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

50 Min Read

[The Marketing AI Show Episode 55]: AI-Powered Content Strategy, Claude 2 from Anthropic, and Major Google Bard Updates

Featured Image

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

Learn More

Aside from Google Bard major updates and a long list of other AI developments in the news, Paul and Mike break down many of this week’s happenings and how they apply to use cases at the Marketing AI Institute, specifically our podcast as well as our upcoming Marketing AI Conference (MAICON).

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

This episode is brought to you by Jasper, On-brand AI content wherever you create.

Listen Now

Watch the Video


00:04:46 — AI-powered content strategy

00:26:08 — Claude 2

00:36:36 — Google Bard updates

00:40:59 — Cassie Kozyrkov and MAICON 2023

00:42:28 — Elon Musk and xAI

00:45:38 — Sarah Silverman OpenAI Copyright Lawsuit

00:49:41 — Google and robots.txt

00:51:19 — ChatGPT disables “Browse with Bing”

00:53:06 — DALL-E and Shutterstock

00:55:33 — NotebookLM

00:57:08 — Shopify Sidekick

00:59:07 — AP + OpenAI agreement

00:46:46 — FTC and OpenAI

01:00:45 — Hollywood and AI replicas

01:03:42 — Hugging Face and fresh VC funds


AI-Powered Content Strategy

In exciting news, this podcast, The Marketing AI Show, recently surpassed 100,000 downloads in 2023 alone. That’s exciting and a huge milestone for us, but the story is the content strategy…and AI’s role in it. Prior to October 2022, the podcast format was interview-style, where Paul was vetting experts, scheduling interviews, and more, causing inconsistency. This led to the podcast being an afterthought in the marketing strategy. Then in October, we put the weekly podcast at the center of our content strategy.

Today, each episode averages more than 4,500 downloads. To put that in perspective, we had 4,800 downloads the entire year of 2022! Plus, we’ve heard from listeners all over the world about how the podcast has impacted them, and how they’ve come to rely on it for their weekly AI news. In this podcast episode, we’re going to talk a bit about how the podcast became key to our content strategy and how AI makes it possible to efficiently produce high-quality episodes that perform.

Claude 2 from Anthropic

Major AI company Anthropic just announced the release of Claude 2, its new foundational AI model. Similar to ChatGPT, you can engage with Claude 1 and now Claude 2 through natural language prompts to perform a range of functions, like generating text, answering questions, and producing code. Claude 2 has improved reasoning from Claude 1, and even scores above the 90th percentile on the GRE reading and writing exam. Some of the changes are impressive, including most notably, the number of tokens you can input into Claude 2 is massive. You can input up to 100,000 tokens into a single prompt, which, says Anthropic, “means that Claude can work over hundreds of pages of technical documentation or even a book.” Is 100,000 a lot? Yes…The Great Gatsby would be about 72,000 tokens! There are also so many other great benefits and enhancements. Paul and Mike break them down in the 2nd story of the podcast.

Major Google Bard updates

Google just announced a ton of new features for Bard, its ChatGPT-like conversational agent. Bard is now available in 40 languages and many more countries than before, including the European Union. Some of the other major updates include changing the tone and style of Bard's responses to five different options, you can drop images into it and have Bard perform a range of tasks related to the image, you can share Bard’s responses with others via shareable links, and more. Tune in to the podcast to hear more of the updates both from a functionality as well as a UX standpoint.

We cover a lot of ground this week, and per usual, it was another big week in AI news stories, so be sure to tune in! The Marketing AI Show can be found on your favorite podcast player and be sure to explore the links below.

Links Referenced in the Show


Read the Transcription

Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.

[00:00:00] Paul Roetzer: What I would say is have your like three to five standard use cases that you want to use these tools for. Transcription, summarization, drafting articles, and every month or two go in those across multiple models because they're all going to keep moving ahead of each other at different times for different capabilities and you really need to stay testing them.

[00:00:20] Paul Roetzer: Welcome to the Marketing AI Show, the podcast that helps your business grow smarter by making artificial intelligence approachable and actionable. You'll hear from top authors, entrepreneurs, researchers, and executives as they share case studies, strategies, and technologies that have the power to transform your business and your career.

[00:00:41] Paul Roetzer: My name is Paul Roetzer. I'm the founder of Marketing AI Institute, and I'm your host.

[00:00:50] Paul Roetzer: Welcome to episode 55 of the Marketing AI Show. I'm your host, Paul Roetzer, along with my co-host as always, Mike Kaput. What's up, Mike? How's it going, Paul? Good. We are doing this on a short week, so anyone who listens regularly knows in episode 54, I realized as we were recording it that I am off the week of, I don't even know what day it is right now, the week of the 17th.

[00:01:14] Paul Roetzer: So we are actually recording this on Friday, July 14th. Normally record on Mondays. And so my concern was we weren't going to have enough to talk about on only a four day gap between recordings. Little did I know that the world of AI would be crazy this week. I should have expected it, but we have plenty to talk about today.

[00:01:32] Paul Roetzer: We're going to move pretty quick, as usual through our three main topics. And then our rapid fire, we actually had to cut some rapid fires because it just gets to be too much sometimes. So, today's episode is brought to us by Jasper, the generative AI platform that is transforming marketing content creation for teams and businesses.

[00:01:51] Paul Roetzer: Unlike other AI solutions, Jasper leverages the best cross section of models and can be trained on your brand voice for greater reliability and brand control. With features like brand voice and campaigns, it offers efficiency with cons, offers efficiency with consistency that's critical to maintaining a cohesive brand.

[00:02:11] Paul Roetzer: Jasper has won the trust of more than 100,000 customers, including Canva, Intel, DocuSign, CB Insights, sports Illustrated, and Marketing AI Institute. Jasper works anywhere with extensions, integrations and APIs that enable on-brand content acceleration on the go. Sign up free or book a custom demo with an AI expert@jasper.ai and.

[00:02:36] Paul Roetzer: Since MAICON is officially when you're listening to this, it will be next week, July 26th, the 28th. It's also brought to you by the Marketing AI Conference. You can still get in MAICON.ai - M A I C O N dot A I. You can use AIPOD100 to get a hundred dollars off. I teased a major keynote announcement in last week's episode.

[00:02:58] Paul Roetzer: We since announced that Cassie Kozyrkov, the Chief Decision scientist at Google, is going to be joining us to open day two. Whose job does AI automate? I can't wait for that talk. And then last week I kind of featured some of the main stage stuff. And if you follow me on LinkedIn, I did a whole post about the main stage talks, but we've got some amazing breakouts as well.

[00:03:20] Paul Roetzer: I think there's almost 30 sessions total, including a workshop from me and a workshop from Mike. We've got AI compliance and all the insights that prompt away from Tim Hayden, great friend of Marketing Institute and a longtime speaker. We have what marketers should know about trademarks, copyrights, and generative AI from an a wonderful IP attorney Sharon Toerek, also a Cleveland.

[00:03:40] Paul Roetzer: We've got Stephan Britton from Shutterstock doing art in the age of ai, decoding technology protecting artists, and championing diversity. I'm intrigued by that one. We've got prompt engineering 1 0 1 for marketers from Jim Sterne, another longtime friend and supporter of the institute, and then Mike is going to be doing the always popular tools Talk 45 AI tools in 45 minutes.

[00:04:01] Paul Roetzer: Dude, that's going to be a lot. That's a lot to cover. 45 minutes. Yeah, I was building, there's no

[00:04:05] Mike Kaput: intro. Yeah, I was building the deck the other day and was like, I have to hustle

[00:04:09] Paul Roetzer: on this one. Yeah, you like, I should have gone with 30 45 is a lot. So, yeah, we'd love to have you again, MAICON is still trending towards a sellout.

[00:04:19] Paul Roetzer: It's going to be close, but there's time to get tickets now. Hopefully when you're listening to this, there's still tickets left, so check that out@MAICON.ai, Jasper will be there. So again, thanks to our sponsor, this show, Jasper. They will be there. Megan Keaney Anderson's doing one of the featured talks on the main stage and they'll be in the, sponsor hall.

[00:04:36] Paul Roetzer: So, with that, let's get onto the show. Mike.

