Google Cloud applications better.
[00:03:46] Mike Kaput: And there were a ton of other announcements around new models, new chips, and new Google Cloud features and services. So what we wanted to do here is actually highlight a few of the ones that appear to be quite [00:04:00] significant and worth diving into further. Now, one of these is all the talk. that was had at the event around AI agents.
[00:04:08] Mike Kaput: So Google Cloud CEO Thomas Kurian, in his keynote to kick off the conference, talked pretty extensively about how AI agents would be used by businesses to automate many different types of tasks. That includes things like analyzing information and helping shoppers find the right items to buy. The primary vehicle for AI agents that was discussed at the event is something called Vertex AI Agent Builder.
[00:04:35] Mike Kaput: which allows people to build their own conversational agents on top of Google Cloud. This is actually a no code product that allows you to build these agents and using Gemini models to have those agents provide answers that Google calls, quote, grounded. So that means something, they are tied to something considered a reliable source.
[00:04:58] Mike Kaput: So in one demo, the [00:05:00] company used this to create an agent that analyzed previous marketing campaigns and understood a company's brand style and then generated new ideas out of it. consistent with that style. Google Cloud also announced Creative Agent, which is an image and text generation AI tool that can help marketers and businesses develop things like storyboards or personalized content for ads and social media.
[00:05:28] Mike Kaput: We also got some more AI powered features for Google Workspace, including AI meetings and messaging add ons that will do things like take notes for you. And we got an announcement of something called Google Vids, which is a new AI powered platform. So, Paul, there's a lot going on here, but first up, maybe talk to us a little bit about any of the updates in particular that jumped out at you, some of the trends that you're seeing from what was discussed at Google Cloud Next.
[00:05:59] Paul Roetzer: [00:06:00] There's lots of interesting stuff around, like, you know, the more practical things I'm looking for, like how it's going to be used in Google Workspace. I mean, so much over the last year, we've talked on this show about once these things are infused into Workspace, once generative AI is, you know, at our fingertips and docs and sheets and all these other applications, that would really start to affect us.
[00:06:20] Paul Roetzer: So certainly I was interested in that area, but the AI agent thing, you know, far and away was the thing that captured my interest. attention. And so I think, you know, I probably want to focus our conversations around that. And I had a few initial thoughts here. So I went back to episode 87, where we talked about the AI timeline.
[00:06:42] Paul Roetzer: And I just want to revisit this, because again, you hear all this, and you might think, oh my gosh, agents are here, it's going to change everything. And so I want to kind of revisit for people where I think we are with this. with agents and then dive a little bit into exactly the kinds of agents that they're envisioning building.
[00:06:58] Paul Roetzer: So in episode 87, when we [00:07:00] laid out this timeline, said, 2025 to 2027 is when we would have the AI agents explosion. at the time I said, we will see lots of talk about agents that can take action. For example, use your keyboard and mouse in 2024, but I think this year is mostly experimentations and demonstrations, maybe the equivalent of GPT 1 or GPT 2 level, if we're thinking about the equivalent in a language model.
[00:07:23] Paul Roetzer: Um, continuing on from episode 87, lots of manual work to get them to function properly, plus lots of human oversight. Not close to full autonomy yet, and most people generally won't be willing to give up the data and privacy needed to get the benefits. Starting in 2025, AI can now take actions reliably with limited human oversight, probably in select domains and verticals initially.
[00:07:47] Paul Roetzer: Then more generally horizontal, horizontally applicable. Some early instances of full autonomy starting in 2025, but not widespread. Disruption to knowledge work starts to become more tangible and measurable. So, [00:08:00] nothing I saw from the Google announcement changes my current perspective on agents. Um, They defined agents in the opening talk as intelligent entities that take actions to achieve goals.
[00:08:15] Paul Roetzer: And they said agents can connect with other agents, which is certainly interesting and expected. What that means is you have an agent trained to do one thing that can work with an agent trained to do another thing, and they can string those agents together to basically do what humans do. So there was, I think it was, let's see, there was one, two, three, four, five, six These are the specific types of agents that they talked about.
[00:08:39] Paul Roetzer: And so I wanted to kind of read through briefly what each of these six agents are. And so again, when you go back to the timeline, we talked about this idea of sort of these like tuned agents that are built specifically to do vertical things. Um, versus just generally able to do everything. And so there's actually a really, useful blog [00:09:00] post that they published on April 12th.
[00:09:01] Paul Roetzer: We'll put in the show notes called 101 Real World Gen AI Use Cases from the World's Leading Organizations. And in this post, they break down the six agents. So the first customer agents, now I want you, as I'm explaining these, to think about the impact this has on, Human jobs, um, as complimentary and assistive, but certainly you can listen to some of these descriptions and start to imagine where agents can do the work of some humans.
[00:09:32] Paul Roetzer: they're obviously they're obviously going to say that in this talk. They're not going to talk about the impact on jobs. But we'll get there. So customer agents, similar to great sales and service people, customer agents are able to listen carefully, understand your needs, and recommend the right products and services.
[00:09:49] Paul Roetzer: They work seamlessly across channels, including the web, mobile, and point of sale, and can be integrated into product experiences with voice and video. Again, I recommend going [00:10:00] and reading this post because they then have. 19 examples of companies that are building these kinds of agents and using these kinds of agents.
[00:10:08] Paul Roetzer: The next one is employee agents. I was really interested to see how they explained this one. Um, Employee agents help workers be more productive and collaborative, collaborate better together. These agents can streamline processes, manage repetitive tasks, answer employee questions, as well as edit and translate critical communications. similar, they have
[00:10:31] Paul Roetzer: 27 examples of companies that are using these. And it's, you know, examples like, um, Bear, Avery Dennison, Syntos, Home Depot, the LA Rams, like they're like talking about LA Rams using it for content analysis to player scouting, McDonald's, Uber. Oh, interesting quote from Uber. so it's a little side note as I'm seeing this.
[00:10:56] Paul Roetzer: So it says, Uber is using AI agents to help employees be more productive, save [00:11:00] time, and even more effective at work. For customer service reps, they've launched new tools that summarize communications with users and can even surface context from previous interactions so frontline staff can be more helpful and effective.
[00:11:12] Paul Roetzer: But they had the CEO of Uber in the opening talk, they showed, they do video clips of like, customers that are using this stuff. He literally said, reduces our agency spending. I just, I mean, we'll get into the agent stuff. We talked about agent stuff last week. like.
[00:11:29] Paul Roetzer: sometimes they're saying the quiet part out loud.
[00:11:31] Paul Roetzer: okay. The next one, creative agents, creative agents can expand your organization with the best design and production skills, working across images, slides, and exploring concepts with workers. Many organizations are building agents for their marketing teams, audio and video production teams, and all the creative people that can use a hand.
[00:11:52] Paul Roetzer: With creative agents, anyone can be a designer, artist, or producer. Interesting. Um They show seven examples there, [00:12:00] including WPP, which we will come back around to in a couple minutes. now as I'm reading these, we've got three to go. This is just Google putting a name to specific AI capabilities. And everybody's building the same stuff.
[00:12:14] Paul Roetzer: So OpenAI is building the same stuff. They may call it something different. They may not. They just may just be ChatGPT. What Google's doing is trying to productize specific sorts of agents. They're trying to make use cases. really obvious to people so that they can commercialize this faster. It's all just Gemini, like it's, it's, it's all the same base models being applied in different ways.
[00:12:37] Paul Roetzer: Or you can go into Vertex AI
[00:12:38] Paul Roetzer: AI and use
[00:12:39] Paul Roetzer: Mistral or Cohere or whatever. It's other models, but basically it's just this Generate capability with a name in essence. Okay. The next one, data agents are like having knowledge, data, knowledgeable data analysts and researchers at your fingertips. They can help answer questions about internal and external sources, [00:13:00] synthesize research, develop new models, and best of all, help find the questions we haven't even thought to ask and then help get the answers.
