From AI assistants pulling pranks to companies taking bold anti-AI stances, join our hosts as they navigate the wild world of AI innovation, controversy, and Rick Astley-inspired shenanigans.
This week, Paul Roetzer and Mike Kaput unveil CampaignsGPT, a cutting-edge tool from SmarterX for analyzing AI's impact on campaign tasks. Our hosts also examine a16z's latest top 100 genAI consumer apps list and explore Cohere CEO Aidan Gomez's insights on AI's future. In rapid fire, we cover Amazon Q's productivity boost, Procreate's bold anti-AI stance, Lindy AI's mischievous behavior, Microsoft Copilot's security concerns, and more.
Listen Now
Watch the Video
Timestamps
00:04:11 — All New CampaignsGPT
00:14:10 — a16z Top 100 Gen AI Consumer Apps
00:24:13 — Cohere CEO Aidan Gomez on 20VC Podcast
00:35:43 — Amazon’s AI Assistant Productivity Gains
00:42:18 — Cursor AI Coding Assistant
00:46:18 — Procreate Anti-AI Stance
00:50:01 — OpenAI partners with Condé Nast
00:52:21 — OpenAI Names First Chief Communications Officer
00:55:33 —Anthropic and OpenAI Release Statements on SB-1047
00:58:51 — Lindy AI Goes Rogue
01:02:20 — Microsoft Copilot Safety and Security Concerns
01:05:05 — LinkedIn AI Disclaimers
Summary
All New Campaigns GPT
SmarterX just dropped their newest tool powered by ChatGPT, focusing on how AI, especially large language models (LLMs), impacts campaigns and the future of work.
Developed by Paul Roetzer, CampaignsGPT breaks down campaigns into tasks, evaluates AI's influence on each task, and gives practical insights to prioritize AI applications and boost productivity.
This tool allows for business leaders to look at how more advanced AI models might affect their teams and organizations, through the exploration of AI’s role in campaign planning and execution.
CampaignsGPT is currently in beta (powered by ChatGPT), so there may be occasional operational hiccups and it can hallucinate at times. We appreciate any feedback and hope you find it helpful as you plan for the future of work.
a16z Top 100 Gen AI Consumer Apps
Venture capital firm Andreessen Horowitz has released the third edition of its top 100 genAI consumer apps list. Every six months, the company looks at data to rank the top 50 “AI-first” web products (by unique monthly visits) and the top 50 “AI-first” mobile apps (by monthly active users).
Compared to their March 2024 report, the company says nearly 30% of the products on the list are new. Perhaps predictably, ChatGPT tops both the web and mobile list as the most popular product. But the similarities end there.
On web, the top 5 products are: ChatGPT, Character AI, Perplexity, Claude, and Suno (the music generation app).
On mobile, the top 5 are: ChatGPT, Microsoft Edge, Photomath (an app that solves math homework), Nova (which appears to be an AI assistant powered by ChatGPT), and Bing.
The company also highlighted some top trends to come out of the data.
Cohere CEO Aidan Gomez on 20VC Podcast
Continuing our habit of paying very close attention to what the leaders of AI are talking about, we have another deep dive this week—this time on an interview with Aidan Gomez, the co-founder and CEO of Cohere, a leading large language model company.
Gomez just gave an hour-long interview to Harry Stebbings on the popular 20VC podcast covering a wide range of topics, including:
- His belief that AI is nowhere near plateauing…
- Small vs. large models…
- How it’s getting harder to train models…
- And his claim that OpenAI has put AGI on the back burner.
Links Referenced in the Show
- CampaignsGPT
- A16z List
- Aidan Gomez on 20VC Podcast
- Amazon AI Productivity
- Cursor AI
- Procreate Anti-AI Stance
- OpenAI + Conde Nast
- OpenAI Names First Chief Communications Officer
- Anthropic and OpenAI SB-1047 Letters
- Lindy AI Goes Rogue
- Microsoft Copilot Safety and Security Concerns
- LinkedIn AI Disclaimers
This week’s episode is brought to you by MAICON, our 5th annual Marketing AI Conference, happening in Cleveland, Sept. 10 - 12. The code POD200 saves $200 on all pass types.
For more information on MAICON and to register for this year’s conference, visit www.MAICON.ai.
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: my daughter's original take on image generation, I thought was genius, when she was 10, where she said it's just stealing people's imagination.
[00:00:06] Paul Roetzer: And, and I like that it's still probably the most profound thing anyone's ever said to me about generative AI is it just goes and learns from our stuff, our creations, our imaginations, and then it just takes them and remixes them.
[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.
[00:00:49] Paul Roetzer: Join us as we accelerate AI literacy for all.
[00:00:56] Paul Roetzer: Welcome to episode 112 of the Artificial [00:01:00] Intelligence Show.
[00:01:00] Paul Roetzer: I'm your host, Paul Roetzer, along with my co host, Mike Kaput. We are recording this Monday, August 26th, about 10 a. m. Eastern time. In case anything major happens today, we'll get, get the timestamp in there. this episode is brought to us by the Marketing AI Conference or MAICON. some people say MyCon, that's fine too.
[00:01:20] Paul Roetzer: Our fifth annual event, it's happening, September 10th to the 12th. This is coming up fast. We are now. just under two weeks to go before the event. So we have, amazing event plan. There's over 69, well, there's exactly 69 sessions currently planned. There's 33 breakouts across two tracks. We've got an applied AI track.
[00:01:41] Paul Roetzer: That's all about use cases, technologies, case studies, how to's, and then we've got a strategic AI track that's. Bigger picture, it's about talent, technology, strategy, budgets, change management. So big picture stuff, kind of, I think of the strategic AI track as more of the director level and above track.
[00:01:59] Paul Roetzer: The [00:02:00] applied AI track is more of that practitioner track. So there's going to be four to five breakouts going on, at each breakout, segment. And then we've got 10 main stage general sessions and keynotes, five lunch labs, three pre conference workshops. Mike's doing an applied AI workshop, all about prioritizing AI use cases.
[00:02:21] Paul Roetzer: I'm doing a strategic leader workshop. That's all about the problem based model and finding ways to drive innovation and growth within the organization through those. we've got mindfulness sessions. We've got a bunch of parties, 16 AI tech demos. It's just going to be an incredible event. So if you can get to Cleveland, September 12th, it's at the Huntington Convention Center, right across from the Rock and Roll Hall of Fame in Lake Erie.
[00:02:42] Paul Roetzer: It's a beautiful venue. it's going to be an amazing few days. So we would love to have you there. use POD200, we'll get 200 off all pass types. You can go to MAICON. AI, that's M A I C O N. A I. Click register and use that code again, [00:03:00] POD200.
[00:03:00] Paul Roetzer: P O D 200.
[00:03:02] Paul Roetzer: get you the 200 off any of the past options. And, Mike and I would love to connect with everybody there.
[00:03:08] Paul Roetzer: It's, you know, we've got this amazing growing audience for this podcast and it's always so fun to get to meet people in person. I always joke, Mike, people I meet who listen to the podcast all the time, who think I talk really slowly in person because they listen to us at like 1. 5 speed and then they hear us talk in person.
[00:03:25] Paul Roetzer: It's like, Oh, you speak really slow. So.
[00:03:29] Mike Kaput: I had someone tell me that I was taller than I sounded, and I was like, I don't know what that even, what that even, what that even means.
[00:03:36] Paul Roetzer: That's hilarious. Taller
[00:03:42] Paul Roetzer: you sound. That's great. I wonder how tall I sound.
[00:03:45] Mike Kaput: have no idea.
[00:03:47] Paul Roetzer: All right. So, yeah, we'd love to meet you in person, show you how tall we actually are. MAICON. AI, check that out. And again, use pod 200. 14 days to [00:04:00] go. I'm starting to get a little nervous as that countdown clock goes down.
[00:04:05] Paul Roetzer: All right, let's get into it. We've got some fascinating stuff today. I had fun preparing for this one. There's some good stuff going on.
[00:04:11] CampaignsGPT
[00:04:11] Mike Kaput: So first up on episode 110, we debuted and talked about JobsGPT, which is a ChatGPT powered tool that's designed to assess the impact of AI on jobs by breaking down different roles into tasks and then evaluating how AI can enhance productivity. Now this tool was built by Paul and through SmarterX. AI, your research, AI research and consulting firm.
[00:04:40] Mike Kaput: And now Paul, you are at it again with a new tool called CampaignsGPT that just dropped. Now this tool uses the same technology and methodology to break down different types of business campaigns into tasks and assess each one's exposure. to AI. So you just give campaigns, [00:05:00] GPT, a type of campaign you want to assess.