[00:04:40] Mike Kaput: Excellent, Paul. So we are going to cover three big topics today in AI and then jump into a bunch of rapid fire. Our first topic today is about AI powered content strategies, specifically our AI powered content strategy. We're very happy to report that this podcast, the marketing AI show recently surpassed a hundred thousand downloads.

[00:05:02] Mike Kaput: Now that is a really significant milestone for us, but it would've been pretty inconceivable a year ago, and that's because a year ago the format was more designed around Paul. You were interviewing AI experts and with your schedule, you know, we can never really get into a consistent

[00:05:19] Paul Roetzer: flow, and that's a very nice way of saying, I never did my job and did this podcast.

[00:05:24] Mike Kaput: You could be forgiven for that though, because honestly like this whole process was really, really time consuming. And honestly, you know, we all kind of, I think, agree the podcast at that time. It was more of an afterthought in our marketing strategy, but that actually all changed back in October of 2022.

[00:05:42] Mike Kaput: And we made a shift that put the new weekly podcast at the center of our content strategy. So today we're really pleased to report that each episode averages now more than 4,500 downloads. And, this is a little tough to read, but to put that in perspective, we had 4,800 downloads, all of 2022. So I'm very, very pleased with the growth.

[00:06:06] Mike Kaput: But yeah, just shows you how far we've come. And you know what's really, really cool is we've heard from listeners all over the world about how the podcast has impacted them and how they've actually come to rely on it for their weekly AI news Now. Today we wanted to spend a few minutes kind of almost as this real world case study, using our own business of talking about how the podcast became key to our content strategy, and specifically how AI makes it possible to efficiently produce high quality episodes that perform really, really well.

[00:06:37] Mike Kaput: So maybe Paul, we could kick things off as you know, a discussion about what is the kind of story behind how the podcast became key to our content strategy.

[00:06:48] Paul Roetzer: Yeah. It really is kind of a happy accident, to be quite honest. So, again, just to hit you with some numbers, we're at 103,000 downloads, year to date in 2023 with 27 episodes.

[00:07:01] Paul Roetzer: In 2021, we had 1600 downloads the entire year in 2022, it was about 4,800 downloads the entire year. Prior to this year, we had done 28 episodes total. So, we made a big commitment to the podcast. The average, or the highest, 24 hour download was actually episode 54, the previous episode at 1800 downloads.

[00:07:26] Paul Roetzer: Our most, popular over a seven day period is actually episode 52 with the 15 AI questions everyone is asking that had a seven day record of 3,442. And to date, the most popular episode overall is episode 40, which was ChatGPT plugins is GPT early a g i, and using AI to do a full product launch in 30 minutes, which was actually talking about Ethan Mollick, who's one of the keynotes at MAICON.

[00:07:52] Paul Roetzer: So that one has 5,100 downloads, so it's actually, you know, very consistent. We don't have some crazy outlier that got 60,000 of the a hundred thousand. These things are all in the four to 5,000 range. And so you know how it all started. Again, it came from a frustration where there was just a lack of consistency.

[00:08:10] Paul Roetzer: The original idea for this podcast when we started in, I think 2019, before the first marketing AI conference, was for me to interview experts. And so I was going to once a week, get somebody interesting, do an interview, do prep, and so realistically, it was going to take probably like five to seven hours of my time every week to pull this off.

[00:08:28] Paul Roetzer: And then we had an outside production team. We didn't really have the support internally to really manage this thing. And those are a lot of the reasons why it just wasn't getting off the ground. It was just too much of a roadblock because of the commitment needed to do it. So then in early spring, I think it was like February, actually, maybe winter 2022.

[00:08:46] Paul Roetzer: I think I'd proposed to Mike, Hey, what if we did like a weekly format and we just picked some topics and we do this interview. And so throughout 2022, we bounced it around. It would show up on the weekly huddle each week and just kept pushing it out like we're not going to do it. And then we finally committed in October of 22 to do this.

[00:09:03] Paul Roetzer: And part of the motivation was, a big part of the motivation was our website traffic was largely flat. So Mike, man, Mike as our chief content officer, he manages the content. He was pumping out all kinds of content we were having following this republishing strategy. We were publishing new content, but we weren't doing the trend stuff.

[00:09:19] Paul Roetzer: We weren't, we didn't have the resources internally to stay on top of all the funding news, all the big AI news. And that was the stuff I always wanted to do. Like I always wanted to be kind of more of a, a center for all the relevant AI news. And there was just no way Mike was going to be able to write 10 articles a week.

[00:09:37] Paul Roetzer: And we weren't planning on hiring to, you know, fill that. So the change in format was actually a content strategy change. It was like, how do we get fresh content on the site? How do we update the blog more regularly with fresh stuff, keep up on the news and the funding, and how do we then take that and amplify that out?

[00:09:55] Paul Roetzer: So we weren't even looking at podcast metrics in October of last year. I mean, we had it, we had megaphone and whatever else, and we had access to the numbers. But I had, honest to God, never even looked at the numbers. I had no idea how many downloads we had. I didn't even honestly know first 24 hours and first seven days were the two KPIs we should be paying attention to.

[00:10:12] Paul Roetzer: I knew nothing about podcast metrics. So the idea was, let's, let's make this switch. Let's go to this weekly format. We'll do three main topics and then we'll do rapid fire. Each podcast, Mike then would turn into a summary blog post. So you'd have one blog post that's like the total summary with the transcript and the timestamps and all that.

[00:10:32] Paul Roetzer: And then each of the main topics would get its own blog posts. That's four blog posts. Each of those would also become a YouTube video. So wed get four YouTube videos, and then we'd get dozens of so social shares. So Cathy McPhillips on our team, our Chief Growth Officer, she heads up that part of it. So Mike takes this, turns this into the content through transcriptions, and he'll kind of walk, maybe walk through some of the ways we're using AI in all of this.

[00:10:54] Paul Roetzer: But basically what happened is we just said, let's just use the podcast as the center of everything. So instead of an afterthought, let's just make it the center of everything, but not as a growth channel and not as like, you know, we're going to stay on these metrics every week. We're just going to do it to enhance the blog, get a YouTube strategy rolling and see if we can't do that.

[00:11:12] Paul Roetzer: I never, in October, considered that the podcast would become our fastest growing platform. So, you know, that's kind of how this came to be. And then it just sort of started taking off. So, Mike, why don't you kind of give us a little overview of how we're using AI in the process. We can kind of mix this story together here about a marketing strategy, content play, plus the AI component to it.

[00:11:34] Paul Roetzer: Yeah, for sure.

[00:11:35] Mike Kaput: And Ithink it's important to note, yeah, this is not strictly AI making all of this possible, that that strategy change is the key to it all. But I would also say that if we were doing, and we were at the beginning, executing that strategy fully, manually, it would've taken us easily. 2, 3, 4 x the amount of time it probably takes us today, that's because we are using AI in a number of ways.

[00:12:00] Mike Kaput: None of these, I would argue, are rocket science or extremely, you know, sexy use cases, but they save a ton of time and free up a lot of bandwidth to focus on what actually matters in this process. So first off, we use it for research. So we're throughout the week manually curating links, following AI developments.

[00:12:19] Mike Kaput: We have an ongoing Zoom thread, so that's really helpful. But even within that, some of the more complex topics are things I may be less familiar with. I do a little research on and things like ChatGPT with the browser plugin enabled. At least formerly until, because one of our other topics is they've temporarily suspended it.

[00:12:38] Mike Kaput: There's another tool that I love called perplexity ai, which is basically just ChatGPT connected to the internet and gives you actual sources for what it. She is responding with, so I can get really quick summaries on different topics or even terms I'm unfamiliar with, perhaps if it's a new development.

[00:12:55] Mike Kaput: That's super, super helpful. That's much, much better than just randomly Googling things like in the past. We also use, Chachi PT and things like Claude, which we'll actually talk about a little later. Claude from an Thropic, which is, Chachi PT like tool as well. To summarize links that we collect each week.

[00:13:13] Mike Kaput: So we're dealing with 20, 30 links as I'm sure anyone has seen in our show notes. I still personally, to create value for the audience, like read through every single one manually, but it is helpful to just summarize, especially if we're a little short on time as well. So I'll use, those types of tools to summarize stories.