[00:13:07] Paul Roetzer: That's a really interesting thing. And honestly, I'm reading through this for the first time. I didn't read through all these before the show. I was thinking about that yesterday, Mike, and I'll just like detour here for a second. I think one of the ways that people are not. Realizing the value of these agents or, or just general is, you know, if you think back to our time together at PR 2020,
[00:13:28] Paul Roetzer: my former agency,
[00:13:30] Paul Roetzer: and the hardest thing for me, I always felt was developing good strategists, like people who could build strategy, analyze business problems, build solutions for clients.
[00:13:42] Paul Roetzer: Like you excelled at that, Mike, your whole career. But the thing that I always found. was an indicator of a great potential strategist. Cause I like to hire people early and like develop these capabilities is people who asked really smart questions, because I feel like to ask smart [00:14:00] questions, you're able to connect dots between things that don't always seem to be connected.
[00:14:05] Paul Roetzer: And so I think going into these models, whether it's Gemini or ChatGPT and saying, what questions should I be asking about how to build a go to market strategy? What questions should I be thinking about when I build my persona? Like leveraging these tools to help
[00:14:22] Paul Roetzer: it grow.
[00:14:23] Paul Roetzer: you figure out what questions you should be asking, because I think by, by asking, learning to ask smart questions, I work with my kids all the time in this, like, Ask, what questions should you be asking of this?
[00:14:35] Paul Roetzer: I don't know. I was just like, I thought, do you use the models at all for that?
[00:14:38] Paul Roetzer: Like, Help
[00:14:39] Paul Roetzer: it figure out what questions to
[00:14:40] Paul Roetzer: ask.
[00:14:41] Mike Kaput: All the time, and I won't spend too much time on it today. Maybe it's a future topic, but there's so many ways to use these as strategy assistants. So, fill in my gaps, my blind spots.
[00:14:51] Mike Kaput: Give me questions that I could ask. a lot of stuff I experiment with is around first principles thinking. Like, help me think clearer about [00:15:00] the fundamental forces driving a potential business or marketing problem. So yeah, absolutely.
[00:15:05] Paul Roetzer: Okay, cool. Yeah. It's just, sometimes I even forget, honestly, to do it. But then when I do like, Oh my gosh, this is so great. So yeah, again, it's like, again, a bit of a sidetrack, but knowing what questions to ask is so critical and that doesn't change with AI, with AI. So in this one, they give 24 examples of companies using these kinds of agents from UPS to Spotify to Mayo Clinic, McLaren Racing, so lots and lots of ideas and inspiration here.
[00:15:35] Paul Roetzer: The next one is code agents. Are develop, helping developers and product teams to design, create, and operate applications faster and better, and to ramp up on new languages and code bases, many organizations are already seeing double digit gains in productivity, leading to faster deployment and cleaner, clearer code.
[00:15:52] Paul Roetzer: Interesting total side note in this one, My daughter is learning how to code in school and I'm not a developer. Like [00:16:00] I've done some coding in my life, but I'm not good at it. And so she was asking for help. I was like, I don't know. So like, I took a screenshot and I put it into a model and I was just like, just what, what is wrong here?
[00:16:09] Paul Roetzer: Not to showing her how to, You know, giving her the answers, but trying to analyze what were we doing correctly that we might need to try different? And it does it like is crazy from a screenshot. It was able to help with code. so code agents, they've got seven examples, including Capgemini, Repl. it, which we love, Wayfair.
[00:16:28] Paul Roetzer: And then the last one is security agents assist security operations by radically increasing the speed of investigations, automating, monitoring, and response for greater vigilance and compliance controls. They can also help guard data and models from cyberattacks, such as Malicious, Prompt, and Jekson. So again, just to make it like tangible, what exactly are they talking about?
[00:16:47] Paul Roetzer: Those are the six ones that they're featuring. Obviously, there can be a bunch more, and what they're enabling, what they want to enable, is using the Vertex AI Agent Builder. They want You and your company to be able to go in and [00:17:00] build things and connect it to data sources that you host in Google cloud.
[00:17:03] Paul Roetzer: Now, I will say went in and tried to use the Vertex AI agent builder just to see, can I build something? And, you know, I keep waiting for like the ChatGPT moment for the no code where you or I have an idea for an app and we just go in and we just build it. Build the app. It, my initial experience is that is not what this is.
[00:17:22] Paul Roetzer: Like you, you still need technical know how. You don't have to be a developer per se, but you definitely need technical know how to use the Vertex AI Agent Builder. It is not, you just go in and give a prompt and you build an app, or an agent in this case. But. Again, I think more than anything, this is a prelude to where we are heading with business and that these agents are going to become more and more a part of what we do, they will impact it.
[00:17:52] Paul Roetzer: Your use of agencies, you know, who, who you're relying on outside. They will impact your internal resources. They will impact your [00:18:00] staffing, your, you know, reskilling, upskilling. Like it's going to have a major impact on everything. I don't think we're at the moment where we should start to really think this is like next month, this is going to be a problem.
[00:18:10] Paul Roetzer: But this is where they're going and they're all working on the same stuff. So I wouldn't be surprised to see this kind of thing from OpenAI in the next couple of months. And as we move into GPT 5 realm, which, you know, could be any day or any month, we're thinking by summer, but I think sooner, this is where we're heading is these things are going to be able to work together and do things.
[00:18:32] Paul Roetzer: And you're going to connect your security, you know, one to your customer agent, and they're going to be able to work together. And it's going to be really interesting how that plays out. ,
[00:18:43] Mike Kaput: our second big topic today is some other Google news, namely Google and the world's biggest.
[00:18:49] Mike Kaput: biggest advertising group are teaming up on generative
[00:18:53] Mike Kaput: AI. WPP has
[00:18:55] Mike Kaput: announced a major collaboration with Google that will see the advertising [00:19:00] group use AI to actually produce ads. According to WPP, quote, As part of the collaboration, Google Cloud's advanced GenAI tools will be used with WPP's proprietary marketing and advertising data.
[00:19:15] Mike Kaput: This will enable WPP's clients to create brand and product specific content using GenAI, to gain deeper insights into their target audiences, to accurately predict and explain content effectiveness, and to optimize campaigns with ongoing adaptive marketing. processes. So this partnership will actually focus on integrating Google's Gemini models with WPP's existing AI powered marketing systems, especially one called WPP Open.
[00:19:45] Mike Kaput: And what this means is that WPP clients like Coca Cola, L'Oreal, and Nestle who use that system already are going to get access to all sorts of AI powered and Gemini powered ad generation and creative [00:20:00] capabilities. Now, for the first phase of this project, there's kind of four big use cases that they're focused on.
[00:20:08] Mike Kaput: First is enhanced creativity. WPP's OpenStudio will integrate Gemini 1. 5 Pro. To help with basic creative tasks, things like writing headlines, turning sketches into images, pairing brand content and guidelines with different types of content, and figuring out actually how the brand sounds and writes using things like Gemini 1.
[00:20:32] Mike Kaput: 5 Pro's 1 million token context
[00:20:35] Mike Kaput: So that ability to have it remember and use tons and tons of information about your brand. The second use case is content optimization. Gemini is going to provide an upgrade to WPP's AI system that predicts how well marketing content will perform. Third is AI narration.
[00:20:54] Mike Kaput: WPP is actually going to use Gemini to automatically create video narration [00:21:00] scripts, then actually send these scripts to a tool called Eleven Labs that we've talked about, to generate synthetic voices for videos. And then fourth is product representation. WPP is going to use generative AI to create product images. So Paul, I guess first I wanted to get your thoughts on how notable is this partnership? How do you see this potentially impacting agencies as a whole and how they're trying to do work for their clients with WPP?
[00:21:30] Mike Kaput: artificial intelligence?
[00:21:32] Paul Roetzer: Well, I was going back, episode 81 is when we talked about Publicis Group and their, you know, 326 investment to revolutionize their ad agency.
[00:21:44] Paul Roetzer: And and, now there's over three years. I think the plan here is, WPP is 320 million per year. Um, it's a race, like it's a race to reinvent ad agencies.