[00:05:04] Mike Kaput: For instance, product launch, lead generation, or change management. And the tool will actually break down that campaign into tasks and subtasks. Those tasks and subtasks are then labeled according to their exposure to AI. So exposure here means the ability for a large language model to reduce the time it takes to complete tasks with equivalent or greater quality.
[00:05:29] Mike Kaput: than an average skilled professional. Now what's really cool is this tool can be used across business units, not just marketing. It includes marketing, sales, service, HR, finance, and more. So Paul, maybe kick us off here by walking us through why you created Campaigns GPT and why people should be using this.
[00:05:50] Paul Roetzer: Yeah, so this was actually built, sort of at the same time as jobs GPT. So when I, when I built jobs GPT, I was playing around with the same [00:06:00] kind of instructions, you know, and building the custom GPT. And the reason we did this is like when Mike and I teach. AI Adoption and Enterprises, you know, the Applied AI Workshop that Mike's leading at MAICON uses the same concepts, these frameworks.
[00:06:15] Paul Roetzer: So what we do is we teach people, if you want to get value out of ChatGPT, Gemini, Claude, other large language models, one of the best ways to do it is to look at what you already do and find ways to infuse AI into those processes. And so. The first thing we'll teach, and again, Mike runs through this in the workshop model, is, to look at your job and say, what are the things you do every day, every week, every month?
[00:06:40] Paul Roetzer: Create a list. What are the tasks that go into that job? And that's where jobs GPT came from. It's like, let's help people accelerate this process. The second thing we'll teach people is look at the campaigns you run. and again, I think campaigns, Our most commonly thought of in the marketing realm, like that we came up in the marketing agency world, we ran campaigns for [00:07:00] clients.
[00:07:00] Paul Roetzer: That's what we did. So whether it's like influencer marketing, lead gen and nurturing, product launches, monthly analytics reports, price optimization, rebranding, SEO, our weekly podcasts, these are all campaigns. They're a collection of tasks designed to achieve a goal. So. We think about in marketing, but it does apply to all these other areas.
[00:07:20] Paul Roetzer: Salespeople run business development, customer referral, market expansion, new product launches, customer service, does customer onboarding, customer retention, customer service training. Like there's all these campaigns, these ongoing repetitive series of tasks that are designed to achieve goals. And so the idea with this product was one, Let, let's just start there.
[00:07:40] Paul Roetzer: Think about your job and then think about the campaigns you run. And if you just do those two things, you're going to find all kinds of use cases for AI. So Campaigns GPT does two things. So one is at that kind of practitioner practitioner level, it identifies and automates use cases. So it helps you find the things you already do and [00:08:00] then figure out which ones AI can help you with.
[00:08:02] Paul Roetzer: The second component is more for the business leader where it's helping people. Analyze what goes into planning and running campaigns, and then look one to two years out as these models get smarter. And more generally capable and figure out how those models are going to affect your staff, your strategies, your budgets, your timelines.
[00:08:26] Paul Roetzer: And so that's kind of how it's structured is you can just go in there and have it identify use cases based on campaigns that you already run. You can upload your existing campaigns if you want. It'll just work. Tell you, you know, your, your own tasks, which ones can be automated and time saved and things like that, or you can just give it a campaign type, like an influencer marketing campaign, and it'll break it into tasks and subtasks.
[00:08:48] Paul Roetzer: So, You know, again, there's these two audiences you're trying to create a single tool for. but from what we've heard with jobs, GPT, people are finding a ton of value in this. So I think that's [00:09:00] now been used like 13 or 1400 conversations with that tool that we launched a couple of weeks ago. And so, yeah, that's the idea here is like, just put these tools out that help people get to adoption faster, but also help, you know, Business leaders start doing the kind of impact assessments that we think are really, really important to the future of work and to preparing, to reskill and upskill our workforces as these models get smarter and they're able to do more and more of the work.
[00:09:30] Paul Roetzer: And so that's the idea with the exposure key is. You know, when we think about what the models are capable of today, they're largely capable of text inputs and text outputs, so they can do some image, you know, obviously with Gemini and ChatGPT, you can do image inputs and image outputs to a degree. Video is kind of a standalone thing right now.
[00:09:51] Paul Roetzer: It's Not baked into these language models. They don't have advanced reasoning capabilities yet. We'll talk a little bit more about reasoning and some of the topics later in this episode. [00:10:00] They're not highly persuasive yet. Like they have some persuasion capabilities, but they don't really have those things.
[00:10:06] Paul Roetzer: And so the exposure key is basically looking at as these models develop these other abilities, then what is the impact that they have on campaigns and jobs? So. Yeah, you can check it out. Just go to smarterx. ai, and then click on tools and jobs and campaigns, GPT are both right there. The way this one works is there's three conversation starters.
[00:10:29] Paul Roetzer: If you've never used like a custom GPT, you can kind of set what the conversation starters are. So this one says, Enter a campaign type to assess. So you can just say, you know, sales pipeline analysis or whatever your campaign type is, and it'll automatically create all the tasks. And then you can say, okay, cool.
[00:10:45] Paul Roetzer: Break it into subtasks. You can, the other Conversation Stars provide a campaign with tasks to assess, where you just upload your own or copy and paste your own. And the other one is just show me an example assessment so you can get a sense of how it works. so the final thing I'll [00:11:00] say is like, it is built.
[00:11:01] Paul Roetzer: It's a chat GPT tool so it can make stuff up. It, it's not gonna be perfect, it's not gonna be right all the time. But the beauty of this is you can talk to it like you would chat GPT, like think of this thing as a planning assistant. Don't just stop it like a single assessment, or output. You know, have a conversation with it.
[00:11:19] Paul Roetzer: Ask it to break into the subtasks or talked about how it's prioritizing tasks or to help you prioritize which ones to focus on. Um. You know, and just try it out. Like it's, it's meant to be kind of a research tool, a planning tool, and it's very much a beta product for us. Let's just kind of get this stuff out into the world and see how it can help people.
[00:11:37] Paul Roetzer: And you've played with it, Mike. I mean, you, you and I kind of went back and forth on it. What were your initial reactions to it?
[00:11:43] Mike Kaput: What I love is that it removes what I would see as the final barriers to getting started with this type of planning, because as we've talked about over the years, there are still some barriers of like, when you sit down with someone in a workshop, for instance, they still can have trouble coming up, like, thinking logically [00:12:00] or sequentially through as, about their job or their campaign as a bundle of tasks.
[00:12:04] Mike Kaput: Not everyone thinks that way. Like, that comes naturally to, in different to different people. So this is literally, you have no real. excuse here when you sit down in front of it, even if you don't have some grand plan of the transformation you want to go through, just type in your title with JobsGPT or just type in what you are working on right now with CampaignsGPT and I, I think in, as the final note here, I would have it's, it's, it's.
[00:12:30] Mike Kaput: That makes it a really good tool to share if you're a listener with your executives too, because as a listener, you're going to probably be a little ahead of the curve, I would think, and you're going to have a great time using these tools and get a lot of value out of them. But these are also really good to just forward to people on your team.
[00:12:45] Mike Kaput: You're trying to convince of the need for change management and also to just show them what's possible today.
[00:12:52] Paul Roetzer: And in the Applied AI workshop, I mentioned like Mike's running that at MAICON, but Mike and I run those things year round for companies and like [00:13:00] private engagements, we'll run these workshops and just to give you a practical sense of how to, how you can use these things. When we teach these frameworks.
[00:13:07] Paul Roetzer: We'll then say, okay, you've got 20 minutes, here's a template, start writing down what you do, or what campaigns you're running. And so we would spend 20 minutes of a three and a half hour workshop, almost 10 percent of the workshop, basically, or whatever that comes out to, with people just staring at blank pages, trying to like, Think of those tasks, like you said, Mike.
[00:13:27] Paul Roetzer: And that's not a natural thing, especially when you're on the spot and like, okay, I only have 20 minutes. I got to think about this. So part of the idea here and how we're kind of like reinventing how these workshops run is like, okay, Now think about the campaigns you run, but put them in here. Just like quick brainstorm, put the kinds of campaigns you run.
[00:13:44] Paul Roetzer: And then like, now all of a sudden you've got dozens or hundreds of tasks where it's like, oh my gosh. So the whole idea is just, again, accelerate adoption by, by getting this manual part out of the way and like helping people, just get those ideas flowing. And that's [00:14:00] oftentimes where you see in workshops.
[00:14:03] Paul Roetzer: That blank page and people just kind of staring around the room or like kind of looking for some inspiration. So that's part of the reason for it.