[00:13:33] Mike Kaput: In some cases not all. I will actually use AI writing tools to. Take those story summaries and write scripts that you hear on this podcast. I almost always tweak them for my own voice. And sometimes AI misses what I would say editorially is a really important point. So I'll drop those in. But especially for really straightforward summaries, it is kind of helpful.

[00:13:57] Mike Kaput: I mean, it doesn't take a ton of time to write a script, but if I can just save a little time on a few summary scripts, I can focus more on fleshing out a more complex

[00:14:06] Paul Roetzer: topic. And real quick, so that's like kind of the pre episode stuff? Yes, that happens. So what we'll do is like, we'll, Mike, and I'll add those links throughout the week to that Zoom thread, just like we just do it, start a new episode, sandbox each week, and then we just throughout the week, drop 'em in there.

[00:14:23] Paul Roetzer: And then Mike, usually like Sunday night, Monday morning, we'll do his part where he is building the whole brief, and then usually on about 30 minutes before we go on, sometimes I'll do it earlier than that, but usually it's about 30 minutes before we go on. I'll actually come in and read through everything.

[00:14:39] Paul Roetzer: Mike's dropped in there, and then sometimes I'll add some notes. A lot of times I just flow with it. It's just like, I'll just look and see what we're talking about and then we just get on and go. Yeah. So for me, again, if you're thinking about. Playing this kind of strategy in your company and maybe you've got a C e O or the person who you know is really the, the point of view from the organization.

[00:14:58] Paul Roetzer: They're the one that's out in the forefront of this stuff. And you wanted to do something like we're doing here. I honest to God, spend about 30 minutes a week on this other than live real, real time reading of things and dropping links in. Cause I'm just consuming that stuff all day long anyway. And then we record this for an hour and then I'm out, like that's it.

[00:15:16] Paul Roetzer: So it's an hour and a half of my time, roughly a week to do this thing that has become the core of our content strategy. And then Mike's probably putting in. Would you say, Mike, for the part we just, you just covered, what do you got like two hours,

[00:15:29] Mike Kaput: three hours prep time? Yeah, I'd say probably three to four hours, max.

[00:15:32] Mike Kaput: And that's really too, like do I, you know, like I'm saying, with some of these summaries, I still prefer to read through everything just from my home. I have to development hundred percent

[00:15:41] Paul Roetzer: the same. Yeah. I have to read 'em myself or I just don't comprehend it the same. So to be

[00:15:45] Mike Kaput: honest, if you're a brand and maybe with maybe following some easier to process developments, you could very easily cut this time down.

[00:15:53] Mike Kaput: But I'd say probably three or four hours. Yeah. Okay. But even that, given what we get out of this is like no time at all in

[00:15:59] my

[00:15:59] Paul Roetzer: opinion. Percent. Yeah. Cause you used to spend, you know, two, three hours plus per blog post writing. Post post. Yeah. Yeah. So just, just if we extract, play that out and said four videos, four podcast or four blog posts.

[00:16:11] Paul Roetzer: All those socials on their own. Just the promo, like the promotion piece of this right, would be 12 to 15 hours easily. Yeah. Okay. Keep going.

[00:16:20] Mike Kaput: Cool. So, you know, once that's all like Paul mentioned in advance and then we record the podcast, and I should mention really quick when I talk about script writing here, none of our like Paul's responses and things here are scripted at all.

[00:16:33] Mike Kaput: I'm talking about the script where introduce topics and even that is more guidelines than anything. So it's mostly a organic off-the-cuff conversation. But once we record the podcast, we then use a tool called script to transcribe the podcast episode. We then also use script, our colleague Cathy, who everyone I believe in our audience knows or has heard of, She does essentially professional grade editing of the audio and video of each episode despite not having, you know, a pro video and audio editing background.

[00:17:06] Mike Kaput: What's really cool about Script, it just gives you this pro grade capability in such an easy to use, ux thanks to the power of ai. And they're also rolling out really cool advanced AI features. There's one that they just announced. We haven't used it yet, but I'm sure we will in the past or in the future, rather called Regenerate, which.

[00:17:26] Mike Kaput: Will actually allow you to change how someone's delivery sounds for a different, a certain line. Like if someone starts trailing off on an important point, you could actually go in and literally edit the text transcript and it will change their voice and their tonality to actually sound much more natural and organic and engaging, which I found jaw dropping.

[00:17:47] Mike Kaput: I watched the demo again this morning,

[00:17:49] Paul Roetzer: actually, and quick context here. We used to use an outside production team to yes, produce these things. So once we recorded, we would just send it to them and Iwant to say it was roughly like $600 a month or something like that for four episodes, 600 to a thousand a month.

[00:18:04] Paul Roetzer: So we don't need that anymore. Like we just do it ourselves. For the $30 a month, we pay for script, basically, and cap.

[00:18:11] Mike Kaput: Yep. Yeah. And I, you know, I don't want to, I'm sure this takes quite a bit of work on Cathy's part, but I think in the grand scheme of things, it's relatively efficient to be doing this versus kind of the traditional ways we would've been doing audio and video editing.

[00:18:26] Mike Kaput: Yeah. Once we have all of the audio, video and transcript, I'll actually use AI writing tools often to rewrite or rephrase some of the scripts that we use in the introductions here of topics to actually become the intros of blog posts for each main topic, because we'll spin off a blog post ideally on each of the main three topics, given how noteworthy they are to the community and to the industry.

[00:18:52] Mike Kaput: So that alone already you have an introduction, a really sound introduction that's compelling to a blog post that I haven't even written yet. We will then use things like Chachi PT and Claude to summarize each section of the transcript on each main topic. I'll often have it put the summary in bullet form.

[00:19:11] Mike Kaput: Tell me, can you find me the main takeaways from this conversation? Usually if it's, you know, a 5, 7, 8 minute segment, it usually fits into ChatGPT. It for sure will fit into Claude too now that the token limit is so high, and we'll talk about that as a rapid fire top, or rather a main topic. So then once I get that summary, that makes it very, very easy for me based on the context I have from our conversation, the research I've done to prepare to actually go in and layer in really useful insights.

[00:19:41] Mike Kaput: So I kind of call it in my head, connecting the dots, right where I'll take what you've said, Paul, what we've talked about, and really try to flesh it out and create actionable takeaways for our audience. So it's not just summarizing what we talked about on the podcast, it's using that as a springboard to create a huge amount of value in a short, you know, a relatively short post and a short amount of time for the audience.

[00:20:06] Mike Kaput: So really, when I write a blog post these days, I am spending all of that time, which is not much time at all. Really just connecting the dots of the material we already have, and that has dramatically cut down the time it takes to produce a 5, 6, 700 word post. That is actually, I would argue, a lot more valuable usually than kind of just like a really surface level informative post.

[00:20:29] Mike Kaput: This is insight, and inspiration that you cannot, like, cannot be duplicated by, say, someone in our space using just an AI writing tool. This is informed by our own experiences,

[00:20:40] Paul Roetzer: right? Unique point of view, which again, we go back to, you know, I always heard me talk recently, I have this, you know, authentic human content wins theory that like, once we can all create a, any AI content we want on any topic.

[00:20:51] Paul Roetzer: It's the people who have unique perspectives and share those through podcasts and videos and live events, even though we're using AI tools to then summarize it and turn it into a post it's originating from a unique perspective that no one else provides. Yes. And so, yeah, I mean this, so we go back to why did we do this strategy back in October?

[00:21:08] Paul Roetzer: It was for fresh, unique, timely content from our point of view, amplify it across YouTube and the blog and you know, podcast networks, and that was our goal. It wasn't to get to a hundred thousand downloads by middle of July, 2023. That was the happy accident part of this. Yeah.

[00:21:26] Mike Kaput: Yeah, absolutely. And I will say that kind of one other area that we're starting, I think to explore more, it's not fully rolled out yet, but is really is in promotion.

[00:21:37] Mike Kaput: We can spin off social shares really, really fast, that are high quality from all this other content we have. I think in the future, we'll probably near future we'll be experimenting more. I would probably guess with, especially as we get into like YouTube shorts and things. There's a couple tools out there that I know we can explore to extract video clips really quickly from the podcast that are like su selected specifically by AI to be the more engaging quotes and moments in a particular episode.