[00:21:59] Paul Roetzer: They're, [00:22:00] they're in a really challenging environment where, the capabilities are accelerating really fast. They obviously need to work with these big, you know, language model, foundation model companies, cloud companies to build next generation services and to try and, keep up and get ahead of where this is going.
[00:22:19] Paul Roetzer: That's obviously extremely competitive between organizations like WPP and Publicis and the big ad groups around the world. I think there's a lot of uncertainty around. intellectual property considerations, know, if they're based in the, in the EU, there, there's considerations around laws and regulations that are ahead of where we are in the US.
[00:22:41] Paul Roetzer: and I think there's just a lot of uncertainty, like something like SORA emerges and the question becomes, well, what isn't the client going to be able to produce themselves? And so and I think there's just a lot of effort on reimagination of the agency model, and the future is uncertain. [00:23:00] Agencies bring strategy and expertise and creative, ability at scale, but I think it's a very real, Question and concern about what won't these frontier models be able do?
[00:23:12] Paul Roetzer: So like, if I'm working with a big ad agency and I want to create a campaign, and I don't know the answer to this. This is a, this is like a bit of a rhetorical question. Like, when GPT 5 emerges and Sora is built into it or Gemini 1. 5 or Gemini 2, and like all of these multimodal capabilities are built into this for images and text and audio, like I can create songs and music, like I can create anything I would need to for a creative campaign,
[00:23:38] Paul Roetzer: . Hmm. what won't I be able to do?
[00:23:40] Paul Roetzer: To just do with one of those models that I would usually pay a big agency for, is a question I have. Like, and so if you imagine Like, let's say I wanted to create an ad for MAICON, I'm thinking out loud here, and I go find like the 10 best ads I've ever seen, like these are just my absolute favorite ads, and I give those [00:24:00] ads to
[00:24:01] Paul Roetzer: Gemini or ChatGPT and say, I want to build an ad campaign, images, video, I want to create music, like everything, soundtracks, based on, like, inspired by these campaigns.
[00:24:13] Paul Roetzer: These are the ones I love. These are the best performing campaigns in history, whatever it may be. Like, let's go. Let's, let's build some images together. Let's build some taglines. Let's build, like, what wouldn't one of those foundation models be able to do? And so I'm not saying like people aren't going to want to work with big ad agencies.
[00:24:29] Paul Roetzer: There's obvious benefits to doing that. And there's going to be a lot beyond just creative that the agencies bring to the table, but I think this is a real consideration for. Agencies, and I do think things like Sora scare them. Like, they look at it and think, holy cow, now they can use those tools as well, but so can you and I for 20 bucks a month.
[00:24:53] Paul Roetzer: So, yeah, I think it's just a, it's a really interesting environment right now where the agencies aren't just competing with each other anymore, [00:25:00] they're now competing with 20 a month technology that, does things that used to require 30 person teams, like at all these different disciplines in creative, from strategy to, you know, design to video production to audio production.
[00:25:15] Paul Roetzer: I mean, we're talking about. Lots of different expertise is brought together in different roles across creative teams that you have single foundation models that can do all of it now. and again, I am not saying better than humans or in place of humans. I'm just asking questions out loud here of like things that I don't have the answers to.
[00:25:36] Paul Roetzer: and I think it just keeps coming back to this idea of like, do we need as many humans doing creative when the AI themselves are really, really Good at creative. and whether I'm still hiring an agency, does the agency need as many people? Like if, if these foundation models are being infused into what they do and they're doing deals with Google and all, they just build their own little agents.
[00:25:56] Paul Roetzer: They build creative agents specific to WPP [00:26:00] knowledge or specific to clients. Like it's a, it's just a really uncertain future here for creativity and workforces, I guess, and agencies.
[00:26:11] Mike Kaput: So it doesn't sound like you're the only one asking these questions. I mean, we had surfaced in some of our research here that it sounds like, you know, WPP is getting asked about things like what is SORA's impact on your business when they're on earnings calls?
[00:26:26] Mike Kaput: Can you talk to us a little bit more about that and about how they're getting, uh, asked about what, how these tools impact the actual business model and what that means for them?
[00:26:35] Paul Roetzer: Yeah, so, you know, again, I think last week with Accenture, we dove into their earnings transcript because you can learn a lot from these earnings calls. and what we, what you learn from WPPs is. Things aren't great. Like they're like, this, this sounds awesome. Do a deal with Google. It sounds great. But then when you dig into the surface, you realize like they're not growing.
[00:26:56] Paul Roetzer: So their WP in, this is a [00:27:00] marketing dive article we'll include in, but they were doing an analysis of Q4 earnings. and WPP is, as it says, is struggling in the evolution. WPP is placing serious investments behind proprietary AI solutions to keep pace with rivals like Publicis that have fared better financially.
[00:27:15] Paul Roetzer: WPP reiterated that it foresees organic revenue growth between 0 and 1 percent in 2024. Given the world of
[00:27:25] Paul Roetzer: Growth potential, like we saw Accenture accelerating 600 million in revenue in one quarter from General AI. It's like, you would think that there'd be some growth here. it said looking ahead, WPP is placing serious investments.
[00:27:37] Paul Roetzer: So we talked about that one. said it would commit 318 million, well, 250 million pounds, about 318 million U. S. dollars annually to drive AI transformation. So they're racing to try and like figure this out. AI could help boost WPP's performance through licensing, assisting with client projects, and improving internal operations, executives have argued.
[00:27:57] Paul Roetzer: and then it did say another potential [00:28:00] problem for agencies is if other offerings outpace their own capabilities with generative AI. WPP released its earnings around previews for OpenAI's Sora tool that can quickly produce complex videos based on text prompts and investors asked about the potential impact.
[00:28:15] Paul Roetzer: They said, I don't think it changes our strategy. I think it reinforces what we are doing. What clients need is work that's copyright proof that, which I'm not really sure how what they're doing solves that. But anyway, that is able to accurately represent their brands and reality. And Sora is not yet at that stage.
[00:28:32] Paul Roetzer: That's a dangerous. That's a dangerous response. Not yet at that stage. You've never seen it. You've never used it other than seeing it on Instagram. You don't know what it's capable of. So I hope that that's not actually like. They don't think that that's a sustainable thing because SOAR is not yet at that stage.
[00:28:51] Paul Roetzer: That could be like three months. That statement could be false. you know, and interesting, like, we keep coming back to these, like, copyright issues. And I [00:29:00] was interested, like, what is WPP's stance on this? And so I did some research on, do they have generative AI policies? Are they talking about copyright?
[00:29:07] Paul Roetzer: To their credit, like, they do have some content about this, but it's really hard to find. Like, I couldn't find there any actual AI policy publicly facing. They may have it. been thinking about and doing AI stuff for a long time. It's not like they were asleep at the wheel prior to 2022. but it's really hard to find.
[00:29:25] Paul Roetzer: But they, I couldn't find any gender of AI policies. They referenced them in like, I, there's an article we'll put in the show notes from 2023, it looks like March, 2023, on AI as a new phase of innovation. And they said that WPP ongoing robust training has been in place since 2019 to ensure users of these tools think about the use of personal data, data, privacy laws, and confidentiality.
[00:29:50] Paul Roetzer: goes in to talk about policing themselves, and we have created our own set of guardrails, but again, I can't find them. We have our own policies for working with creative [00:30:00] AI, and we have a security and privacy charter. and then, keep in mind, I believe they're based in England maybe? I think they're based in the EU at least.
[00:30:09] Paul Roetzer: So they're, they have more, you know, stricter guidelines than we do
[00:30:12] Paul Roetzer: the U. S. and then it said, let's not forget that General AI tools that are making the headlines are trained on all the content in the world. There's a big question about content curation. It's a challenge. Increasingly, you will see managed data sets for training purposes, particularly in relation to copyright material.
[00:30:26] Paul Roetzer: So they're definitely like Aware and working on this and building policies and turning around it. What that means to these kinds of partnerships with Google, where it's making it more readily available to create stuff, I don't, I don't know. And I couldn't really find anything about it.
[00:30:39] Mike Kaput: All right. So for third topic this
[00:30:44] Paul Roetzer: Speaking of copyright
[00:30:45] Mike Kaput: issues.