[00:14:10] a16z Top 100 Gen AI Consumer Apps
[00:14:10] Mike Kaput: All right. Our second big topic this week, venture capital firm Andreessen Horowitz has released the third edition of their top 100 generative AI consumer apps list. So every six months, Andreessen Horowitz looks at data to rank the top 50, what they call quote AI first web products, and they do this by looking at unique monthly visits and the top 50 AI first mobile apps, which they.
[00:14:38] Mike Kaput: rate based on monthly active users. So a couple of interesting things to come out of the most recent one compared to their report that they released in March, 2024, the company says nearly 30 percent of the products on the, on the list right now are new ones. Now, predictably, perhaps ChatGPT. It tops both the [00:15:00] web and mobile list as the most popular product, but the similarities kind of end there.
[00:15:05] Mike Kaput: So, on the web, the top five products are ChatGPT, Character AI, Perplexity, Claude, and then Suno, the music generation app. On mobile, the top five are quite different. ChatGPT is the only similarity at number one, then Microsoft Edge, an app called PhotoMath, which solves math homework with AI, Nova, which appears to be an AI assistant powered by ChatGPT, and Bing.
[00:15:35] Mike Kaput: Now, the company also took a look at some of the trends coming out of their data and out of the rankings list. These include things like, first, creative tools are dominating the list. 52 percent of the tools on the web list are related to creative tasks. 58 percent of the new tools on the list this time around are creative tools.
[00:15:56] Mike Kaput: For instance, content editing is the most popular use case [00:16:00] for products coming out on mobile. Second, there was a new category of products that appeared this time around. What they call Aesthetics and Dating. This includes an app called LooksMax that analyzes a user's photo to offer tips on how they can become more attractive.
[00:16:18] Mike Kaput: It will even go as far as to analyze your voice to help you do that. And then another big trend is that Discord is driving a lot of the growth of these products. In many cases, Discord Discord traffic, says, Andreessen Horwitz, is a leading indicator for apps that will climb the web and mobile ranks, especially when it comes to content generation.
[00:16:39] Mike Kaput: So Paul, as you are looking at this list, and we've tracked this, multiple of these lists over the last few months, did any apps or trends stand out to you here? Hmm.
[00:16:51] Paul Roetzer: so the, from a trends perspective, I'll, I'll start there. the rise of kind of multimodality is, is very apparent. [00:17:00] You know, you mentioned it, you know, how things are evolving from a standpoint of like image and video. so they said the majority of content generation tools previously focused on image over the last six months, other modalities have gained steam image generally represents 41 percent of the top gen, content gen sites.
[00:17:18] Paul Roetzer: The five generation tools that made the list for the first time, there was C Art is the only image one. You had Luma, Vigil, and Vidnaz, from video. Music had Udio, Suno's on the list as well. So it's like you're starting to see those modalities and even go back to the campaign's GPT and the exposure level.
[00:17:38] Paul Roetzer: This is part of it. There's an exposure level for image. There's an exposure level for video. Exposure level for audio, including music. And again, the idea here is many of these modalities are standalone apps right now. So we're seeing advancements in video. Honestly, aren't that good? Like video gen tools, you see these amazing demos and it like looks [00:18:00] awesome.
[00:18:00] Paul Roetzer: But the reality is it's really really slow. So if you want to create something in Gen 3 from Runway or Loom or whatever, you may be waiting hours if you're in a queue for these things to generate and then they'll do like five seconds at a time. And it's really hard to get the consistency. So like These are very early.
[00:18:17] Paul Roetzer: Like, don't think that just because these tools are rising that they're all, this is all solved and like, you know, this stuff is consumer grade and ready to go, but it's early indicators that these tools are becoming more and more popular, even though these are like first generation versions of what we're seeing in these capabilities.
[00:18:34] Paul Roetzer: it's kind of a surprise not to see Runway on the list. Like, we talk a lot about Runway and they've got the Gen 3 video, but, you know, I mean, maybe it's just more of a business tool, or maybe that's more cost prohibitive at this point. the other trend was the rise of perplexity, the continued rise of perplexity, you know, becoming more popular with consumers, not just within kind of the AI circles and business circles, but becoming more of a consumer application, [00:19:00] Anthropx, Claude, you know, continues to rise.
[00:19:03] Paul Roetzer: The one thing I found fascinating, I guess I didn't even thought about this is TikTok's parent company, ByteDance. making a push, and I, there's a couple of reasons I thought this might be happening. Um, one could be the threat to TikTok, and so they're diversifying into all these other areas. But they had, ByteDance had three of its apps debut on the list.
[00:19:23] Paul Roetzer: There's the ed tech platform, Goth, number 44. There's a bot builder called Goat. Coase or Cozy, and then a general assistant called Dubao, and then there was, they actually had a total of six, so they had three other apps. So this was news to me, like I didn't realize ByteDance was kind of infiltrating through all these AI apps, but apparently they launched an R& D division called Flow in 2023 that's focused on building generative AI applications.
[00:19:52] Paul Roetzer: So. That kind of made sense. you mentioned the looks max and you max like, Oh my, I just, [00:20:00] I, again, sometimes I just feel really old. Like I can't even, I totally get that this is doable and the technology allows this, but people are paying five bucks a month apparently to have AI show you what you'd look like as a 10 out of 10, like how your looks need to change.
[00:20:16] Paul Roetzer: So you would rate as a 10 out of 10 and how to make your voice more attractive. Like. I just, I don't know. I was just
[00:20:23] Mike Kaput: The voice thing was wild to
[00:20:24] Paul Roetzer: Yeah, I, maybe I can get my voice to make me sound taller.
[00:20:29] Mike Kaput: Yeah.
[00:20:30] Paul Roetzer: Riz, then there was one called Riz, which, you know, I have a 12 year old and 11 year old. I hear the term Riz like 20 times a day.
[00:20:37] Paul Roetzer: Riz on the other hand, focus on upleveling dating app messages. Users can upload a conversation screenshot or profile and get suggestions on what to say. Then there was one that came in at number 11 on the Gen AI web products, Spicy Chat. I did not go to that app and see what that one's all about, but I can imagine.
[00:20:56] Paul Roetzer: Um, so yeah, I just, [00:21:00] I guess my one overall take is like my personal use and awareness is quite low. of the top 50. Like, you know, when you look at the top 10 in web products, ChatGPT, Character. ai, Perplexity, Claude, Suno, I'm familiar with all those, Quillbot, Poe, I know those. But once you start getting beyond that, I mean, Hugging Face, Eleven Labs, Luma, I know all those, but when you go into the mobile app side, I don't know, like I was familiar with like four of them.
[00:21:27] Paul Roetzer: So there's this whole generation of these, you know, Gen AI mobile apps that are emerging and I I'm blissfully unaware, I think, of a lot of them because I think a lot of them are, are being more for that kind of the dating scene and the personalized characters and things that I'm just not living in that world right now.
[00:21:47] Paul Roetzer: I would say I would love to see a business list like this. I don't know. I don't think they've done that. I haven't seen one, but like our data from the state of marketing, AI report this year, Kind of [00:22:00] jives with some of this. So we had ChatGPT is by far the most popular Gen AI platform in businesses.
[00:22:06] Paul Roetzer: And then when we asked favorite AI tools, you had ChatGPT, Perplexity, Claude, Google Gemini, Adobe Firefly, Canva, Descript, Grammarly. so yeah, I just, I don't know, it's cool. Like I think it's, I would recommend people go check it out. They've got some really simple visuals you can look at. And then the way they did it, I don't, I don't think I said, did I say this yet?
[00:22:26] Paul Roetzer: the way they determine it is for web apps, they look at unique monthly visits per similar web and their data is from July, 2024. So it's fresh data. And then for the mobile apps, they look at monthly active users per sensor tower, which I've never heard of, but that's apparently the data source for the, how they evaluate these things.
[00:22:46] Mike Kaput: Yeah, it's quite eye opening to me and really just proves a point that we've talked about previously is like clearly these apps are skewed towards like Gen Z or younger. Your [00:23:00] kids, kids of any of that generation, are going to be, are living an AI first lifestyle whether they know they're using it or not.
[00:23:08] Paul Roetzer: Yeah. And I don't know if you've seen on like TikTok and Twitter just in the last week, I've been seeing a lot of these videos of college today where they've got their laptops and this is like filmed in classrooms and they've got, they're taking a video with their phone of their laptop. The laptop has AI running.