[00:22:06] Mike Kaput: I think that would be a hugely valuable promotional use case and I'm sure there'll be some others, until we just get to fully synthetic voices and you and I can be, playing golf or something while our, our avatars are doing the podcast. But yeah, that's where we're at today, kind of with how we use it and, you know, Again, it, it's kind of hard to approximate, but I would, I would say this stuff, if we didn't have even these tools to do these and the process to do these with aIwould say it would take us three times as long Yeah.

[00:22:36] Mike Kaput: To go through this process.

[00:22:38] Paul Roetzer: And so just to kind of wrap this main topic, so the, a few lessons learned, I always like to kind of, you know, some takeaways from this. So the first thing for me is just be consistent. Keep showing up and creating value. Like, again, we didn't do this for the metrics, we didn't do this for the revenue, like, we did this because people are trying to figure this stuff out.

[00:22:56] Paul Roetzer: We're trying to figure it out, and let's just show up andcreate value. I often think of like the, the band analogy, like, just, just show up and play. The small venues like every night you just keep showing up and that's how they, they even eventually build an audience and a following, like a loyal fan base.

[00:23:10] Paul Roetzer: And for us, that wasn't, again, the goal. It was just like, let's just create the value and let's just keep showing up every week. And then the numbers started going. It's like, oh, okay, this is actually resonating with people. Let's keep doing it. Let's double down on this. So we've definitely seen that from a marketing perspective, be consistent and create as much value as possible at all times.

[00:23:27] Paul Roetzer: The second is, We just kind of covered drive efficiency, creativity, and performance with intelligent automation, what we always teach people as a use case model, look at something you already do and say, where can we use AI to make this more efficient, more creative, you know, higher probability of working.

[00:23:42] Paul Roetzer: And we just went through that. We have dozens of tasks involved in every podcast episode and we have AI infused in probably at least half of them. And as Mike said, it's driving massive efficiencies for us. And then the final thing, and this is something I just keep coming back to after 23 years, as a marketer and as an entrepreneur, you never know what's going to work.

[00:24:01] Paul Roetzer: So marketing is a very humbling profession. Sometimes the things that work are just, are by design. So we had a strategy here like the, the metrics aren't as a result of the strategy necessarily, but we at least had a strategy in like a why we were doing it. Sometimes the reason things work is just pure perseverance.

[00:24:19] Paul Roetzer: You just stay with the thing long enough and don't give up. And sometimes it's luck. But there was a great tweet, earlier this week from Paul Graham, kinda renowned in investor and entrepreneur, founder of Y Combinator. He tweeted, there's an interesting form of luck that affects startups. You start a successful startup.

[00:24:36] Paul Roetzer: Then a few years into it, some innovation appears and you're uniquely poised to benefit from it. Smartphone did this for Facebook. Like many kinds of luck, this kind only benefits people who are already out there doing stuff. And also like many kinds of luck, it's not completely random. If you work on an idea that's aligned with the future, it's more likely to happen to you.

[00:24:58] Paul Roetzer: So I saw that. I thought, oh my gosh, like, that's us. Like, andso I rep replied that the new innovation that changed everything for us was ChatGPT. It woke the world up to potential of ai. So it just so happens. That we changed this strategy in October of 2022. It was actually October 20th, 2022 was the first weekly format about a month before chat, c p t came out.

[00:25:19] Paul Roetzer: So if chat c p t hadn't happened, would we have hit a hundred thousand downloads this year? No way. Like, but because we were there, we were doing this, we were creating value and being consistent. We had like a vision for what could happen when that moment hit, our luck occurred. So it is this, it's this crazy combination in marketing and business and your careers and your companies where you just gotta be there and sometimes, you know, things happen andyou're in the right place to take advantage of it.

[00:25:47] Paul Roetzer: So, we'll kind of, you know, wrap this section. Just say thank you to all of you who listened regularly, who we've heard from, you know, about the impact this has on you. We're going to keep doing it obviously, and hopefully you know, this kind of brief about how we're doing it and why we're doing it. As helpful to you as you think about your own content strategies moving forward.

[00:26:07] So

[00:26:08] Mike Kaput: I referenced Claude a couple of times. This is a major topic for this week because AI company, an thropic, just announced the release of Claude two, which is its new foundational AI model. And so like chat CH pt, you can engage with Claude one and now Claude two through natural language prompts to perform a range of different functions like generate text, answer questions, produce code.

[00:26:34] Mike Kaput: Now, Claude two is just vastly more powerful than its predecessors, so it actually has improved its reasoning from previous, the previous model, and it now scores above the 90th percentile on things like the G R E Reading and writing exam. Now, a big, big thing with this model is that the number of tokens you can input into Claude two is massive.

[00:26:57] Mike Kaput: You can input up to a hundred thousand tokens in a single prompt, which according to anthro quote, means that CLA can work over hundreds of pages of technical documentation or even a book, and I looked this up for reference. The novel that I'm sure a lot of people read in like High School, the Great Gatsby, is about 72,000 tokens, which means you could upload the whole thing to Claude if you had it in the right file format and literally summarize or ask questions or write content off of that book.

[00:27:28] Mike Kaput: And you can actually also upload PDFs and C C S V files to Claude two and use it to query that information as well. An THROPIC also says the model is much better at coding tasks and math problems than the previous version. So I, I've played around with this a bit, but Paul, I'm going to definitely get your thoughts on how you've been experiencing it.

[00:27:49] Mike Kaput: But first off, how much of a leap forward is this model, in

[00:27:52] Paul Roetzer: your opinion? Well, having played with it, it's quite significant. I mean, I would say it's probably the closest thing to GPT four that I've personally experienced. They have a bunch of money, you know, we talked about before they raised four 50 million series C n A.

[00:28:07] Paul Roetzer: They're over 1.4 billion total raised, estimated to be worth over 4 billion. So, I mean, this is a, they're a major player, you know, it's them and inflection and OpenAI and Google and, you know, a few other key players. But they're, they're doing some really interesting things. And I know, you know, we wanted to talk about.

[00:28:25] Paul Roetzer: The example Ihad just this week where I was using a test case of, I'm interviewing Ethan Mollick. So at Maycom we're doing a fireside chat and anybody who doesn't follow his one US useful thing, newsletter, go subscribe. He has probably the most practical advice on the web right now of how these tools are playing out the large language models and comparing 'em and running tests on these things.

[00:28:46] Paul Roetzer: So we learn a ton from 'em. So in preparation for that interview, I thought, okay, well I've got about a dozen of his articles that I wanted to use as inspiration for the interview, and I thought this is the perfect use case for these tools. So my process was, as you and I were saying like, when I'm going to talk about something, I have to read it myself.

[00:29:04] Paul Roetzer: I cannot just give this yeah, to a, to one of these tools and have it summarized for me and process it. So what I did is I went through my own process for each of them. I had my notes, which often is copy, paste, bullet points of like key things that they say, and then I had my questions and I would draft my own questions based on the article.

[00:29:21] Paul Roetzer: Then I took the first article, which was, that one was about, let's see, which one was it? Oh, setting time on fire and the temptation of the button. So the whole article was about the, the Google help me Write button that's coming to Google Docs, which he already has access to and the impact that that's going to have on knowledge work and you know, the meaning of work and all of this stuff.

[00:29:43] Paul Roetzer: So I took that and started, I tested Google Bard inflection Pi Claude two ChatGPT 3.54, and code interpreter. So what I was doing with the first article was trying to figure out which of these models is actually the most usable for doing a summary. The prompt I gave each of them was, I need your help preparing for an interview with Ethan Malik, associate professor of the Wharton Business School.

[00:30:07] Paul Roetzer: I'm going to give you a series of Ethan Malik's articles one at a time, and I would like you to write a brief summary of each article and include five to seven potential interview questions for each article. Can you do that? And then I would get the response from each of them. So compared to the outputs of some prompting experimentation, I would look at kind of all, all that they gave me.

[00:30:29] Paul Roetzer: And so I ended up after the first one, centering on Claude two and Chet GPT four. andthe, the experiment was interesting in a number of ways, but Claude two absolutely performed on par with GT four for this specific use case. I, I'll give a little more context on the couple other ones. So the one Ithought was kind of shocking was Google Bard, which is I think the only one that's actually connected to the internet.