[00:30:45] Mike Kaput: topic, yes, speaking
[00:30:47] Mike Kaput: of copyright issues, there are several hot topics. AI startups that are getting a bunch of attention this week. And not all of it is what we would call good attention.
[00:30:58] Mike Kaput: So Paul, in this segment, [00:31:00] we're going to do things slightly differently and actually tackle each of three startups that are getting a ton of buzz one at a time and then unpacking why they're getting all sorts of attention and what that means for business leaders following this space. So first up are the early reviews of Humane's AI pin,
[00:31:21] Mike Kaput: company called
[00:31:21] Mike Kaput: Humane that makes a product called AI Pin, and the first reviews of this product are quite frankly, brutal.
[00:31:29] Mike Kaput: The Humane AI pin is intended to be a small, think of a tab like wearable that attaches to your front, like a lapel or a strap on your backpack. And acts as a voice activated AI assistant designed to kind of replace your phone on the go. Now, the promise of the pen is something the company calls, quote, ambient computing.
[00:31:52] Mike Kaput: So this ability to access multimodal AI that performs a range of tasks for you in the real world. [00:32:00] Unfortunately, the device does not seem to be anywhere near that promise. The Verge wrote a very long, in depth review, but did not mince words in that review when summing it up. They said, quote, should you buy this thing?
[00:32:13] Mike Kaput: That one's easy. Nope. Nuh uh. No way. The AI PIN is an interesting idea that is so thoroughly unfinished and so totally broken in so many unacceptable ways that I can't think of anyone to whom I'd recommend spending the 699 for the device and the 24 monthly subscription. So, despite being kind of long term bullish on this idea of wearable computing, And devices like the iAPN, the Verge found that the device is extremely slow processing even basic commands.
[00:32:45] Mike Kaput: It's missing basic features, and half the time it just did not work as intended. I mean, for instance, it can't even do things like put things on your calendar if you ask it to, at least at this stage.
[00:32:59] Mike Kaput: Now, [00:33:00] despite a really rough rollout, Humane has some pretty significant backing. According to CB Insight, some research I did before the podcast today, the company has raised 242 million over six funding rounds, Altman, among others, as an early investor.
[00:33:17] Mike Kaput: So Paul, first up with the Humane AI pin, you seemed somewhat skeptical of the overall premise behind the pin when we talked about it on episode 72 when they kind of pre launched it. What did you think reading these early reviews?
[00:33:31] Paul Roetzer: Listen, I don't want to, like, I don't want to badger them worse than they're already getting it. I said on episode 72, I just didn't think this was a viable concept. and I also, I think at the time said, I love entrepreneurship. I love a willingness to take risks. I like that they tried this, but, my initial feeling was it just wasn't a [00:34:00] form factor that there would be demand for, and that was assuming it even worked.
[00:34:06] Paul Roetzer: The reviews basically say, to save you time of having to go look at them, it's extremely slow. Like, takes, you say, I saw a picture where, Marcus, Brownlee did a review and he's like, just, he said, what is, what am
[00:34:19] Paul Roetzer: I looking at?
[00:34:19] Paul Roetzer: And it basically was a, it was a Cybertruck. And it took like 12 seconds to come back and say it was a Tesla Cybertruck.
[00:34:25] Paul Roetzer: Meanwhile, he went on his phone, searched it up and had images and videos of the Cybertruck. So, it's really, really slow. It overheats. overheats all the time.
[00:34:33] Paul Roetzer: And then it just says, like, I can't do this because of, you know, it's overheated. You can give me five minutes. Basically, it costs 700 and you have to have a separate 24 a month plan, as you said, with T Mobile, which also requires a different phone number.
[00:34:46] Paul Roetzer: So it's not even like it's attached to your, your cell plan. It's not apparently better at the phone than anything. Like there's no. Use case where this is actually better than your phone. I [00:35:00] get that they raised 240 million, but that's nothing for a hardware company. Like, again, we talked about inflection a few ago is basically tanked and they had 1.
[00:35:08] Paul Roetzer: 5 billion, like 240 million is six rounds is not a lot. it's like borderline. Insulting to your customers to release something like this as a V1. Again, I get the whole Silicon Valley mentality of move fast, put stuff out, ship, ship, ship, like just get things into the market, but this is borderline irresponsible to put out a product like this and you just don't.
[00:35:34] Paul Roetzer: usually recover from launches that are this bad. you know, you go back to the Vision Pro, there was people who didn't like the Apple Vision Pro and yes, the price tag was crazy, but it was an insane piece of technology that has all of these incredible advancements. This is not. That and the Mark has Brownlee reviews.
[00:35:56] Paul Roetzer: So if you're not familiar with him, he has 18. 6 million subscribers on [00:36:00] YouTube. The guy does like insane, tech reviews, like awesome stuff. he, the, as of this morning, his YouTube video has 1. 9 million views. And the headline is the worst product I've ever reviewed for now.
[00:36:14] Paul Roetzer: So again, I don't want to
[00:36:16] Paul Roetzer: like, you know, make this worse for them than it is.
[00:36:20] Paul Roetzer: Their executives were all on Twitter over the weekend saying like, Hey, basically give us a chance. This is really hard. We've worked really hard at this. And I don't doubt that. And again, not to take anything away from what that team has tried to accomplish. I just don't. Know how you come back from something like this.
[00:36:37] Paul Roetzer: I don't know how you get more funding. Um, maybe they've got some IP, some patents and some people that are worth rolling into a, another tech company. Like what Sam's trying to do with, is it Johnny from
[00:36:51] Paul Roetzer: Ivy? Is it from Apple? Like
[00:36:53] Paul Roetzer: maybe there's some saving face play to do something with this tech, but as a standalone product that [00:37:00] is going to make an impact on the world, it does not seem to be heading in that direction.
[00:37:05] Mike Kaput: So another tool or app getting a lot of play is something called UDO. in a previous episode, we had talked about. Some of the stunning AI music generators coming out, one of which is Sunoo AI, that takes, generates complete songs from scratch using AI, and UDIO does the same thing, and it's being seen as a
[00:37:26] Mike Kaput: serious
[00:37:26] Mike Kaput: competitor to a tool like Sunoo, because it has the ability to generate Catchy, realistic sounding music and singing in any style.
[00:37:36] Mike Kaput: What's notable about UDIO is that it's founded by four former researchers at Google DeepMind. They have 10 million in total seed funding from VC firms that include Andreessen Horowitz and also some. Prominent, individual investors like Instagram's co founder and some musicians like Will. i. am. And I listened to some sample songs created with [00:38:00] Udio and was pretty blown away.
[00:38:01] Mike Kaput: It's currently under a wait list to use it. I think they're just flooded with people trying to sign up. I would say it's certainly sounds on par. With Suno, but just like Suno, there are still some big unanswered questions here about how these tools are able to
[00:38:18] Mike Kaput: such realistic sounding music.
[00:38:21] Mike Kaput: So, Rolling Stone says, quote, Though neither company will directly confirm or deny it, There is substantial reason to believe that both UDIO and SUNO were trained on copyrighted music without permission. So Paul, what are your kind of initial thoughts on some of these well funded music generator startups like SUNO, like UDIO?
[00:38:43] Mike Kaput: Does anything jump out at you about kind of their business models, impact, copyright concerns?
[00:38:48] Paul Roetzer: I will say I was able to get in and get an account Sunday, so I was able to
[00:38:52] Paul Roetzer: go and I was showing my daughter last night, how how, how that one works. So, Yeah. I mean, it's impressive. [00:39:00] My, my initial feeling, like they have an interesting, founding team, certainly, certainly top investors. I don't understand what the defensible mode is
[00:39:11] Paul Roetzer: Pierre, like it, this is going to be copied by everybody. Like my guess is all the research labs already have these capabilities or could quickly do them if they're also willing to take copyrighted songs and train models on it if they haven't already. So it doesn't seem like there's any. Real technological breakthrough that's going to like keep them differentiated that Suno or somebody else isn't going to be able to replicate and do.