[00:23:26] Paul Roetzer: That's recording everything the professor is saying and even seeing the board and then doing analysis in real time, solving problems in real time on there. So they're using Gen AI assistance. In the classroom, in real time to solve things and learn things and like explain things in different ways. I mean, we're, we're in a paradigm shift in education, and I just don't think we're, as a society, generally aware of how transformational this is going to be.
[00:23:54] Paul Roetzer: And then when you look at these lists, you start to realize how throughout society, communications, information [00:24:00] gathering, relationship building, like, All of it is just getting fundamentally reinvented before our eyes, and I, I don't know that we're, until you zoom out, you become aware of how significant the change is.
[00:24:13] Aidan Gomez on 20VC Podcast
[00:24:13] Mike Kaput: Alright, so our third big topic this week, we are continuing our habit of paying very close attention to what the leaders in AI are talking about, and we're doing another deep dive into that this week. This time we are looking at an interview that was just given by Aidan Gomez, who is the co founder and CEO of Cohere, which is a leading large language model company.
[00:24:38] Mike Kaput: Gomez just gave an hour long interview to Harry Stebbings on the popular 20VC podcast, and covered a ton of really interesting topics, including things like he believes that AI appears to be nowhere near plateauing in its capabilities. He talks about small versus large models and the advantages of each.
[00:24:58] Mike Kaput: He mentions how hard [00:25:00] it's getting to train models to be better and more powerful. And he claims that OpenAI has put AGI on the back burner in favor of becoming a consumer consumer. So, Paul, what jumped out to you as worth paying attention to in this interview? Maybe kind of also just give us a really quick sense of like who Gomez is and why he's worth listening to.
[00:25:23] Paul Roetzer: Yeah, so we have talked about Aidan, you know, a number of times before on the podcast. We've talked about Cohere, quite a bit. So Aidan was one of the nine authors of the Intention is All You Need paper in 2017 that Google Brain published that invented the transformer. You know, so we've talked about the transformer as the basis for GPT.
[00:25:41] Paul Roetzer: It's the basis for large language models today. So he was an intern at the time, I believe in the Google brain team, but he was one of the nine cited authors of that paper. So he left to co found Cohere. they've raised about a billion dollars at somewhere between a five and a 6 billion valuation. As we've talked [00:26:00] in recent.
[00:26:01] Paul Roetzer: Like weeks and months though, that's not a lot of money, in the gen AI space. I mean, a billion sounds like a lot, but I mean, inflection had raised a billion and a half before they got acquihired. Character. ai had raised probably close to a billion. I don't remember their exact number before they got acquihired recently.
[00:26:17] Paul Roetzer: So, These numbers sound big, but in reality, in order to compete with the frontier model companies, the OpenAIs and Anthropics and Meta and Google, a billion is not much. And so that's actually kind of part of the basis for the interview is like, well, how are you building a large language model company differently that allows you to, you know, Keep competing and not get acquihired.
[00:26:42] Paul Roetzer: And I'm not saying they're not going to get acquihired, but up till now, they have not. Um, so I'll hit on a few of the trends. Again, I think it's a worthwhile listen. Like if you are a listener who enjoys kind of going a little deeper on these, as Mike said, hearing from the people that [00:27:00] are building the models, building these, you know, the major Gen AI companies, it's, it's worth your time to really hear their full perspective.
[00:27:07] Paul Roetzer: But I'll just highlight some of the key themes that jumped out at me. So on the topic of making bot models bigger. Kind of this idea of scaling laws that we keep talking about, that if we just keep giving more NVIDIA chips to them, more data and more time to train, they just get smarter. That's the basic premise of the scaling laws.
[00:27:25] Paul Roetzer: And he said, it's definitely true. And I'll just quote here. It's definitely true that if you throw more compute at the model, if you make the model bigger, it'll get better. It's kind of like its most trustworthy way to improve models. It's also the dumbest. and he said, He said, I just think it's extremely inefficient.
[00:27:42] Paul Roetzer: There are much better ways. And then he said, there's a lot of pressure on making smaller, more efficient models, smarter via data and algorithms, methods, rather than just scaling up due to market forces. There's pressure on price. So his main thing is like, yeah, the models are going to keep getting smarter and more generally [00:28:00] capable.
[00:28:00] Paul Roetzer: If you are one of the five frontier model companies that has billions of dollars to spend, if not tens or hundreds of billions, then you're Then yes, you can keep building much larger models. And he is a believer in the scaling laws. He doesn't see them ending anytime soon. So then he went on to talk about these smaller, more efficient models.
[00:28:17] Paul Roetzer: and he says, we live in this world of unbundled verticalized models, which are much more efficient and smaller designed for specific use cases. He says, there will be both. We will have these massive general models, and then we will have these smaller verticalized models. and that's just how it's going to go.
[00:28:35] Paul Roetzer: So you have some verticalized and then the horizontal generally capable ones. On the question of competing with frontier model companies, specifically Microsoft, Amazon, Google, Facebook, and then you add OpenAI and Anthropic to that list, he said, I think you're just doing the scaling project. You have to be one of those, or you have to be an effective subsidiary of one of those companies.
[00:28:54] Paul Roetzer: So in other words, If you want to build the big, generally capable models, you're only, it's only going [00:29:00] to come from one of those five companies or someone who's basically getting a bunch of money from them and functions like a subsidiary, but isn't being acquired because of, you know, constraints right now.
[00:29:10] Paul Roetzer: now. he said pretty much all of the major gains that we've seen in open source space have come from data improvements though. Models getting much better by taking higher quality data from the internet, better scraping algorithms, parsing those web pages, pulling out the right parts. Upweighting specific parts of the internet, because there's a lot of repetition and junk.
[00:29:29] Paul Roetzer: So this goes, we've talked about this idea before, but when these models are trained, when GPT 5 is trained, it just sucks in all available content. All, just imagine all the text on the internet going into it. Maybe it's multimodal, maybe they're training it on images and videos and audio too. But it's basically taking all of this data.
[00:29:48] Paul Roetzer: Then what these researchers have to do is apply weights to the importance of different sources. So what they're going to do is say like research papers are better than Reddit boards. And so they may give more weight in the [00:30:00] training of these models to content that comes from research papers that have been published in journals, as an example, or authored books.
[00:30:07] Paul Roetzer: It's why they take from books three, the pirated, you know, 180, 000 books, because the assumption is stuff that's been published. through actual publishing houses, through publishers, is better. It's why they take from the New York Times and all these other places. So what he's saying is, they are definitely seeing that.
[00:30:24] Paul Roetzer: That the higher the quality of data, the less volume of data you need. And that's a really key thing that they've learned now in the last couple years. So he says, pulling out the most valuable knowledge, rich parts of the internet and emphasizing those to the models, telling it like, focus on this.
[00:30:40] Paul Roetzer: This is the good stuff. He talked about system two thinking. This is a topic we've touched on many times, on the podcast. So he says the status quo with models is I ask you a question and the model is expected to respond immediately with the right answer. That's an incredibly high burden to place on the model.
[00:30:58] Paul Roetzer: Like you couldn't do that to a [00:31:00] human. So again, system one thinking is I asked Mike a question. Mike replies immediately. System 2 thinking is, I ask Mike a question and Mike ponders it for a while. He says, let me think about that. And maybe it takes 30 seconds, maybe it takes 30 minutes, and then he comes back.
[00:31:13] Paul Roetzer: Well, what happened is, Mike thought through things. He went through a chain of thought, he applied reasoning and logic, he drew on his experiences, and then he came back and gave me an answer. That's what they're trying to do with these machines. They're trying to give them the ability, when they have a harder problem, To stop and think about it.
[00:31:30] Paul Roetzer: That's system two thinking. And so he said, um, that there's this very obvious next step for models, which is where you need to let them think and work through problems. and he says, right now there's no real true notion of problem solving. Now, the thing I thought was fascinating here, cause he was asked like, why is that so hard?
[00:31:49] Paul Roetzer: And he said, because they learn from data on the internet and the internet is full of outputs, not the actual reasoning process. So when you read something on the internet. [00:32:00] You're just seeing the final product. You don't see the work that went into the final product. So what they're having to do is actually train these models on the thought process that goes in to that final output.
[00:32:13] Paul Roetzer: So if they want to train it on an article, it's like, well, what did that person do to write the article? What process did they go through? How did they make these decisions? And he said, that stuff's not freely available. And so that's where they're investing a lot of their time is actually teaching the machine how to think.
[00:32:27] Paul Roetzer: Because there's no representative data of that on the internet, that led into the importance of data. So they asked him like, what have you changed your mind about in the last 12 months? And he said, it's the data. I underrated it dramatically. I thought it was just scale. So up until 12 months ago, he thought just build the bigger models.