[00:30:58] Paul Roetzer: . So PLOT two is not connected to the internet. Chat g PT is not connected to the internet at the moment. I believe it's because, well, we're going to talk about this later, but because it was bypassing paywalls and extracted content from stuff it wasn't supposed to have access to. So when I gave this to Google Bard, which again connected to the internet, asked it that question, It said, I do not have enough information about that person to help with your request.

[00:31:22] Paul Roetzer: I'm a large language model, blah, blah, blah. So basically like I didn't understand what I was asking it, that I'm going to give you an article and then you're going to give it to me. It didn't realize that. So I said, are you able to read webpage and provide summaries? Yes, I can do that. Blah, blah, blah, blah, blah.

[00:31:34] Paul Roetzer: So I said, okay, great. Here's the first article, and I gave it the link. So now keep in mind, this is a blog post about the Google Docs. Help me write button, the summary from Google. Bard doesn't ever mention Google or Google Docs in the summary or the questions. So it was like completely missing the point of like what this thing was, and it was basically unusable.

[00:31:57] Paul Roetzer: So what I realized about Bard at that moment, now again, we'll talk about the updates they just made, was it had some decent insights. But it basically just used a bunch of filler words to sound smart. Like that was my interpretation of it. If I asked someone, like an intern or an associate to do this project and they came back, I was like, I this, I would set it aside and start asking them questions about what they actually read to see if they really understood it.

[00:32:19] Paul Roetzer: because it was pretty apparent to me that it didn't really understand the context of this article at all. So Bard was like immediately out inflection pie. Oh my God. So I have, I have, I like Inflection Pie. It's a really interesting, very conversational agent. It is free to go use, so test it for yourself. It's supposed to be more personal, more empathetic, more conversational, but it just asks you questions over and over and over again.

[00:32:46] Paul Roetzer: So for this use case, it was absolutely maddening to use. So I, it didn't, you know, it wouldn't, it couldn't do any of the url. So I pasted the whole article in, like, take, copy, paste, put it in there, and then it would just ask me questions about it. And I would say, can you please just give me the summary that I asked for?

[00:33:05] Paul Roetzer: And it would ask me another question and give me emojis, and I would say, okay, Ineed the five to seven questions. Can you please give that to me? It would give me emojis. Again, it, it would never do the thing you asked it to do. So inflection was out 3.5, chat J p D 3.5 did a good job, code interpreter was amazing.

[00:33:23] Paul Roetzer: So that was the first time I actually applied this to code interpreter. But again, keep in mind it's not connected to the internet, so you have to give it the P d F. So in the case of code interpreter, you can actually upload the pdf, d f, or give it the full text. And then the same with ChatGPT four, and Claude.

[00:33:38] Paul Roetzer: So I would give it the full text with Claude. You can upload the PDF f as well. And it actually did a little better with the full PDF d upload than it did with just the text. I tested both. So overall what I would say is for this use case for some summarization of articles, and this is the beautiful thing for me, is I've been looking for this, this exact use case for years.

[00:33:57] Paul Roetzer: You and I read a lot of AI research papers. They're dense, they're scientific, they're long. I haven't had anything to help me summarize that. I can now take a pdf d f, upload it to Claude two and say, summarize this and provide, you know, maybe the 10 bullet points the key to this thing and it'll do an incredible job.

[00:34:14] Paul Roetzer: So if you're looking for this use case, again, I haven't expanded my uses of CLA two yet beyond that, but it is g pt, g PT four level. So I would say code interpreter g gpt four and Claude two are roughly on par for the use case that I did. And I'm, I'm saying like these summaries were incredible and the questions were deep.

[00:34:34] Paul Roetzer: And so what I did is I then took the 10 articles, put 'em all into Claude two and Chad GPT four, had it do its own summary and export of questions. And then I did my own for each. And then I merged all of those. And I would say out of the, you know, 20 or 30 questions I ended up with as possible questions for the interview.

[00:34:51] Paul Roetzer: About half of them were from GT four or Claude two. Sometimes it was just, they asked it better than I asked it. Other times they brought in context of things that weren't in the article at all. Deeper meaning behind what he wrote. And I was like, damn, that's really good. So yeah, I mean, my first take on Claude two is extremely impressive.

[00:35:11] Paul Roetzer: I, but I, again, it's early in terms of how I'm using it. So it's also of

[00:35:17] Mike Kaput: note that anthro partners with companies like Jasper to bring models like Claude two to their customers. So you may end up be being using some of these models, through other tools that you use now. It sounds like you would recommend marketers take time to familiarize themselves with Claude Two.

[00:35:34] Mike Kaput: Kind of given the strength of the model. In my view, it's like, I mean, I know there's a lot of competing products out there, but so far my tests with it, Ithink it's worth paying

[00:35:43] Paul Roetzer: attention to. Yeah. And as we've said many times, you have to keep coming back to these tools. . Like we're going to talk about Google Bar next, like, it failed this time, but then they updated it yesterday morning.

[00:35:54] Paul Roetzer: I don't know if they made an update to the core algorithm too, like the, the core model that like, Makes the outputs better, better understanding. But you know, Claude too is brand new. It's way better than Claude. Just as GT four was a leap forward over GPT three in my opinion. So these things keep evolving and you, you really have to stay up on 'em.

[00:36:13] Paul Roetzer: What I would say is have your like three to five standard use cases that you want to use these tools for. Transcription, summarization, drafting articles, and every month or two go in andtry those across multiple models because they're all going to keep moving ahead of each other at different times for different capabilities and you really need to stay testing them.

[00:36:36] Mike Kaput: So speaking of Google Bard, we just got a announcement for a ton of new features that Google has announced for Bard, which is, of course it's chat j pt, like conversational agent,

[00:36:48] Paul Roetzer: and also what they're building into Google Workspace.

[00:36:51] Mike Kaput: Yes. So Bard is actually now available in 40 languages, and it's available now in a bunch of other countries that it wasn't before, including significantly the European Union where Google had not previously released it due to concerns around some of the data privacy rules in the eu.

[00:37:09] Mike Kaput: Other major updates include things like you can now listen to Bard's responses. Instead of simply reading them, you can actually change the tone and style of Bard's responses. There's five different options for tone and style, simple, long, short, professional, or casual. You can also name or rename and pin conversations so you can reference those really quickly later.

[00:37:32] Mike Kaput: A new feature allows you to export Python code to repli, which is a really popular browser based integrated development environment, or i d e share shareable links make it easier to share bard's responses with other people. And Bard now supports images, at least for English users right now. You can drop images into it and have Bard perform a range of tasks related to the image.

[00:37:54] Mike Kaput: It can tell you what's in it, write a caption for you, things like that. Now, Paul, as you're kind of hearing these announcements, what did you make of them?

[00:38:03] Paul Roetzer: Well, one Sundar Phai tweeted it. So, you know, it's, they think it's a big deal. So usually if I see Sundar tweeting, it's like, okay, this is a major launch for them.

[00:38:12] Paul Roetzer: They're trying to make a big deal out of it. I think it's interesting from a perspective of what they're choosing to launch . And how it differs from some of the other models, because I do think that there's this race to differentiate from each other. And I think they even called it augmented imagination was what they're saying, like a new take on this stuff.

[00:38:35] Paul Roetzer: andso I think like the multimodal, like being able to, you know, give it an image and ask what's in this image, being able to prompt it based on images. It, it's certainly interesting. I, Ihope that they're not doing all these other things in lieu of solving for the fact that the outputs are subpar usually.

[00:38:55] Paul Roetzer: So my experience with the Bard has been largely disappointing in terms of what it creates. Just as I explained with the, like the summarization thing, it's nowhere close, not even GPT three level close. Like it's just not on the same level. I know they have better capabilities that they're holding back for whatever reasons.

[00:39:12] Paul Roetzer: So I keep waiting for Bard to actually be on par with GT four or Beyond. It is not, but it has a bunch of features now that those ones don't. So in, Yeah, I mentioned in my testing. So it's got solid insights make, but otherwise, like it's very literal. It's not very advanced in terms of what it outputs.

[00:39:33] Paul Roetzer: So, yeah, I mean, I think obviously it's interesting features, until they solve the quality of the outputs and get them on par with some of the other models. Ijust don't see it, you know, being a true competitor yet. But as we've always said, the big play here is who has the data and the distribution.