[00:39:37] Paul Roetzer: the way these labs work is like, once somebody has done something, it's copied, you know, really fast. And people move around all the time between labs and bring that knowledge somewhere else. So cool tech for sure. I think it's going to be everywhere, not just from Woodio, but like literally everywhere is going to have this capability.
[00:39:56] Paul Roetzer: And then just like a background, not only on Oudio, but [00:40:00] also the next one we're going to talk about, just so people, cause Mike and I get pitched all the time to like review products. And I think I've mentioned before, we've thought about playing in this world of like tech tools and reviews, and we'll do more of it.
[00:40:12] Paul Roetzer: We, we, you know, we're going to do a lot of it, but, I think just like,
[00:40:16] Paul Roetzer: um, sometimes people just jump to like the hot new tool and they don't really think about. Like why it is potentially relevant or if people want to pitch Mike and I on products, just to give you a sense of like how we think about these and how we filter them and decide which ones are worth bringing to you.
[00:40:32] Paul Roetzer: So there's a lot that we look at, but just a few of the variables we consider. One is the founding team. So in this case, like Google DeepMind people, I'm going to pay attention every time. Like if people have left one of the major research labs, the founded thing, it gets our attention regardless of whether they have funding or not.
[00:40:46] Paul Roetzer: So we look at the founding team. We look at how much funding they have and what series of funding they're in. We look at the investor list, who are the people behind these companies. And if they're, you know, major players, then again, that captures our attention. In this [00:41:00] case, it was, um, there was a, not only were the founders impressive, but the investors, you had the co founder and CTO of Instagram and the head of Gemini at Google were two of the investors in this company.
[00:41:11] Paul Roetzer: So that catches our attention. Product market fit, you know, it's just obvious from the get go, this is something that's going to hit the market, which I would not. Say, the AI pin would fit into, personal experience with the tech. So if we've tried it and then just like buzz, chatter in our community of influencers, the people we consider sort of tastemakers in AI.
[00:41:30] Paul Roetzer: So, you know, from Oudio to the next one we talk about and all the ones we'll talk about in the future, just so people have a little context as to how we decide which AI tools are worth talking about.
[00:41:40] Mike Kaput: So on that note, this third AI startup in the spotlight today is a tool called Captions.
[00:41:46] Mike Kaput: This is a AI powered creative studio that initially gained a bunch of massive popularity as an iOS app, and now it's rolled out capabilities for web and desktop users. So the Captions AI powered creative [00:42:00] studio has a bunch of features to help you basically edit video with the power of AI. So you can do things like edit AI dubbing a video into 28 different languages while preserving the user's voice.
[00:42:12] Mike Kaput: AI eye contact, which corrects eye contact in just one click. AI shorts, which extract short based on
[00:42:20] Mike Kaput: viral potential from long form videos. And a feature called AI denoise, which removes background noise automatically. Now our team, Claire and our team, has been impressed with testing out tool
[00:42:32] Mike Kaput: and some initial use cases as we're kind of looking into using it for podcast, video creation, event video footage.
[00:42:41] Mike Kaput: According to CB Insights, Captions has 37. 5 million in funding from VC firms that include Kleiner Perkins, Andreessen Horowitz, and Sequoia Capital. Now Paul, what were your initial thoughts on captions? I know we've been kind of exploring whether or not this is a replacement for [00:43:00] something like a descript, a complement to it.
[00:43:03] Paul Roetzer: I think this one's a good example of how You can think about the impact of AI at innovation within your own company. So the way this plays out, I see this getting, you know, a lot of love last like Tuesday or so from, again, like the tastemakers within our community, people that I follow closely, starts kind of blowing up.
[00:43:22] Paul Roetzer: I go check it out. Looks really impressive. They've got AI scriptwriter, AI avatars, voiceovers and voice cloning, AI trimming, AI enhanced speech, AI eye contact. They have like 12 to 15 AI features, some of which to me seem complimentary or redundant to what we already have. So I send it to Claire and our team was like, Hey, can you take a look at this?
[00:43:42] Paul Roetzer: because Claire's the one that does the production of the podcast and the webinars and some of our other content. So she's the one that plays around most with this, is most adept, especially on the video side of things. So Claire goes through it, builds brief. comes back to me like 24 hours later with this great analysis, like three to five [00:44:00] pager.
[00:44:00] Paul Roetzer: Here's ways we could use it. Screenshots of how did it. and I go back and say, okay, cool. Like, how does this affect the script? And there's another tool. I think Opus Clips is another one we use to do clips. and basically in this very dynamic way, like, okay, go, like, if you think this is a better tool, go use it.
[00:44:17] Paul Roetzer: And so I think this is the kind of. way that companies need to approach when you find a tool that, again, go back to our criteria, impressive founding team, major investors, product market fit. They have 3 million users of the app. Um, we know this is legit technology. This one's worth it. You know, spending some brain cycles on and some potential resources.
[00:44:39] Paul Roetzer: And for us, we work in this world. It's like, okay, what's it cost? 50 bucks a month, 20 bucks? I don't, I don't actually don't know, but whatever it is, okay, clear. Is there, can you justify the cost? Is there going to be time saved, increasing creativity? Are we going to be able to do new things? Yes, go. This is, that's how decisions are made.
[00:44:55] Paul Roetzer: Now, again, we're small, we can be nimble like that. But I think that's what you [00:45:00] have to do. You have to have a process to stay up on the latest technology that is actually legitimate. Don't get caught up in the next shiny thing that is just a distraction. But if you have criteria to assess these tech, and again, I share with you like six of our variables, but there's others.
[00:45:15] Paul Roetzer: Once you have that criteria, you can look at a tech and say, okay, this fits our model of something we need to pay attention to. Let's have someone on our team who can go explore this. Ideally, it's the domain experts, the person most capable of assessing that technology and then, you know, here, here you are five days later, and we can be integrating that into the podcast this week.
[00:45:34] Paul Roetzer: and so that's how this stuff works. And I think that's how a lot of organizations need to get structured to make it work.
[00:45:39] Mike Kaput: Yeah, and just a final note there, I realize everyone has different budgets, but like, we're not breaking the bank doing this. If you can't spend 20 bucks to test something out for a month, you are thinking about this the absolute wrong way. And it really, you really need to maybe Take a step back and think through that, because you cannot even [00:46:00] think about the cost of testing out a tool that's 20, 30 bucks a month.
[00:46:03] Paul Roetzer: Yeah, and I think this is a good point, Mike. And maybe another way to think about it is like, we're doing this kind of one off, more dynamic. If you're in a bigger environment where you do need some more guardrails or you have more processes, then try and get that in advance. Say, hey, as part of our AI council or AI innovation team, or like whatever you're calling it.
[00:46:20] Paul Roetzer: we would like to have the freedom to spend up to 500 A month, a quarter to R& D technology.
[00:46:28] Paul Roetzer: We want to be able to play around with things. We will have a benchmark of current performance. We'll have a goal of what it's going to do. We're going to test it over a 90 day period or a 30 day period. Like here's the system and get approval on that overall approach to testing AI tech versus, Hey, can I go try this one?
[00:46:43] Paul Roetzer: Can I go try this one? Can I get a 20 approval for that one? Or I'm just going to put it on my personal credit card and try it.
[00:46:48] Paul Roetzer: It's get a system that
[00:46:51] Paul Roetzer: That encourages. Innovation. and I think to your point, Mike, like if you go to a leadership and say, listen, here's three tools we've already [00:47:00] found that can make us more efficient, more creative, more productive.
[00:47:02] Paul Roetzer: We would love to keep experimenting with this. Mike and Claire are going to lead our internal effort on this. Here's how they're going to do it. Here's the framework we'll use. Can we get approval to ad hoc spend what we need to spend, to try this stuff?
[00:47:16] Mike Kaput: Absolutely. All right, let's dive into some rapid fire topics
[00:47:21] Mike Kaput: So related to kind of some of the talk we've had around agencies, Um, brands are appearing to actually start cracking down in some cases on agency usage of AI, according to some new reporting from AdAge. So that includes some brands demanding stronger AI safeguards and contracts with ad agencies, and in some cases, even restricting any AI usage without prior authorization.