[00:32:44] Paul Roetzer: Um, The other one I liked is he talked about voice. So we talked about like the advanced voice mode coming from OpenAI and how people aren't really ready for that. He said, anyone who has tried having a voice based conversation with one of these models, it's a stunning experience. You're [00:33:00] left in shock when you hear the model.
[00:33:02] Paul Roetzer: Exhibiting emotion and inflection, and you hear it breathe, before it speaks and you hear it lip smacking. So this is someone at the frontier who is telling you like, this is crazy stuff. And that's been my experience having seen it so far. And then the couple other ones, the models are getting much faster, and much smarter.
[00:33:23] Paul Roetzer: So he talks about how they are obviously becoming way, faster, a lot due to the data and the improvements in the algorithms, get into enterprise adoption. he said that they're really scared. So one of the reasons why these, you know, lack of adoption enterprises, they're scared people are going to take their data, train on it and put in some security vulnerability or that they lose their IP.
[00:33:44] Paul Roetzer: So it's a, it's a valid concern, but he's seeing a major shift. Last year was a lot of proof of concepts. This year, he's seeing an urgency to adopt. He touched on AI agent hype and he said it's 100 percent justified. These things are going to transform productivity and they're [00:34:00] coming. We've all been working on them for the last year to two years and we're going to see a massive change there.
[00:34:05] Paul Roetzer: And then the last two, what's coming soon that people are missing. He talked again about reasoning and planning. As the major thing that is, these things are going to be capable of. And then he touched on robotics, which we've talked about is the embedding of these models, this intelligence into humanoid robots and what that'll enable in the economy.
[00:34:23] Paul Roetzer: and he said, I think robotics is the place where there will be big breakthroughs, cost needs to come down, but it's happening. And then the models, that are there are much more robust. So someone's going to crack these general purpose humanoid robots. And then the final thing, which I thought was really interesting, cause we were all this, like, I want to pursue AGI.
[00:34:40] Paul Roetzer: We're going to solve intelligence. He's very practical. So he was asked like, where do you want it to go? And he said, I think that the world is super supply constrained and pretty much every luxury we have today. Has come from technology that is developed to increase productivity, boost the supply of things, make them more abundant, make them cheaper.[00:35:00]
[00:35:00] Paul Roetzer: And so what I really care about with this technology is driving productivity for the world and making humans more effective, able to do more. So that's, he's building the company focused on these smaller, verticalized models. They have open source models. They have proprietary models. They do on prem installations for enterprises.
[00:35:17] Paul Roetzer: They do cloud installations through the major cloud companies. So they're kind of playing all, but he's very focused on let's just augment humans. Let's drive productivity to create a world of abundance. So yeah, it's, I mean, Aidan's great. I love listening to his stuff. And like I said, I think it's a worthwhile listen for anybody who wants to understand the different approaches people are taking and the different leaders who are out there.
[00:35:43] Amazon AI Productivity
[00:35:43] Mike Kaput: Alrightthat's awesome. Let's dive into some rapid fire topics for this week. So first up, Amazon CEO Andy Jassy has shared some impressive results from the company's AI assistant, Amazon Q, in a recent post on LinkedIn. In it, [00:36:00] he talked about how this tool has dramatically improved efficiency in software development within the company.
[00:36:06] Mike Kaput: The company, he reported that tasks like upgrading applications to Java 17, which previously took something like 50 developer days, can now be done in just a few hours. And overall, this impact of Amazon Q on this type of work has been huge. Jassy says that it has saved the equivalent. of 4, 500 developer years of work.
[00:36:31] Mike Kaput: He also said the quality of the AI generated code is noteworthy. 79 percent of code reviews are being shipped without additional changes. That efficiency is translated into significant financial benefits. Jassy estimates 260 million in annualized efficiency gains through enhanced security and reduced infrastructure costs.
[00:36:56] Mike Kaput: So, looking ahead, Amazon is naturally planning to expand its [00:37:00] use of Q and trying to drive even more productivity for developers. This is particularly notable because, quite a while ago, we reported on this. Amazon Q had kind of a rocky start. It was facing a bunch of criticism internally for producing incorrect outputs.
[00:37:17] Mike Kaput: The company has addressed those initial challenges by expanding its team of human reviewers to fine tune the bot's performance, getting the results that we just talked about here. So, Paul, this is a pretty impressive Case study. Is this typical of the type of productivity gains we can start to expect when you apply, apply AI at scale to things like this?
[00:37:41] Paul Roetzer: think coding is sort of the leading edge here. So we've been hearing about. time savings through co pilot like GitHub, co pilot for, you know, the last couple of years, it was one of the first ways people started applying the Gen AI, language models. So it's not surprising that we're starting to see some of the greatest [00:38:00] returns from this.
[00:38:01] Paul Roetzer: But I think there's a bit of a moral of the story here goes back to the campaign's GPT ideas, find repetitive tasks and automate them. And you don't have to automate them a hundred percent. So in that article, it said 79 percent of code reviews being shipped without additional changes. Well, that means 21 percent are requiring human in the loop changes.
[00:38:20] Paul Roetzer: And that's okay. Like that's a pretty high error rate when it comes to coding. Like you wouldn't push code out the door where 21 percent of it was wrong. So. That's okay, though. Like, the whole point here is you're automating enough where you're saving all that time and money, but the humans are still required to be there, like, verifying the code is correct and fixing things that aren't.
[00:38:41] Paul Roetzer: Now, that code review rate is going to keep going up. Like, they may be able to eventually get that to 90, 95%, maybe even 99 percent accuracy. Still going to need the human in the loop, but it's now kind of beyond human capability. And so I think that paying attention to these gains in [00:39:00] coding is incredibly important because it is an, it is a, a leading indicator of how other tasks and professions will be affected.
[00:39:10] Paul Roetzer: And so coding is, there's massive money, it's an enormous market, lots of value to be created by improving coding capabilities. But you can look at this and say the same thing can now happen in in writing, in accounting, in legal industry. So as these models become better, as they become more precise, the error rate drops, and you can start to invest in building these kind of capabilities into all sorts of knowledge work professions.
[00:39:39] Paul Roetzer: Coding is just, you know, A leading example to monitor. the article also mentioned that, the, let's see, Amazon web services CEO, Matt Garman in June, suggested AI could soon dominate coding tasks, fundamentally changing what it means to be a developer by 2025. So, I mean, they're looking out ahead and the question [00:40:00] becomes like, what, what do organizations do with this?
[00:40:02] Paul Roetzer: Like Amazon didn't get in as far as I know, I looked at, the LinkedIn post and the, the Yahoo article. They didn't say what they're doing with all this time saving. I thought they're just making more product or are they gonna not need as many coders? Like, I don't know. But I think that's the other thing we watch for now is like what, when organizations can start having this kind of time saving.
[00:40:25] Paul Roetzer: That is a major shift in the implications to workforces. And again, that's the whole idea of JobsGPT and CampaignsGPT is take the key roles in your company, take the things that those people do and start to try and look ahead and say, okay, when these models are truly multimodal and they have the ability to code and create audio and video and all these things.
[00:40:46] Paul Roetzer: What changes, because those changes are like one to three years out in most industries and it's important we start looking at this. So yeah, it's a, it was a fascinating story and it's certainly some eye popping numbers.
[00:40:59] Mike Kaput: [00:41:00] Yeah, I also thought it was pretty notable that, like we talked about in the intro to this topic, people kind of dunked on Amazon Q a little bit internally when it first came out. Rightly so. Like, you get a tool, you're forced to use it, it's not working well. That's frustrating. But people often seem to use that as an excuse to write the whole thing off, and it's like, look where we are a year later.
[00:41:21] Mike Kaput: Some fine tuning. And we've suddenly achieved this kind of productivity game. That's impressive to me.
[00:41:27] Paul Roetzer: Yeah, and I think that's, we're going to see. You know, the same thing probably with these AI agents, you know, I keep stressing, they're just demos right now. They're really impressive demos, but it's not there yet. Um, but just wait, like it's, it doesn't take long before all of a sudden stuff just starts working.
[00:41:43] Paul Roetzer: And you're right, some people just write stuff off because it doesn't work today. And they forget how fast these models are improving through, you know, just pure scaling laws, but also through the hard work, the post training that happens after the fact. The bringing in experts, [00:42:00] PhDs, data scientists, domain experts, and having them work with the models to make them smarter in a specific domain.
[00:42:07] Paul Roetzer: So yeah, if you're not drilling in and realizing how fast this innovation is happening, you can, you can kind of get caught off guard for sure.