[00:39:53] Paul Roetzer: Mm. And while Claude two may be better right now, and even GPD four is better, at the end of the day. Those aren't going to be the things that are built into Google Workspace. So you're going to have the distribution of Google Workspace to build Bard into. And Google obviously has more data than anyone, especially once they start training or infusing video into these as well.

[00:40:14] Paul Roetzer: They have YouTube to, to integrate. They have Google images, they have, they just have more multimodal data than anyone. And if, if they can truly succeed at infusing that into this model andmaking Bard what it's capable of being, you just, you continue, cannot bet against what they end up doing. I just hope we see a leap forward in its quality of output in the near future.

[00:40:39] Paul Roetzer: Gotcha.

[00:40:41] Mike Kaput: That is a really good take, and I think that's helpful to help kind of our audience parse through all these different options because like you mentioned, you know, you have to keep revisiting these tools. Things could be night and day from the last time you've used one of them for a specific use case.

[00:40:56] Mike Kaput: Let's dive into our rapid fire topics. So first up, you. Alluded to this in the beginning of the podcast, but we are thrilled to announce that Cassie Kozyrkov Chief Decision Scientist at Google is going to be on the MAICON stage in just a week or two here. Koff founded the field of decision intelligence at Google and advised leadership on decision process AI strategy, and how to build data-driven organizations.

[00:41:28] Mike Kaput: So Cassie Kozyrkov is kicking off day two of MAICON with the keynote, which you mentioned. Whose job does AI automate? Paul, can you tell us a little bit more about Cassie's appearance at MAICON? Kind of the keynote, just kind of what we might want to, might be expecting from her.

[00:41:45] Paul Roetzer: Yeah, I'm just, I'm super excited about this one.

[00:41:47] Paul Roetzer: She's been on my radar for three years, like she's been in my kind of sandbox wishlist of speakers and, just sort of serendipitously this one came together kind of last minute and. We were extremely excited to be able to add her as kind of the final keynote and announce that. So it, it's going to be incredible.

[00:42:06] Paul Roetzer: I mean, just to hear her perspective, she has amazing online classes, YouTube videos. You can go check her out and see some of her training. It, it's just a remarkable addition to the lineup, so I can't wait for that, that talk. That's awesome.

[00:42:20] Mike Kaput: So, another notable person, in ai, Elon Musk has not speaking at MAICON, not not speaking at MAICON.

[00:42:27] Mike Kaput: To be clear, Elon Musk has launched a new AI company called X a ai, and it includes a team of top researchers from companies like Google, Microsoft, DeepMind, and others. And the company's stated goal in their own words is to quote, understand the true nature of the universe. Now, if you're wondering what that means and wondering how the company defines that, Let me know if you find the definition cause they do not elaborate on that mission.

[00:43:01] Mike Kaput: So not a lot of details about this, but this is Musk's kind of latest AI venture after previously co-founding and then leaving OpenAI. And obviously we have talked about in the past, he has been a bit critical of OpenAI and certain AI efforts going so far as to be the lead signer of the Pause AI research for six months letter that we talked about a few weeks or a few months ago, rather.

[00:43:25] Mike Kaput: Now Paul, there's no doubt Elon Musk is a publicity machine, makes some big promises that don't always come to fruition. Is this announcement

[00:43:33] Paul Roetzer: legit? Yeah, I think it is. I mean, he's always talked about x.com. X ai is like, you know, the grand play and I think the, the key is, When you look at the totality of the companies he owns or runs, we talk about the importance of data.

[00:43:51] Paul Roetzer: They have a really interesting potential play here. So one, he has Twitter now and he shut off access to their data to everybody else, basically. So he can hoover up all of the data within Twitter. He obviously has Tesla, which gives the real view. We've talked about Jan Lacoon and the idea that true intelligence will come from understanding the world around you.

[00:44:10] Paul Roetzer: Teslas have eight outward facing cameras and tens, 10 billion plus miles, where it's learning from its outward environment. He has optimists, which they're trying to build within Tesla, which is a humanoid robot that takes that out world, out world learning, gives it language and all these capabilities. He has Neuralink, which is embedding, fibers into human brains to try and, you know, affect neurological processes.

[00:44:33] Paul Roetzer: So he's absolutely playing in this, in a lot of different areas and can bring together, and SpaceX obviously, so they can bring together a lot of learning. And so my guess is this is the grand play outside of getting to Mars and being interplanetary species. All of these pieces that may seem like totally unrelated companies probably all feed up to his grand vision of solving, you know, what this is all about.

[00:45:01] Paul Roetzer: . And so he's almost taking it like beyond AGI and super intelligence and saying like, we need to develop this understanding of the universe. I don't know, I I didn't get a chance to listen to his interview. I think he did a, a Twitter sphere. Yeah, that's what they're called Thursday night. I didn't, I didn't get a chance to listen to that, so he probably expanded on it a bit.

[00:45:19] Paul Roetzer: So I don't know. I mean, it's, it might be one of those things like you launch and you don't hear, like Neuralink, it's been, you know, supposed to be a big deal for a really long time and yeah. Shows up every, you know, 10 months and does a little press conference. So this might be like that. We don't hear anything about it now for six months.

[00:45:33] Paul Roetzer: Or it might be the big thing you just never know with him. So another

[00:45:38] Mike Kaput: news comedian Sarah Silverman, is actually suing OpenAI and Meta claiming that they used a book she wrote in 2010 in an unauthorized way to train their models. Now she's actually joining several other authors in class action lawsuits against OpenAI and meta.

[00:45:56] Mike Kaput: So these lawsuits essentially claim that Silverman and the other authors involved did not consent to the use of their copyrighted books to train OpenAIs ChatGPT and Meta's Lama model Now. Paul, it seems like OpenAI, and maybe to a lesser extent meta, but mostly OpenAI is getting hit on all sides with lawsuits.

[00:46:15] Mike Kaput: Now, is there a chance that one or more of these legal actions forces OpenAI or another major AI company to alter their models or take one or more of their models off the market? As I believe researcher Margaret Mitchell has speculated could happen.

[00:46:31] Paul Roetzer: I have no idea. All I know is the, the lawyers for OpenAI are very, very busy people right now and it shows no signs of slowing down.

[00:46:42] Paul Roetzer: So I, you know, I don't know how many lawsuits that are active that we know about. Yeah. But the, the other one, you know, that we were going to touch on too is the FTC open investigation against OpenAI as well. So it's like been a very busy legal week for OpenAI and I'll say Sam Altman did tweet about this one, which Iwas, kind of caught me off guard that he commented on the F TC one.

[00:47:02] Paul Roetzer: Yeah. Tweeted on July 13th. It's very disappointing to see the FTCs re request start with a leak and does not help build trust. That said, it's super important to us that our technology is safe and pro-consumer and we are confident we follow the law of, of course, we'll work with the ftc. He then went on to say we built GPT four on top of years of safety research and spent six plus months after we finished initial training, making it safer and more aligned before releasing it.

[00:47:30] Paul Roetzer: We protect user privacy and design our systems to learn about the world. Not private individuals, which I imagine carries over to people's books and other things like Sarah Silverman's complaining about. We're transparent about the limitations of our technology, especially when we fall short. And our cap profit structure means we aren't in incentivized to make unlimited returns.

[00:47:49] Paul Roetzer: A pretty hollow argument, honestly. Like, it's basically like, Hey, but we're doing all this stuff, right? Like, oh, you're causing all this harm and doing all these things wrong, but okay, you're doing a few things right? Or like seemingly right. Not going to help them with the ftc. I mean, the FTCs not going to care about stuff like this.

[00:48:06] Paul Roetzer: They're going to get into like, okay, yeah, great, but here's the 10 things you're doing that are causing harm. Like, let's talk about those not the few things you're doing that you think are correct. So that's why I have no idea how this plays out. I think it's going to be a story we will be following for years to come.

[00:48:20] Paul Roetzer: And I, like I said, I think the lawyers for OpenAI and stability, AI and all these other players, Google I'm sure is going to get yanked in and e everybody. It's just going to be an ongoing issue in legal battles. So who knows?

[00:48:35] Mike Kaput: So I don't know if it's just reading all these stories at once or if this is a real thing, but my sense is like pressure is ratcheting up with OpenAI from some legal sources.