[00:47:47] Mike Kaput: According to one ad agency CEO that AdAge interviewed, quote, recently we won three new pieces of business and in the master service agreement, it says you're not allowed to use AI of any kind without [00:48:00] prior authorization. So brands that AdAge talked to seem to be rightly worried about all the ways AI can go wrong if it's used by a third party partner.
[00:48:09] Mike Kaput: So, Being concerned about how proprietary data is used and the ethics around using AI in certain ways. For instance, some brands worry that training models on their data can then inform outputs used by other competitors using the same types of AI systems and models. Or some are starting to worry that AI impacts the creative work.
[00:48:32] Mike Kaput: being done by humans in the ad industry. So an example of this that Ad Age mentioned is there was recently controversial
[00:48:39] Mike Kaput: Under Armour ad that came out this past month. It used generative AI trained on the company's brand assets to generate new visuals and a voice over of boxer Anthony Joshua. And it did all this without actually shooting any new footage, which had some creatives a little upset.
[00:48:58] Mike Kaput: So paul, you have [00:49:00] extensive background in the agency world, you know, former agency owner, like how should agencies be thinking about or solving for these concerns?
[00:49:08] Paul Roetzer: Yeah, we've talked about this on a few recent episodes, but I think first and foremost, you just have to get the attorneys in the room. Like this is a, we need to just kind of reset how contracts and agreements are structured between clients and agencies and freelancers, like just outside service providers and do it once, get on the same page and then have a regular review cycle.
[00:49:31] Paul Roetzer: Like you can't. And so, you know, I'm not going to be scrambling to figure this out with each client and a client finally asks you about it and you don't have an answer. So, like, for example, and this kind of aligns with some of
[00:49:41] Paul Roetzer: the you were just talking about, Mike, Clients and agents are going to have non disclosure agreements that's going to cover confidential proprietary information.
[00:49:49] Paul Roetzer: So the question becomes, like, as one sample where this could go wrong, If I take client data and I put it in to train a model or to build a GPT or to help [00:50:00] with creative, have I breached that non disclosure agreement by putting it into that model? If I don't know if the model will be trained, use it for training data for a future version.
[00:50:10] Paul Roetzer: Um, that's one possible legal concern. The other, kind of more on a policy side is just to make sure up front that the agency and the client are aligned on their generative AI policies. So if you have a master service agreement saying you're not allowed to use it, but the internal policy tells the account team you're allowed to use it for ideation and whatever, you have to make sure that that trickles down.
[00:50:33] Paul Roetzer: So when a master service agreement is in place, that the account team needs to be aware from the top people down to the interns. That actually our agreement is we're not allowed to follow our own generative AI policies. So it's just, I think it's just messy right now. There's not a lot of clarity.
[00:50:48] Paul Roetzer: There's not a lot of like standardization around these things. And I think for agencies and for, for brands, the client side, it's just really important to get this stuff straight and [00:51:00] get it right. I don't know that I would be signing master service grants where I'm not allowed to use it at all. That
[00:51:05] Paul Roetzer: seems very restrictive and not ideal.
[00:51:10] Paul Roetzer: So, yeah, I mean, then it gets into, you know, right clients working with the right agencies that trust each other. So I say, again, it just further, shows just the the, uncertainty of the agency world right now. There's just so many implications, again, so many opportunities, but there's just a lot of operations and legal stuff that needs to be happening behind the scenes. for
[00:51:35] Mike Kaput: we just got two new pieces of legislation that are being proposed that could have an impact on U.
[00:51:40] Mike Kaput: S. citizens relationship with AI. So the first is a proposed bill being dubbed
[00:51:46] Mike Kaput: American
[00:51:46] Mike Kaput: Privacy Rights Act. This is a proposed bipartisan bill that would, in effect, establish a
[00:51:53] Mike Kaput: right to privacy law. for American citizens, which we do not have at the moment. We only state [00:52:00] privacy laws, which this act would actually supersede if it passes.
[00:52:05] Mike Kaput: So according to some reporting on the discussion draft of the bill by CNN, it would ban the transfer of an American citizen's personal data to third parties unless they provide explicit approval or if the data is for one of the specific. purposes that are allowed and called out in this bill, stuff like fraud prevention.
[00:52:26] Mike Kaput: Interestingly, it would also let users opt out of targeted advertising altogether and require companies to collect only enough data as they need to do their business. So this is still in draft form, and when it's introduced, it must clear its committee. And also passed both chambers of Congress. Now, a second piece of legislation getting attention is called the Generative AI Copyright Act.
[00:52:52] Mike Kaput: This act was introduced in the U. S. Congress this past week and is designed to force companies to share what copyrighted [00:53:00] material, if any, that they use to make copyrights. their generative AI models and products. The bill says it does this by forcing companies to submit a notice that details how they used copyrighted works in their training data to the register of copyrights before they release or update a new generative AI system.
[00:53:19] Mike Kaput: So this notice details how detailing how copyright works are used must be submitted at least three times. 30 days before somebody publicly releases AI tools or the companies can face a financial penalty. Now,
[00:53:34] Mike Kaput: I read in the bill, this is detailed as quote, a civil penalty in an amount, not less than 5, 000.
[00:53:41] Mike Kaput: So it's not clear exactly like what cap they might go up to, because that's not quite that much for people like opening.
[00:53:48] Paul Roetzer: anybody.
[00:53:49] Mike Kaput: You're right. So with the caveat here, Paul, that neither of these laws have yet passed, like. What do you make of each one's goals? Like, what does this tell us about what [00:54:00] lawmakers might be thinking about here? I
[00:54:02] Paul Roetzer: think it just shows that there is stuff happening behind the scenes that we're not hearing about every day, that the people in Congress are certainly aware of a lot of these issues. we've said before that at some point this stuff is going to get politicized. There's going be, Battle lines drawn on, on each side.
[00:54:20] Paul Roetzer: You know, someone's going to to start a fight on this stuff, and it's likely not going to be bipartisan. Once we know which side the public is on, you know, I think that's, and I still don't necessarily feel like we're at that point going into the election cycle in 2024 in the US. Like, I don't know that it's a clear bipartisan issue on either side, but I think progress would be made.
[00:54:41] Paul Roetzer: And I just made a note to myself, like I keep saying one of the safest jobs of the next decade are IP attorneys that work in generative AI. I just realized like, Lobbyists that work in generative AI.
[00:54:51] Paul Roetzer: I even imagine the amount of money being spent right now on lobbying efforts around this kind of stuff in Washington. So, yeah, [00:55:00] definitely worth paying attention to, but as you said, just they're early stages both. nothing definitive, but progress is definitely being made, it seems.
[00:55:09] Mike Kaput: So, hot on the heels of our segment last week about Demis Hassabis, comes some more news about him and his relationship with Google. we saw a new report by The Information titled, Google's Demis Hassabis Chafes Under New AI Push, and it details the obstacles and the challenges that Hassabis is facing internally at Google.
[00:55:31] Mike Kaput: So these include him dealing with a decrease in independence within the firm since his DeepMind team merged with Google Brain. That created a single overarching AI unit within the company. It sounds like he's under pressure to show commercial results from AI research, and he's having some difficulty sometimes preventing open AI from poaching Google researchers.
[00:55:54] Mike Kaput: Now, This comes after we reported last week on a series of articles that appeared to be [00:56:00] intentionally raising Demis profile within Google, kind of by casting him as one of
[00:56:04] Mike Kaput: key individuals helping the firm win aI arms race. Now, Paul, in the light of that, like, do you think this is Demis doing this, like, in order to get leverage internally at Google to solve some of these challenges?
[00:56:18] Mike Kaput: Is this Google spearheading these stories? What's going on here, given this new article?
[00:56:22] Paul Roetzer: I can't imagine it's Google, like I, it just, just makes no sense. I, obviously I don't know, but as I said previously, it just seems strategic. Like they're, Shane Legg I've also noticed has been out, the other co founder of DeepMind, he's, been doing, Quite a number of interviews lately as well. Now part of it might just be raising the profile of Gemini, but you know, I think the article just does a good job of highlighting.