[00:42:16] Cursor AI
[00:42:16] Mike Kaput: So related to that topic, we also saw a hot AI startup just raised a 60 million Series A round to revolutionize how programmers code using AI. The company is called Cursor. It is a coding assistant that allows you to quickly and seamlessly code with help from AI. That includes the ability to use natural language prompts to create or edit Code.
[00:42:42] Mike Kaput: Cursor has been around since 2021 and it says it has over 30,000 customers, but it's been getting a bunch of rave reviews online and increased attention this week because of the funding announcement. The investors in the series A include some heavy hitters in the world of ai, [00:43:00] including Andreessen Horowitz, OpenAI, and Google's, Jeff Dean.
[00:43:05] Mike Kaput: So Paul, can you maybe break down for us, like, why are tools like this getting so much attention? Especially, I mean, given the productivity gains we're seeing internal tools like Amazon Q provide.
[00:43:19] Paul Roetzer: Yeah, so, u this one seemed to just take off. A little bit before the announcement of the 60 million because Andrej Karpathy tweeted about it. So Andrej Karpathy, we talked about, legendary AI researcher, ran AI at Tesla for five years, couple of stints at OpenAI. he tweeted a few days ago, programming is changing so fast.
[00:43:38] Paul Roetzer: I'm trying Cursor plus Claude Sonnet 3. 5 instead of GitHub Copilot again. And I think it's now a net win just empirically over the last few days. Most of my programming is now writing English. Prompting and then reviewing and editing the generated diffs and doing a bit of half coding where you write the first chunk and then you just keep basically tab, tab, tab, because it just keeps writing it for [00:44:00] you.
[00:44:00] Paul Roetzer: so imagine how like, you know, a few years ago, smart compose in Gmail, same deal, you just tab over as it finishes the sentence. It's like, oh, it just, it's doing it well. Like, I don't really need to say much more. Um, so people, I saw him get accused of like shilling for the brand. He's like, I got nothing to do with this brand.
[00:44:14] Paul Roetzer: I'm not invested in it. Like, it's just great. And so when you start seeing people like that, just objectively saying, this is changing how I code, and I'm not getting paid to say this, and I don't have a stake in the company, then people start paying, you know, close attention. And so, you know, I went in and kind of like looked at the Series A announcement.
[00:44:35] Paul Roetzer: And the thing I found fascinating is just kind of how they're positioning this. So they say our aim with cursor. is to continue to lead the shift by building a magical tool that will one day write all the world's code. It's a pretty ambitious vision. but like we've talked about, like, Repl. it's in a similar boat.
[00:44:50] Paul Roetzer: Like, everyone's trying to make the future of coding words. Like, you need no code. I, I mean, I built campaigns GPT and jobs GPT. I have zero coding ability. So, [00:45:00] Stuff I couldn't have built a year ago, six months ago, I can now build in a weekend with some knowledge and some words, basically. And so that's what their aim is.
[00:45:08] Paul Roetzer: Like, we just want everybody to be able to code, build whatever they envision with words. So they went on to say, already in Cursor, hours of hunting for the right primitives are being replaced by instant answers. Mechanical refractors are being reduced to single tabs, meaning again, like tab, tab, tab, it's, it's just doing it.
[00:45:24] Paul Roetzer: Terse directives are getting expanded into working source, and thousand line changes are ripping to life in seconds. Going forward, we hope Cursor will let you orchestrate AI powered background coders, view and modify systems in pseudocode, instantly scan your creations for any trace or bug, and much more.
[00:45:40] Paul Roetzer: Now again, I think of this as, okay, that's code. Now do this in HR, Legal, Finance, Marketing, Sales, Service, this same idea of just simple outputs that the AI is there assisting the humans still in the loop. But a year, two years from now, it's, it's going to be insanely [00:46:00] accurate with all the training and tuning that's going to go into these things.
[00:46:03] Paul Roetzer: So yeah, it's, it's wild. Now Mike and I aren't coders, so we can't go in and like personally use this and tell you how transformative it is. I'll take Andrej's word for it.
[00:46:11] Mike Kaput: That's usually a safe stance, isn't it? I'll take what he says is probably right.
[00:46:18] Procreate's Anti-AI Stance
[00:46:18] Mike Kaput: so our next topic is a popular iPad illustration app called Procreate has taken a bold stance against generative AI in creative software. In a recent announcement, the company said it will not incorporate Gen AI into its products.
[00:46:36] Mike Kaput: Procreate CEO James Kuda did not mince words in this announcement, stating, I really effing hate generative AI. He did not say effing. I don't like what's happening in the industry, and I don't like what it's doing to artists. He said that generative AI is, quote, ripping the humanity out of things, and steering the creative world's direction.
[00:46:57] Mike Kaput: world toward a, quote, barren future. [00:47:00] Now, this obviously stands in contrast to existing creative apps like Adobe's products, which have integrated AI features into all sorts of areas of their software. The company acknowledged this stance may make them an exception in the industry, but they think it is an exciting path forward for their community.
[00:47:20] Mike Kaput: So, Paul, this is certainly a bold stance. Is this A sustainable stance for any software business, much less a creative tool, in 2024 or beyond.
[00:47:31] Paul Roetzer: Well, I mean, I totally respect the position, you know, I think coming out with a very strong stance against it, that's great, like, that, you know, I think that people have that choice to make, Um, there, there's parts of this, like, I agree with them, like, I, I mean, I've said before, like, my daughter's original take on image generation, I thought was, Genius when she was 10, where she said it's just stealing people's imagination.
[00:47:54] Paul Roetzer: And, and I like that it's still probably the most profound thing anyone's ever said to me about generative AI is it [00:48:00] just goes and learns from our stuff, our creations, our imaginations, and then it just takes them and remixes them. And so I share his, in some ways, the spies for what it's doing. I think I just choose to be very realistic.
[00:48:17] Paul Roetzer: Like it's just. It's the future. It's like trying to move forward in business and saying, we're just not going to use the internet. Like, you know, and again, it's like all respect to them for that stance. And I understand where he's coming from. but you have to live with those beliefs and that generative AI is a very, very broad, All encompassing technology and to basically say you're a software company that's never going to use any components of generative AI is, is a really hard thing to adhere to when in the future, like everything's going to be powered by it.
[00:48:58] Paul Roetzer: So, I don't know. I [00:49:00] mean, I hope it works. I think maybe there is a niche for them in this world. They got a lot of love from their community for saying these things. We'll see how it plays out. I wouldn't, if I was an investor, I probably wouldn't invest in this company. Again, not cause I don't respect them or like, I think it might not be a winning play, but it's what they believe in and then it's what they need to do.
[00:49:22] Paul Roetzer: And all respect to them for doing that. But I think it's just going to be a hard thing. to stick. And I go back to that A16Z research we talked about up front. It's like 52 percent of companies on the web list are focused on content generator editing across modalities of the 12 percent are in the creative tool space.
[00:49:39] Paul Roetzer: Like it's going to be a tsunami of generative AI in this space. And I think it's going to be hard to compete if you don't offer the convenience and features that some of the gen AI components, like, again, I'm not saying go all in on gen AI, but To just say you're not going to have any of it is a tough, it's a tough, competitive [00:50:00] environment to win in.
[00:50:01] OpenAI partners with Condé Nast
[00:50:01] Mike Kaput: So next up, Condé Nast, which is the media giant behind publications like The New Yorker, Vogue, Vanity Fair, and Wired, has entered into a multi year partnership with OpenAI. Now, this deal allows OpenAI to use content from Condé Nast's portfolio in both ChatGPT and the new SearchGPT platform. As part of the agreement, OpenAI will be able to surface stories from Condé Nast outlets in its AI products.
[00:50:35] Mike Kaput: Condé Nast CEO Roger Lynch framed the deal as a way to quote, meet audiences where they are, unquote, while ensuring that there's proper attribution and compensation for the company's intellectual property as it's used in AI models. Specific terms of the deal, including the financials, have not yet been released.
[00:50:54] Mike Kaput: been disclosed. So Paul, this is not the first deal like this that we've seen. [00:51:00] It seems like licensing deals may be the future for some big publishers. Is that kind of how you look at a development like this?
[00:51:08] Paul Roetzer: The thing is, you know, we've talked about this a number of times, this is the only option. Like, what, what is the alternative? You as a publisher, you don't do the deal. Like I, it, the AI model companies have a lot of the leverage right now. And so the option is, you can sue us because we took your copyrighted material to train our models, or we can find a way to work together and build a new revenue stream for you as part of the next generation of search and information consumption and buying behavior.