[00:48:47] Mike Kaput: I mean, obviously it's not new that they're being sued, right? But between F T C in investigation being open and these new lawsuits coming to light, I mean, it does seem like the pressure could be on.

[00:48:59] Paul Roetzer: Yeah, I mean, again, we have to keep in context. This tech is to most of the world new as of November 30th of last year.

[00:49:06] Paul Roetzer: And so everybody tries to figure out what it means and the play is, and by spring of this year, these lawsuits were probably getting rolling. There was probably many people complaining. Yeah. And open eyes probably known about the FTC investigation for a while, because Sam mentioned it was a leak. So, yeah, I mean, I, I'm sure there's way more going on behind the scenes and probably a lot more lawsuits coming that we're not hearing about yet.

[00:49:29] Paul Roetzer: But it, yeah, I mean certainly if Sam's commenting on it, It's at least top of mind at the moment.

[00:49:35] Mike Kaput: Yeah, so actually kind of interestingly a little bit related to this, Google announced that it's time to explore alternatives to its robot txt web standard for controlling how search engines crawl web content.

[00:49:52] Mike Kaput: And they actually said, quote, we believe it's time for the web and AI communities to explore additional machine readable means for web publisher choice and control for emerging AI and research use cases. Now, there wasn't a ton more commentary aside from that, but basically the company is kicking off a public discussion among web and AI communities to weigh in on possible alternative or complimentary protocols to the standard web crawling that it's been doing for the last couple decades Now.

[00:50:25] Mike Kaput: Paul, I kind of read that. Is this a response to some websites kind of crying foul over how companies are scraping their content for training or AI tools being able to get behind paywalls, like we're going to discuss, what was your kind of read on what's going on here?

[00:50:40] Paul Roetzer: That was my assumption, is this, it's, I mean, part it, it probably deals with just overall search and online content and publishers wanting to protect their assets.

[00:50:49] Paul Roetzer: Part of it might be trying to be proactive about regulations. . You know, by being able to say, here's the steps we're taking. So definitely a development worth continuing to watch. I think this idea has been floated for a while, but there was going to need to be some other protocol. . As AI starts, you know, these things are, you know, getting, consuming so much content on the web and learning from it that we're going to need other ways to protect, that content from being ingested into foundation language models.

[00:51:15] Paul Roetzer: So, yeah, we'll see, see where it goes. And to kind of

[00:51:19] Mike Kaput: show a practical example of what's at stake here. OpenAI has temporarily turned off ChatGPTs browse with Bing feature, which is available to ChatGPT plus users or was, they actually express concerns that the feature which basically just browse as the web to respond to user prompts could sometimes show content that is behind paywalls or content that's otherwise not fully accessible by the public.

[00:51:46] Mike Kaput: So, you know, this obviously comes, as they're facing all these other legal challenges over data privacy and copyright. So OpenAI says they're doing this out of quote, an abundance of caution. Is that your read on this as well? Basically they're like, oh shoot, this tool is actually violating people's protected content rights.

[00:52:05] Paul Roetzer: Yeah, an abundance of caution sounds like legal terminology for we're not admitting fault, but Yes, it, it does what you said it does. Yeah, Imean, it, it, it's, Idon't know that it's truly caution. I think it's maybe we got busted with this thing doing something we weren't aware it was doing, and now we gotta fix it.

[00:52:25] Paul Roetzer: So I'm not, I'm honestly not sure when the browse with B thing went away. Like I hadn't tried it until this week, and that's when I noticed it. Yeah. And I did a search and I found, you know, the article about it, and that's when I realized like some of these, like clawed too and, some of these other weren't connected to the internet either and Right.

[00:52:42] Paul Roetzer: Yeah. It's, I'm not, I don't, I haven't heard any official thing from Open. I was scouring their, like Twitter feed and post, like, they haven't said anything about it that I could see. So I don't know when it actually happened. Yeah. I think

[00:52:52] Mike Kaput: their announcement about wasn't even really an announcement. It was kind of buried in something else.

[00:52:56] Mike Kaput: Like, by the way, we're,

[00:52:58] Paul Roetzer: by the way, the main valuable plugin that we provided before, code interpreter is no longer functional.

[00:53:06] Mike Kaput: So another interesting development, this one's I guess, a little more positive for OpenAI, is that Shutterstock is extending its existing partnership with OpenAI for six more years, which is allowing OpenAI to continue training its AI models on Shutterstock's image video and metadata library.

[00:53:24] Mike Kaput: Now, the two companies actually started partnering in 2021 to train the Dali image generation model. At the same time, Shutterstock created a fund to compensate artists whose work was being used ar that they had put on their platform. Now basically Shutterstock has as of today, integrated Dolly into its site.

[00:53:47] Mike Kaput: And strange, interestingly, not strangely banned non dolly generated AI images. It now plans to let users edit any Shutterstock image with ai. So basically, they are seemingly embracing partnership with OpenAI, as a kind of mutually beneficial partnership. That's a bit of a different approach from one of their competitors.

[00:54:09] Mike Kaput: Getty Images, which has banned AI generated content entirely and is also suing, I believe, stability AI for copyright issues. We're seeing two very different experiments with AI play out in real time here. It sounds like, you know, on one hand, shutter stock is formally partnering with OpenAI committed for the next six years to be sharing the.

[00:54:33] Mike Kaput: Their images, their data. Whereas Getty has just gone, you know, guns blazing against AI companies. Is there a right approach here? What are some of the kind of pitfalls, nuances

[00:54:45] Paul Roetzer: to this issue? I guess time and consumer usage will tell. You know, I think the Shutterstock approach, I mean, they did this last year.

[00:54:54] Paul Roetzer: I think they first announced this, cause I believe Stephan from Shutterstock was at our, our conference last year. Yeah. And you talked a little bit about this and, so yeah, I mean, it's divergent approaches. It'll be interesting to see, where it goes. But I think we're going to see a lot more stuff like this where there's licensing deals.

[00:55:13] Paul Roetzer: People are going to try and get out ahead of the laws and regulations and try and do it as, you know, properly as possible. So they have more cover when the time comes and they have to explain how it all works. So, yeah, we'll see. It's, Imean, Ithink the Shutterstock approach is smart. . So far.

[00:55:30] Paul Roetzer: Yeah,

[00:55:31] Mike Kaput: so at this year's IO conference, Google announced something called Tailwind, which is an AI powered note taking tool. The news here is that that tool is now being rebranded to be called Google Notebook, lm, and it's launching to a small group of US users. And right now it seems like you'll be able to use Notebook, LM to essentially select one or more Google Docs and ask questions about them, create new content spun off of them, do all the things you might be able to do by putting them into a language model.

[00:56:03] Mike Kaput: For instance, you could summarize documents or spin off a script from, say, a video outline you have in a Google Doc. Now, Paul, we see features like this in a lot of third party AI tools and startups. I mean, is the trend here that we're just seeing them baked more and more into the actual applications than themselves that we use to create documents?

[00:56:25] Mike Kaput: You know, Microsoft 365, Google Workspace, et cetera.

[00:56:29] Paul Roetzer: That's going to be the story of the second half of 2023, and probably in the early 2024 in my opinion, is as Microsoft and Google come online with the gener AI features that are just baked into the platforms, what does that do to the app ecosystem? And nobody knows, but I mean, certainly they could build any feature like this they want, and again, they have the distribution of Google Workspace and anyone using Google Docs.

[00:56:53] Paul Roetzer: So, yeah, I Ithink it's going to be interesting to see how this all plays out with the, you know, Microsoft and Google launching these capabilities wide scale and what it means to us as users and what it means to the kind of the startup ecosystem.

[00:57:08] Mike Kaput: So in another interesting AI product announcement, Shopify, the e-commerce business platform just announced something called Sidekick.

[00:57:16] Mike Kaput: This is an AI assistant embedded right into its platform. So if you are a Shopify user, sidekick can do everything from answer questions about your business to actually perform tasks for merchants within their Shopify instance. It can also do things like analyze sales trends and look at your website data and tell you interesting things about it.

[00:57:38] Mike Kaput: Now Paul, this seems like a no-brainer for a platform like Shopify. Do you think every platform or application company should be building something like this into

[00:57:48] Paul Roetzer: their product? Yeah, it's smart. And you know, I think what caught my eye about this is I saw, because Toby lucky, the c e O of Shopify had tweeted out a video like showing this and the way they positioned it was really intelligent, which is.