[00:56:45] Paul Roetzer: It's not like you can just mash two research labs together that historically were kind of abrasive to each other. So again, Google Brain was one. DeepMind was acquired in 2014. They functioned very independently from each other and in some [00:57:00] ways kind of battled with each other for resources. And they were forced to come together last year to to really try and bring a single frontier model which became Gemini 1.
[00:57:11] Paul Roetzer: 5 as we have it today. Um, combine the best of their internal talent, their, their computing power, their models. They're trying to mash it all into one thing. And that's, it's a hard thing to do, especially when DeepMind was always much more of an independent research arm that was working on the next frontiers.
[00:57:30] Paul Roetzer: All their, from day one, their plan has been to build the first AGI. It's literally on the cover of their business plan. Build the first AGI. and now you got to build a language model. Do Gemini in workspace with like, it's not what they were there for. And so it's kind of natural that there would be some challenges.
[00:57:50] Paul Roetzer: We obviously have no idea whether, you know, it's going to work out internally or not. I will just say, Demis and his team are absolutely critical to that organization. [00:58:00] If there were ever an instance where Demis did actually leave, and this article said he considered leaving to build his own lab again, and that's not the first time we've heard that.
[00:58:08] Paul Roetzer: If he were to actually leave, that would be catastrophic to Google, in my opinion. So, I hope it all works out. Love, I I have always admired Demis, and his DeepMind team and what they're doing. And I really hope it works out and they can make the breakthroughs they've sought to make.,
[00:58:27] Mike Kaput: So
[00:58:28] Mike Kaput: in some other Google news, Google actually just released a 40 plus page prompting 101 guide that is packed with really useful advice on how to prompt Google Gemini using Google Workspace apps and by directly interacting with it at gemini.
[00:58:44] Mike Kaput: google. com. Now the guide I would highly recommend anyone check out. It's very readable, it's got a lot of overall prompting advice, and a ton of example use cases prompts
[00:58:54] Mike Kaput: by role. Bye. Now, some of the advice in this guide may actually be a little counterintuitive if [00:59:00] all you're kind of following is those prompting gurus out there who are selling specific prompts and specific frameworks for doing things.
[00:59:08] Mike Kaput: Because Google really just kind of advises us to keep it simple. They say there are four main areas to consider when writing an prompt and that
[00:59:17] Mike Kaput: you don't even need to use all four if you don't want. The four areas are give it a persona, give it a task, give it context, and tell it what type of format you want.
[00:59:28] Mike Kaput: Now, Google also says, quote, prompting is an art, and based on what we have learned during our Workspace Labs program, the most successful prompts average around 21 words, and that people who don't know this are using even fewer words on average, about 9 words per prompt. So, as part of this guide, Google doesn't just offer that overall advice, but it also But gives a bunch of sample use cases
[00:59:53] Mike Kaput: prompts by role. Now this is the majority of the guide and worth
[00:59:57] Mike Kaput: into
[00:59:58] Mike Kaput: because they give you [01:00:00] things like here's how CMOs should consider prompting AI tools for different use cases. So things like prompting Gemini right within Docs to generate new data. blog post ideas based on your meeting notes with, say, your social media team.
[01:00:14] Mike Kaput: Or, if you're a brand manager,
[01:00:17] Mike Kaput: using
[01:00:18] Mike Kaput: Gemini do, market research. So you can say, I need to do market research on a certain industry to identify new trends. Use these URLs to uncover emerging trends and shifting consumer preferences.
[01:00:30] Mike Kaput: So,
[01:00:31] Mike Kaput: all in all, highly recommend checking out.
[01:00:33] Mike Kaput: um, Paul, this certainly seems to kind of confirm the perspective you've had and talked about many times, which is look, prompt engineering, or at least putting thought your
[01:00:42] Mike Kaput: prompts is important, but it's likely to not be something long term you really, really, really need to get good at as these models just get better at understanding your inputs.
[01:00:53] Mike Kaput: Do you kind of still believe that after seeing what's in this guide?
[01:00:56] Paul Roetzer: Yeah, I do. And I think, again, it just keeps [01:01:00] going back to just talk to it like a person. Talk to it like you're giving instructions to an associate, an intern, whatever it may be. And like we said earlier, ask it questions like, what else does it need to complete this task? And this is where I think that, you know, we're not going to, you know, get rid of the need for humans to be experts and have experience and have points of view and have instinct and intuition.
[01:01:24] Paul Roetzer: Like, all of those things make you good at talking to these AI.
[01:01:29] Mike Kaput: Mm .
[01:01:30] Paul Roetzer: And so, yeah, and I think, again, it just further um, shows that there is no one right answer. A lot of times when it comes to AI, you can go read the Anthropic Cloud prompts. You can read the OpenAI prompts. You can read the Google prompts.
[01:01:44] Paul Roetzer: This stuff is constantly evolving.
[01:01:47] Paul Roetzer: No one knows
[01:01:48] Paul Roetzer: all of the answers. And a lot of times it's just educated guidance because someone has a little more experience at it than you. But you can get in and become an expert in Gemini in, you know, a couple of weeks. Like if you're, if [01:02:00] you're just trying a bunch of stuff and that's an exciting thing to me is, you know, get in there, read this thing and start playing around with it and become the person on your team who actually knows how to work with Gemini, that might be really valuable.
[01:02:11] Paul Roetzer: So yeah, I don't know. It's just, I like these, these guides. I think they're, when they come from the people that are building the tools, it's very helpful to get any level of guidance from them of how to work with their systems.
[01:02:23] Paul Roetzer: very
[01:02:27] Mike Kaput: Tiktok is actually considering using AI generatedinfluencers that could compete with human creators for ad deals.
[01:02:32] Mike Kaput: According to some reporting in the information, this feature is still in development, but it would consist of generating AI influencers to perform in a video based on a prompt submitted by an advertiser. While the AI Influencer feature has been in
[01:02:48] Mike Kaput: for several months,
[01:02:49] Mike Kaput: The information reports that TikTok staffers say it is far from ready, and testing shows that AI influencers, at least right now, generate far fewer e commerce [01:03:00] transactions than humans.
[01:03:02] Mike Kaput: Now, Paul, it sounds like we're probably a little far off from getting fully AI influencers in this context on TikTok, but it sounds like the potential could be there. Do you see this as a potential? Marketing strategy or trend that's likely to catch on and stick around.
[01:03:18] Paul Roetzer: I do, I don't know if that's good or bad,
[01:03:20] Paul Roetzer: but , I mean,
[01:03:21] Paul Roetzer: you and I wrote about, I can't remember the influencer's
[01:03:24] Mike Kaput: Yeah, it's
[01:03:24] Mike Kaput: A little McKayla in the book.
[01:03:25] Paul Roetzer: go, who had millions of followers on Instagram and had her own agents, like,
[01:03:31] Paul Roetzer: they, yeah.
[01:03:32] Paul Roetzer: so, yeah, I think not only, like, AI influencers built from nothing, but I think you will, if we aren't already there, um, will have celebrities and existing influencers who just create deepfakes of themselves and do deals with that thing.
[01:03:48] Paul Roetzer: Like, hey, you can get me for 50, 000 for a 30 second ad, or you can get my AI version of
[01:03:54] Paul Roetzer: me for 20, 000,
[01:03:56] Paul Roetzer: and you can have it say anything. I'll, I'll approve [01:04:00] everything. And people are just going to make. Crazy money creating like deepfakes of themselves and then licensing those deepfakes. so yeah, it's going to be a weird world when it comes to the influencer stuff.
[01:04:11] Mike Kaput: Yeah, and I would just say that There's plenty of things to critique here. And who knows if this will be effective in terms of influencer marketing on tik tok But if you're someone that is saying of course, it's impossible people will respond well to AI influencers or digital avatars, I think I would take a step back and consider that
[01:04:32] Mike Kaput: because some of the, of the, some of the companion AI tools, character.