[00:51:40] Paul Roetzer: And they don't need this data just to train on, they need it at inference. They need it in real time for search and, you know, outputs from the LLM. So It's tough. Like, I think it could work out for publishers, but, I, I just feel like the language model companies, specifically OpenAI and [00:52:00] probably Google and others, they hold a lot of the cards here and so it's, you know, I think they're trying to, put deals in place to get maybe some of the legal headaches out of the way down the road and to be able to make these models more real time, and more valuable to users.
[00:52:18] Paul Roetzer: So yeah, I think you'll just keep seeing deals like this done form
[00:52:21] OpenAI Names First Chief Communications Officer
[00:52:21] Mike Kaput: So some other OpenAI news. OpenAI has appointed Hannah Wong as its first Chief Communications Officer. Wong, who joined OpenAI in 2021 after a stint at Apple, will oversee a comprehensive range of comms functions including media relations, internal communications, strategic messaging, events, brand design, social media, and more.
[00:52:45] Mike Kaput: Now, she's well equipped for this. She was previously the company's head of public relations and the VP of communications. She's also reporting directly to CEO Sam Altman. So, there's no doubt, you know, as part of this, Wong will [00:53:00] be kind of typically tasked with keeping the company out of hot water from a PR perspective or spinning.
[00:53:05] Mike Kaput: News about the company, but another part of the job, specific to AI that they mention, is explaining the technology in a way that audiences can actually understand. This comes at a time when OpenAI has been quietly expanding its comms team according to Axios. It's gone from 8 members when ChatGPT was launched to more than 50 today.
[00:53:27] Mike Kaput: So, Paul, we've talked quite a bit about OpenAI having kind of its fair share of let's say like communications and PR issues in the last couple of years. Like, is this type of appointment meant to change that, get more diligent about how the company is communicating with the market and with consumers?
[00:53:47] Paul Roetzer: Yeah, I think, again, I don't have any, like, inside information here, but I spent 24 years in communications and marketing, and a few things seem pretty apparent here. So, one is definitely, they have a lot of [00:54:00] self inflicted wounds, You know, with the rollout, with talking about the training data, with, Um, product launches that haven't, you know, come to market as quick as they wanted.
[00:54:09] Paul Roetzer: There's just lots going on, and so it makes a ton of sense to sort of get someone in the role that can make them more strategic, keep people on message. The second is they're trying to play to the enterprise now. I don't know if I agree with Aidan Gomez that They've like given up on AGI and they're now just like a consumer play, but they're definitely trying to message more to enterprises.
[00:54:29] Paul Roetzer: And so you need more focus on that. And then the third is, for better or for worse, OpenAI is kind of the prominent voice for the AI industry. I mean, Google plays their, Anthropic has, you know, kind of a lesser, position, in terms of awareness. But OpenAI is. Like, kind of it when it comes to how people view AI overall and where we're going with AGI, like, they're a major, major player and they need to step up how [00:55:00] they're doing that and the messaging they have.
[00:55:02] Paul Roetzer: Um, plus I think there's, they're trying to play a greater role in laws and regulations and legislation and you need messaging. You need to. You know, be more strategic in how you're approaching that. And that's what communications leaders do. They think about those different stakeholders and what the messaging is to them and how to deliver that messaging and, how to monitor, you know, sentiment and, you know, they're, she's got her work cut out for her.
[00:55:26] Paul Roetzer: It's going to be a big job. staff of 50 helps, but you know, you're going to need a lot more than that. Probably.
[00:55:33] Anthropic and OpenAI SB 1047 Statements
[00:55:33] Mike Kaput: So, Anthropic and OpenAI have both released some statements related to California's controversial AI safety bill. We've talked about this a bunch. It's called SB 1047. First up, Anthropic CEO Dario Amodei penned a letter that says, while the company is happy that the state of California took some of its suggestions to improve this bill, The new version of SB 1047 [00:56:00] also lacks a bunch of suggestions that the firm made to the government.
[00:56:05] Mike Kaput: He writes in our assessment, the new SB 1047 is substantially improved to the point where we believe its benefits likely outweigh its costs. However, we are not certain of this and there are still some aspects of the bill which seem concerning or ambiguous to us. Now at the same time, OpenAI sent a letter.
[00:56:22] Mike Kaput: to California State Senator Scott Wiener's office opposing the bill. Wiener is a lead sponsor of the bill. OpenAI said it would hurt innovation and that regulation should come from the federal government. It also said the bill could have big implications for U. S. competitiveness in AI as a whole and on national terms.
[00:56:42] Mike Kaput: Security. As an interesting aside here, Bloomberg reports that the company has even put conversations on hold about expanding its San Francisco offices amid concerns about uncertainty in California's regulatory landscape. So Paul, how concerned are AI [00:57:00] companies getting about this bill? It seems like we're hearing an increased amount of rhetoric around it.
[00:57:04] Paul Roetzer: Obviously Barry, and Anthropics has been pretty involved, I think, in, you know, influencing the direction of the bill. yeah, I, I mean, obviously this is a really important topic. We, we touched on it a few times, you know, a month or two ago. and I think we've now talked about it like three straight episodes.
[00:57:22] Paul Roetzer: it's something that people need to be paying attention to. Obviously, if the big AI companies are, very vocally voicing their opinion and making those opinions public and starting to threaten a lack of expansion within California, like now we're starting to play the leverage game. So yeah, I don't know.
[00:57:41] Paul Roetzer: It'd be interesting to see how this all plays out. And I don't know what the timing is, like when we should expect decisions and, you know, go or not go. I think they've changed this bill like eight times already. So I don't, I don't know. Maybe for the next episode, we'll take a little bit of like, what exactly is the timeline on this thing?
[00:57:56] Paul Roetzer: Because it just feels like a lot of uncertainty and keeps evolving. [00:58:00] I did see one tweet, I forget who it was from, but they said like, had this bill been put in place like 12 months ago, I don't know. We would have already stopped the innovation because the current models we're doing would have been basically not allowed with where this bill was a year ago.
[00:58:15] Paul Roetzer: And so it's just this whole idea of like how, how little we know about the future models. And so we can just over regulate too quickly. When we think they're dangerous and yet, you know, nobody's worrying right now about GPT 4 as being overly dangerous to anybody, but a year or two ago, they might've thought anything bigger than this is like, we got problems.
[00:58:35] Paul Roetzer: So I don't know. So I said in the last episode, like I'm just sort of in the middle trying to understand both sides. I don't really have a strong opinion on, you know, whether the bill in its current form should be passed or not, but it definitely seems like there's still a lot of questions about it.
[00:58:50] Lindy AI Goes Rogue
[00:58:50] Mike Kaput: All right, so there's an AI startup out there. This is going to be good. There's an AI startup out there called Lindy, and it [00:59:00] allows customers to build AI assistants. And the company just shared a really bizarre incident of AI having a mind of its own. So here's what appears to have happened. Gone down, according to Lindy's CEO, who posted on X about this whole thing and showed a video about it.
[00:59:18] Mike Kaput: A customer requested video tutorials from Lindy the company via email. One of the company's AI assistants, which are called Lindy's, autonomously responded via email with a link to the company's video tutorials. So far, so good. The only problem is that the company does not have video tutorials. And the AI, instead of saying, Hey, we don't have tutorials, instead provided what looked like a real link to a video tutorial, that instead went to the iconic Rick Astley 1987 hit, Never Gonna Give You Up.
[00:59:57] Mike Kaput: Which is a classic internet prank known as [01:00:00] Rickrolling. Lindy's CEO, Flo Crivello, posted the following about the incident. Quote, a customer reached out asking for video tutorials. We obviously have a Lindy handling this and I was delighted to see that she sent a video. But then I remembered we don't have a video tutorial and realized Lindy is literally effing Rickrolling our customers.
[01:00:21] Mike Kaput: Crivella theorized that the behavior could have emerged from how the AI makes predictions. It may have predicted that when it's prompted to send a video, YouTube was the next logical step, followed by another common thing that happens on YouTube, which is people sending Rick Astley videos. So Paul, this is like a hilarious and, you know, relatively harmless way for AI to go wrong, but is this like, Kind of autonomy, something we need to increasingly worry about as we kind of see agents and more autonomous AI tools take the reins.
[01:00:56] Paul Roetzer: Yeah, I mean, it is hilarious. And by the way, my kids apparently know [01:01:00] what Rickrolling is. So it's still a thing,
[01:01:02] Mike Kaput: That's good. Okay.
[01:01:03] Paul Roetzer: to me after all these years. but yeah, I mean, this is the point of the human in the loop. Like, and I get asked a lot about like Gen AI and customer, service, customer experience use cases.