[00:58:01] Paul Roetzer: You know, these small business owners, these e-commerce shop owners, they need help. Like they need strategy and they need guidance. They can't afford marketing agencies, they can't afford consultants all the time. And so what they're doing is saying like, we're just going to build it right into you, and 24 7, whenever you're working on your business, it'll be there for you as a sidekick.

[00:58:18] Paul Roetzer: So I think. The positioning of it is great. The value proposition of it is great. I think every SaaS company is going to have to have something like this, and I haven't seen many do it well yet. . It's kind of what I want chat spott to be. You know, we've talked about HubSpot and chat spott before.

[00:58:37] Paul Roetzer: You just want that assistant there at all times to help you run your platform, answer questions about reporting data, answer, you know, give recommendations of what to do. That's what I think we're looking at as a future where all of these software systems have this built in intelligent assistant that is there for you 24 7, like a strategist or consultant would be.

[00:59:01] Paul Roetzer: So I haven't tested it, but it looks really promising. So the

[00:59:07] Mike Kaput: Associated Press and OpenAI have also announced a partnership, and this is a new partnership where they say they'll share news, content and technology and explore generative AI use cases in news. Now this is nothing new for the ap. Way back in 2014, they actually began automating corporate earnings reports with, intelligent automation and they've been experimenting with AI technology ever since then.

[00:59:33] Mike Kaput: Now, as of right now, basically rating the announcement, it's not super clear exactly how AI or how AP will use OpenAI technology or how OpenAI will use an access the APS content, but I would say this relationship is pretty significant given the size and clout of the players involved. So I know details are pretty thin right now, but what do you think is going on with this partnership?

[00:59:58] Paul Roetzer: Again, it seems like with Shutterstock they're making a very concentrated effort to go to the data sources and build licensing deals and partnerships and so that their future foundational models. So GT five for example, I would imagine there's a race to probably do all kinds of licensing deals with content providers.

[01:00:17] Paul Roetzer: I would think other major media companies would be, you know, in line for this, where they're going to license the access to the data. . So it's going to cost 'em a bunch of money. They may have to raise more money. But it's the right way to do this. it's how they'll get around the, the eventual if issue of what this data is trained on.

[01:00:34] Paul Roetzer: If they can train models in the future based on stuff that they're more confident they're allowed to have access to. So Iwould assume we're going to see more deals like this coming. So right

[01:00:45] Mike Kaput: now in Hollywood, writers are on strike and have been for some time, and we just got confirmation that Hollywood actors are also going to go on strike.

[01:00:55] Mike Kaput: Now, that's significant in and of itself, but what's really interesting is as part of this news, the chief negotiator for the Screen Actors Guild sag. The Union essentially revealed that Major Hollywood studios had offered. A proposal, an AI related one to the guild. And according to the chief negotiator, the proposal suggested that studios be allowed to create digital scans of background performers in exchange, they would pay them a day's pay to digitally scan them and then create and retain rights to their digital likeness forever without further compensation.

[01:01:36] Mike Kaput: Now, use of AI to replace human professionals is also a core concern of the writer strike. So Paul, this sounds a bit honestly like science fiction, but it's actually a huge and very real issue that many performers and creators are navigating right now. Essentially, how can you own and protect your likeness, your voice, your writing style, et cetera, when we can realistically recreate it them with AI tools?

[01:02:02] Mike Kaput: What are your thoughts on this?

[01:02:04] Paul Roetzer: This is a space I haven't spent a ton of time thinking about, but just as you're talking, I'm thinking about all the extras that go into these, yeah. You know, making movies. So I don't think, I mean, you're not replacing like the lead actor actresses right now, but if you think about all the extras behind and how easy it would be most likely to develop digital versions of those people, and not have to have all these extras.

[01:02:26] Paul Roetzer: So with the way these guilds or the unions work is you gotta protect, you know, the, the lowest people on the poll. So these lead actors, you know, are going to come out and say, no, no, no. Like we're all part of the same group. . You're, you can't replace them with digital beings. Like, I, it'll be fascinating to see how this one plays out.

[01:02:43] Paul Roetzer: I, it does sound sci-fi for sure, but we know the tech is there to do exactly what they're worried about happening. So I don't know how you negotiate that without a very prolonged thing, because we're talking about entirely new technology and precedence here. That many of these people probably don't even comprehend and now they have to negotiate these key things and this is, it could be a very important part of the future.

[01:03:05] Paul Roetzer: Like they, they can't mess up this negotiation. So I have no idea how this one plays out. It'd be fascinating to watch. So,

[01:03:11] Mike Kaput: Yeah, no kidding. And as someone, this doesn't spoil anything but the first, 20 or so minutes of the new Indiana Jones movie take place. In the past when Harrison Ford's character was the age of the original movies, and I have to say it is breathtaking, deep fakes of him, just you honestly could have made the whole movie with that, what they did there.

[01:03:34] Mike Kaput: It's truly incredible. It's wild. It's come so far and I was like blown away. Wow. All right. Last but not least, Forbes is reporting that hugging face, a major open source AI library of machine learning models, which is used by a bunch of major players to build new models, is actually raising a new funding round that values the company at 4 billion.

[01:03:59] Mike Kaput: Now reportedly this Series D Round will raise at least 200 million, though none of this is, confirmed yet as they're still kind of shopping around for funding. Hugging face might be something, it's very well known in more technical circles in the AI world, it might be something a little less well known to some of our audience members.

[01:04:19] Mike Kaput: Can you maybe just walk us through kind of what is the importance of hugging face in the larger AI ecosystem? I

[01:04:25] Paul Roetzer: think a simple way to think about it is as we move forward, everyone is going to be building AI tools and capabilities. Every brand, every web shop, every developer, and they accelerate the capability of that.

[01:04:37] Paul Roetzer: They allow people who can build things to go in and get code and get, you know, get systems that they can use to build tools, build capabilities. And so as that spreads, I mean, they're going to just keep growing in popularity. So again, one of those, like, go back to the Paul Graham quote, you know, about luck, sort of, being in favor of the people who were already in the arena.

[01:04:57] Paul Roetzer: They, they would fit that. Description. You know, they've been here doing this for a while and now AI's having its moment and that moment appears like it's going to go for an extended period of time, and they stand to benefit greatly from that and accelerate the adoption and creation of AI technology.

[01:05:14] Paul Roetzer: Awesome.

[01:05:14] Mike Kaput: Well, Paul, as always, thank you for the time and insights breaking down the latest in ai. I'm pleasantly stunned by how many topics happened in the last four days to cover. I believe this is four days worth of news, so we appreciate you as always helping us demystify

[01:05:32] Paul Roetzer: some of this. All right, well everyone have a great week.

[01:05:35] Paul Roetzer: We will be back, I don't know what day that's going to be, the 27th, 24th, something like that. 25th with the next fifth episode. Yeah. Yeah, so have a great week and we'll be back with a lot more next time. I'm sure. Thanks for listening as always.

[01:05:51] Paul Roetzer: Thanks for listening to the Marketing AI Show. If you like what you heard, you can subscribe on your favorite podcast app, and if you're ready to continue your learning, head over to www.marketingaiinstitute.com. Be sure to subscribe to our weekly newsletter, check out our free monthly webinars, and explore dozens of online courses and professional certifications.

[01:06:13] Paul Roetzer: Until next time, stay curious and explore AI.

Related Posts

[The Marketing AI Show Episode 44]: Inside ChatGPT’s Revolutionary Potential, Major Google AI Announcements, and Big Problems with AI Training Are Discovered

Cathy McPhillips | April 25, 2023

This week's Marketing AI Show covers ChatGPT’s potential, more announcements from Google, and problems with AI training.

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

Cathy McPhillips | August 1, 2023

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

[The Marketing AI Show Episode 43]: AWS Gets Into the Generative AI Game, AutoGPT and Autonomous AI Agents, and How AI Could Impact Millions of Knowledge Workers Sooner Than You Think

Cathy McPhillips | April 18, 2023

This week's episode of The Marketing AI Show talks about generative AI announcements from AWS, AutoGPT, and how AI could impact millions of workers.