[01:04:37] Mike Kaput: ai, all this stuff is proving that behavior might be changing a little faster than you think.
[01:04:43] Paul Roetzer: Yes, I think that's a very good point, Mike.
[01:04:46] Mike Kaput: All right. So we've got a couple more items to wrap up here.
[01:04:50] Mike Kaput: ChatGPT actually just got a major upgrade, so much going on that, you know, this is far down the docket, but,
[01:04:57] Mike Kaput: h,
[01:04:57] Mike Kaput: but h, OpenAI announced that [01:05:00] users with a ChatGPT Plus team or enterprise license are now able
[01:05:04] Mike Kaput: use
[01:05:04] Mike Kaput: an updated version of GPT 4 Turbo within ChatGPT, so it makes it much more powerful and better at writing math.
[01:05:13] Mike Kaput: Writing, Math, Reasoning, and Coding. It also updates ChatGPT's knowledge base, so
[01:05:19] Mike Kaput: instead, now it is trained on publicly available data up to December 2023, so getting closer to the current moment.
[01:05:28] Mike Kaput: The company said this update should also make quote, more direct, less verbose, and use more conversational language.
[01:05:37] Mike Kaput: So based on the chatter online, Paul, it kind of seems like OpenAI
[01:05:40] Mike Kaput: feeling
[01:05:41] Mike Kaput: a little bit of the pressure, at least, as everyone's talking about advanced models
[01:05:45] Mike Kaput: Claude3 Opus, Gemini 1. 5
[01:05:48] Mike Kaput: Pro. Do you think that's what prompted this update here?
[01:05:52] Paul Roetzer: I don't think so. I just become more convinced by the day that GPT 5 is going to be just [01:06:00] So crazy. Like, such a leap over. What we expect it to be like so much smarter than we expect it to be. And I think they're just putting out a prelude to it. I don't think that they feel the pressure that other people probably do, their models.
[01:06:19] Paul Roetzer: yeah. Oh, total side note, but I think I mentioned this to you on Friday, Mike, but just related to these models, I was listening to an interview with Jack Clark, who's co founder of Anthropic, and on a recent episode, we talked about how I was. Confused that open or that Anthropic was like racing ahead with Claude3 Opus.
[01:06:37] Paul Roetzer: And they're just building these like frontier models. So powerful when I thought the original idea of their company was safety and security of models ends up that that actually isn't the story. And so maybe we get into this in another episode, but Jack Clark, for the first time I ever heard, and.
[01:06:52] Paul Roetzer: Admitted that that is not why they all left open AI. That they actually just wanted to get in and build their own frontier model before this got too big and cost too much money. [01:07:00] And the safety and security thing didn't come in until a couple of months after the founding of Anthropic, the way they built a constitutional AI.
[01:07:06] Paul Roetzer: And it was just like, oh yeah, we'll make it safe, but. Anthropic, to my understanding, and everyone else I'd ever heard interview them, was that they left to build a safer version. And Jack basically nixed that idea and said, that is not why we build Anthropic. So it solves my conundrum of like, why, why are they moving so fast and building such powerful models?
[01:07:27] Paul Roetzer: It doesn't seem like their own guidelines ends up that. That wasn't why they
[01:07:32] Paul Roetzer: did it
[01:07:32] Paul Roetzer: it in the first place, according
[01:07:33] Paul Roetzer: to Jack. So, fascinating
[01:07:35] Paul Roetzer: interview. I'll, we'll put that, podcast in the show notes. It was a really good interview.
[01:07:39] Mike Kaput: All right, so
[01:07:41] Mike Kaput: our last topic today is that Adobe is coming under a little bit of scrutiny for how it has trained its Firefly image generation software.
[01:07:49] Mike Kaput: This is according to some reporting from Bloomberg. So, since Firefly's release, the company has been touting this system as something that's trained using only [01:08:00] content that
[01:08:00] Mike Kaput: company
[01:08:01] Mike Kaput: has the rights to or content that's in the public domain. So, Adobe is routinely promoting Firefly as the ethical, legally sound option for customers since it claims Firefly was not generating content that was based on intellectual property that people or brands had out on the internet.
[01:08:20] Mike Kaput: However, it now appears some images that were created using MidJourney, another AI image generation tool, have made it into Adobe's training data. So the problem here is this. MidJourney, among some others, have come under fire because their systems are trained on possibly copyrighted work online. So MidJourney alone is embroiled in a number of lawsuits.
[01:08:44] Mike Kaput: By using some of MidJourney's images in its training data, Adobe has then opened itself up the
[01:08:51] Mike Kaput: same criticisms, because it's at least in part possibly using copyrighted material to train its models on. Now according to [01:09:00] Bloomberg, it turns out AI generated content from MidJourney made it into Firefly's training data because, quote, creators were allowed to submit millions of images into Adobe's stock marketplace.
[01:09:12] Mike Kaput: that use the technology from other companies. Adobe says that a relatively small number of images were used to train tool
[01:09:21] Mike Kaput: that were generated by other AI platforms, about 5%. So Paul, as you go through the details here, how bad is this terms of violating Adobe's
[01:09:32] Mike Kaput: previous comments and commitments to ethical AI?
[01:09:35] Paul Roetzer: It doesn't look good, but in terms of how blatant this is versus everybody else, that's just straight up BS. Stealing everything. Well, I use stealing as like a yet to be defined
[01:09:50] Paul Roetzer: legal term. um, use, let's go back. Using copyrighted images and not admitting to it. How about that? so I, [01:10:00] you know, is it going to change anything?
[01:10:01] Paul Roetzer: Are they going to lose clients because of it? I highly doubt it. It just doesn't look great considering they were taking this very ethical stand. So I think whatever is going to learn in the future is like at the end of the day, as long as you can get away from the legal liability stuff. um, we always said in PR, like I used to do crisis communications work.
[01:10:18] Paul Roetzer: Tell the truth. Tell it first, like just control the narrative by just being straight up. I get that these companies can't do this, but everyone is just like hiding behind these like statements and just say it, like get it over with, say we did it, or this is how we're doing it. And we think it's legal, and like, prove us wrong, we'll pay our fines, but like, if this is how the industry's going to be built, and they think this is a viable, ethical, legal way to do it, then just say it, and own the story.
[01:10:49] Paul Roetzer: I don't know, like, again, it's
[01:10:52] Paul Roetzer: of me is just like an observer of it, part of me is like the PR person in me from my days of Crisis comes, like, but it [01:11:00] drives me nuts that we're just at this stage where all these big tech companies are so afraid to just say what they're actually doing. and it's not for competitive reasons.
[01:11:08] Paul Roetzer: Give me a break. They all know what each other's training on. They're
[01:11:11] Paul Roetzer: just saying.
[01:11:12] Paul Roetzer: they just won't tell the public what they're doing.
[01:11:14] Mike Kaput: All right, Paul, that's all we got this week in, this week in AI. I would encourage everyone listening, if you are enjoying the content of the podcast, if you have gotten value out of it, please, please leave us a review.
[01:11:30] Mike Kaput: This helps us get the podcast in the hands and earbuds of as many people possible. So please give us a review on the platform of your choice.
[01:11:40] Mike Kaput: Last but not least. Please if you have not already, check out Marketing AI Institute's newsletter. Go to marketing ai institute.com/newsletter. You can subscribe to our this week in AI newsletter.
[01:11:54] Mike Kaput: It not only includes all of the topics that we discussed on the podcast today, but [01:12:00] also all the topics we didn't get to because there's a lot more going on in AI than a single podcast episode can cover. So we give you that in a comprehensive brief.
[01:12:10] Mike Kaput: Every
[01:12:10] Mike Kaput: single week so you can keep up to speed on this fast moving field. Paul, thanks again for helping us demystify what's going on in AI this week.
[01:12:20] Paul Roetzer: Thank you, Mike, and thanks everyone for listening. We'll talk to you next
[01:12:22] Paul Roetzer: week.
[01:12:24] 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.
[01:12:47] Until next time, stay curious and explore AI.