[01:01:16] Paul Roetzer: And I always tell people just be careful. Like if you're. Allowing the chatbot to interface or send emails or reply to people's chats and text messages. They're not 100 percent accurate. They don't do what they're supposed to do all the time. And sometimes they come up with creative solutions that aren't the thing you're supposed to be doing.
[01:01:35] Paul Roetzer: So, I do think it is just, again, a reminder in a very funny way. that these things don't function like normal software. They don't just do what you tell them to do. Sometimes they do other things and in this case it's funny and their customers probably got a good laugh out of it, but in other cases it can be far more damaging and lead to safety and security concerns which sort of is a [01:02:00] segue to our next rapid fire topic.
[01:02:03] Mike Kaput: Indeed, and I would just say also, too, like, I give Lindy a lot of credit for having a sense of humor as a company, but I could see, I don't know, let me know if you agree. There are definitely enterprises we've talked to where even this funny thing happening would get the whole thing shut down immediately.
[01:02:17] Paul Roetzer: enterprises we have worked with, it's, it's a non starter
[01:02:20] Microsoft Copilot Safety and Security Concerns
[01:02:20] Mike Kaput: Yeah. All right. So to your point. The segue here is that a recent report now suggests that some large enterprises are hitting the pause button on Microsoft Copilot due to significant data governance and security concerns. So this comes from a guy named Jack Berkowitz, chief data officer at a company called Security, who shared his findings about this with a publication called The Register.
[01:02:47] Mike Kaput: So Berkowitz talked to a bunch of chief data officers at major companies, and he said that approximately half of those chief data officers he surveyed have either completely turned off Copilot or [01:03:00] severely restricted its use within their organization. It seems the biggest issue stems from the AI's ability to access and summarize information That employees might technically have access to, but should not, such as salary information or other sensitive data.
[01:03:17] Mike Kaput: This gets exacerbated as a problem in companies that have a bunch of commission structures in Microsoft 365, and those conflicting authorizations can lead to inappropriate data exposure through co pilot interactions. Now, Berkowitz emphasized this is not an unsolvable issue, but it seems to require cleaner data and more robust security measures to do kind of out of the box what you're hoping Copilot does for your company.
[01:03:45] Mike Kaput: So Paul, this definitely in one way or another broadly seems like an issue we're hearing about from enterprises, where even when these tools that they know. that they have access to, that they think they can trust, like there are still these [01:04:00] unresolved issues. Are you seeing this when you talk to leaders about either co pilot specifically or tools broadly?
[01:04:07] Paul Roetzer: Yeah, these are very valid concerns, you know, if you're in an organization and you're giving this thing access to all your knowledge base, and within that knowledge base is highly segmented confidential information that you don't want the average employee chatbot about saying, how much is my boss making and things like that.
[01:04:26] Paul Roetzer: Then there's a, there's a lot more science that goes into how you roll these things out. It's not as simple as just getting a license and turning it on. And I think this is a big part why you see slow adoption in enterprises. Either just, we're not going to do this until we solve it completely, or we're only going to roll this out to these 20 people who have all these data privileges.
[01:04:47] Paul Roetzer: It's not an easy thing in the enterprise. And so this is, I think, why these teams are starting to build. You know, these companies are trying to build out these enterprise teams to drive education, onboarding, sales, because there's a [01:05:00] lot that goes into the adoption other than just buying the licenses. And this is a good reminder of that.
[01:05:05] LinkedIn AI Disclaimers
[01:05:05] Mike Kaput: All right, our last topic this week. On last week's episode, we covered our thoughts on AI in schools, and that was inspired by a LinkedIn post that Paul, you had written about that. But what we didn't cover at that time was another interesting aspect of the post that you published on LinkedIn. It was automatically labeled by LinkedIn using the content credentials system, which basically indicated the image, Paul, that you shared with the post was generated with AI, which it was and was intentionally and obviously meant to be generated with AI.
[01:05:39] Mike Kaput: Yeah, right, right. So, Content Credentials basically applies a small label with the letter CR within a bubble that you can click on to see metadata about where an image came from and how it was made. Now, importantly, this is not LinkedIn itself, Natively being able to detect and label AI generated [01:06:00] content, the company says, quote, it's not yet possible to do that.
[01:06:03] Mike Kaput: Instead, content credentials are displaying existing metadata applied to content from something called C2PA, and
[01:06:12] Mike Kaput: Content Provenance and Authenticity. This is a project that enables consumers to trace the source and authenticity of media content. C2PA is being used by some major AI companies to label the content created by their AI tools, so that when you do go post it somewhere, it may show up with this content credentials watermark.
[01:06:36] Mike Kaput: This actually includes OpenAI, which attaches this metadata to images created in ChatGPT. So Paul, like, how likely are we to start seeing things like content credentials appear on AI generated content moving forward?
[01:06:53] Paul Roetzer: I expect it'll become much more prominent, especially on the social media channels. I think they're going to put it under a lot of pressure to identify stuff [01:07:00] that's not Real? That was, you know, generative AI. Yeah, I had no idea what it was. you know, I was obviously familiar with the organization, but I literally like looked at the image on my post.
[01:07:10] Paul Roetzer: We'll put the link to my post in if you haven't seen this yet. We'll put the link to the original post and go check it out. But yeah, there's this little bubble in the top left and it's lowercase C R and I sent it to Mike. I was like, Oh man, I never seen this. Have you ever seen this before? And yeah, you click on it and it just says issued by Web Claims Signing CA, issued to OpenAI, issued on January 30th, 2024.
[01:07:32] Paul Roetzer: That's it. It's just like, Oh, okay. Like this is obviously AI generated. so yeah, it just kind of caught me by surprise. And then when I went and looked into it was the article on content credentials from LinkedIn was from. Two weeks ago at that time. So I think it's a relatively new thing that they're doing.
[01:07:47] Paul Roetzer: So if you're seeing it, that's, that's what it is. And we'll put the link to the LinkedIn page that explains content credentials and you can go check it out, but yeah, just something to kind of be aware of. If you start seeing it pop up on images of your own or on other people's images, that's where it's [01:08:00] coming from.
[01:08:01] Mike Kaput: Alright, Paul, that's another week in AI wrapped up here. just a couple final quick notes here. If you have not subscribed to our newsletter at MarketingAIInstitute.
[01:08:10] Mike Kaput: marketingainstitute.com/newsletter, we cover all these stories more in depth and all the stories we didn't get to on this episode. And last but not least, if you have not yet left us a review and you are able to leave us a review on your podcast platform of choice, it would be very helpful to help us improve the show and reach more listeners.
[01:08:32] Mike Kaput: Until next week, Paul, thank you so much for wrapping up all the critical news happening this week in artificial
[01:08:41] Paul Roetzer: I'll add two quick notes. One, if you have subscribed on SmarterX. ai to the Executive AI Insider Newsletter, it is coming soon. We actually have, we have, the format locked in. We've got issue one, ready to roll. And it's either going to launch this coming weekend or the weekend after, I think is the plan.
[01:08:59] Paul Roetzer: So it's [01:09:00] going to start coming on Sundays and I'm going to write it basically every Friday and I'll send it on Sunday. So if you don't know what I'm talking about, go to smarterx. ai. It's a new newsletter being written for leaders to try and keep them on track. in the loop and kind of leading AI literacy and transformation in their organizations.
[01:09:15] Paul Roetzer: And then the second thing is get those MAICON tickets. Mike and I would love to see you in person. Go to maicon. ai. It's MAICON.ai. Use that pod 200 promo code and get 200 off any pass. And hopefully we'll see you in 14 days in Cleveland. Thanks, Mike. We'll talk to you all next week. Oh, final note.
[01:09:33] Paul Roetzer: We're actually, Labor Day next week. Right, Mike? So we're going to drop our episode on Wednesday, the, what is that? The 4th, I think we decided. yeah, Wednesday, September 4th will be the next episode. So we're going to do a one day delay, come out Wednesday, September 4th. So, if you're watching for the podcast on Tuesday, the 3rd, you'll be disappointed.
[01:09:54] Paul Roetzer: You'll be really happy to find it on Wednesday, September 4th though. All right. Thanks everyone.
[01:09:59] Paul Roetzer: [01:10:00] 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.
[01:10:22] Paul Roetzer: Until next time, stay curious and explore AI.
Claire Prudhomme
Claire Prudhomme is the Marketing Manager of Media and Content at the Marketing AI Institute. With a background in content marketing, video production and a deep interest in AI public policy, Claire brings a broad skill set to her role. Claire combines her skills, passion for storytelling, and dedication to lifelong learning to drive the Marketing AI Institute's mission forward.