GPT-4 is changing the game. Access is easier, outputs are better, and technologies connecting to it are increasing exponentially with the help of a new plugin system. What will the rest of this week bring us?
OpenAI launches a plugin system for ChatGPT
OpenAI just announced a plugin system for ChatGPT, enabling it to interact with the wider world through the internet. The plugins, developed by companies like Expedia, Instacart, and Slack, will allow users to perform a variety of tasks using these sites from right within ChatGPT.
It’s not just companies wanting to embed AI into their sites. OpenAI itself is hosting three of the plugins: one that gives ChatGPT access to up-to-date information on the internet, a Python code interpreter, and a retrieval plugin that allows users to ask questions of documents, files, notes, emails, and public documentation.
Of particular note, one of the plugins available integrates with Zapier, which itself integrates with thousands of other tools. Right now, there’s a waitlist to access the plugins for developers and ChatGPT Plus users.
Did we just open a whole new world of AI use cases?
Artificial General Intelligence…one step closer
“OpenAI’s mission is to ensure that artificial general intelligence (AGI)—by which we mean highly autonomous systems that outperform humans at most economically valuable work—benefits all of humanity.” We read this in episode 36 of The Marketing AI Show, just over a month ago.
Now, OpenAI is saying, “Our mission is to ensure that artificial general intelligence—AI systems that are generally smarter than humans—benefits all of humanity.” A team of Microsoft AI scientists claims that GPT-4, the latest iteration of OpenAI’s Large Language Model, exhibits “sparks” of human-level intelligence, or artificial general intelligence (AGI).
The researchers argue that GPT-4’s impressive performance in a wide range of tasks, such as mathematics, coding, and even legal exams, indicates its potential as an early version of an AGI system.
While some argue that AGI is a pipe dream, others believe that it could usher in a new era for humanity, and this research indicates GPT-4 might just be leading the way.
Are these thoughts and findings legit? How seriously should we take it?
It only took 30 minutes to market a product launch
Imagine leveraging the power of AI to complete a massive business project in just 30 minutes, accomplishing tasks that would take humans hours or even days.
In a remarkable experiment from Wharton professor Ethan Mollick, a combination of AI tools was used to market the launch of an educational game, conduct market research, create an email campaign, design a website, and craft a social media campaign, among other tasks—in just 30 minutes.
The results demonstrated the unprecedented potential of AI as a multiplier of human effort, with vast implications for the future of work, productivity, and creativity. Over the course of half an hour, Mollick used no more than 20 inputs, actions, or prompts to generate 9,200 words of content, a working HTML/CSS file, 12 images, a voice file, and a movie file across a marketing positioning document, email campaign, website, logo, script and video, and social campaigns.
As he put it “AI would do all the work, I would just offer directions.”
Is this the new normal for marketers?
Listen to this week’s episode on your favorite podcast player, and be sure to explore the links below for more thoughts and perspectives on these important topics.
00:05:40 — OpenAI announces ChatGPT plug-in system
00:21:39 — Microsoft scientists claim GPT-4 exhibits “sparks of human-level intelligence”
00:31:20 — 30 minutes to launch an educational game using AI
00:45:16 — character.ai funding led by Andreessen Horowitz
00:49:01 — University of Maryland computer scientists reveal methods for detecting AI-generated text are unreliable
00:50:47 — Adobe announces creative generative AI models, Adobe Firefly
00:54:28 — Nvidia announces a suite of cloud services to accelerate enterprise adoption of generative AI
00:57:48 — The secret history of Elon Musk and Sam Altman
Links referenced in the show
- ChatGPT plugins
- Is GPT-4 early AGI?
- Fully marketing a new product launch in 30 minutes using AI
- Character AI Series A
- Can AI text be reliably detected?
- Adobe Firefly
- NVIDIA Foundations
- The Secret History of Elon Musk and Sam Altman
Watch the Video
Read the Interview Transcription
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
[00:00:00] Paul Roetzer: My biggest fear at the moment is a lack of understanding of this technology, what it’s capable of, what its limitations are, and that the people at the top of companies and agencies and media, you know, outlets and publishers, whatever, the ones that are going to make the decisions about the impact on staffing probably don’t understand the technology driving those decisions.
[00:00:18] 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.
[00:00:39] Paul Roetzer: My name is Paul Roetzer. I’m the founder of Marketing AI Institute, and I’m your host.
[00:00:47] Paul Roetzer: Welcome to episode 40 of the Marketing AI Show. I’m your host, Paul Roetzer, along with my co-host Mike Kaput. Who, where were you at last week? I was in San Francisco. Where were you? I was in Florida, . Okay. You were talking AI with agencies, right? Like a agency event? I was, yeah.
[00:01:03] Mike Kaput: At the, agency Builders Retreat, which was a small kind of exclusive event for about a hundred, 150 agency owners and executives.
[00:01:13] Mike Kaput: Okay. And it was, an eye-opening session for both myself and the audience about what was possible with AI and marketing agencies. Definitely. Got tons of questions after I, think I scared a few people, but they were scared in a good way. , like they, someone told me I’m scared, but in a good way because I just need to move with some
[00:01:33] Paul Roetzer: emergency
[00:01:34] Paul Roetzer: Well, that’s good. I was out at, the Pavilion CMO summit, so it was really cool. It was a group of, I don’t know, it was like a hundred, 150 people maybe, I think at the event. But it was great. I mean, like lots of questions at the CMO level. Lots of people, you know, really trying to figure this out.
[00:01:51] Paul Roetzer: But yeah, I will say for our listeners, like if you’re early in this and feeling overwhelmed, you are not alone. Like the , the more Mike and I are kind of out doing talks and meeting with people and engaging with the community, the more you realize like, people are, are really struggling to move on this and do more than like a use case.
[00:02:12] Paul Roetzer: So, yeah, I don’t know. I, and you know, big topic, obviously this week we’ve got our AI for ri writers summit. It’s coming up on, is it Thursday? Yep. It’s like, okay, really gotta build that deck. , anybody who’s a regular listener knows that, I have not built my deck yet. And, that that has continued.
[00:02:32] Paul Roetzer: So it’s partially really good. I didn’t build it, not because I’m lazy and, didn’t want to, I haven’t built it because everything keeps moving so quickly. I’ve been trying to figure out like what is the most impactful thing I can say about the state of AI and writing. And it’s obviously changed just in the last two weeks, like we had.
[00:02:54] Paul Roetzer: Yeah. You know, G B T four comes out and now we have plugins, which we’re going to talk about, and it’s just such a rapidly moving space and there’s so many. I want to say, and I think need to say, and yet I got like 25 minutes to do it after. I do like the introduction , so there may be a continuation of the AI for Writer Summit, but we have a great event planned.
[00:03:15] Paul Roetzer: This, this, episode is brought to us by the AI for Writers Summit. It’s just AIWriterSummit.com. Is that, what is it? AIWriterSummit.com. There we go. So I’ve gotta talk, Mike’s gotta talk on tools and technologies may have be from writers. going to be talking about, you know, the impact of AI and writing teams.
[00:03:33] Paul Roetzer: We have a fireside chat with Ann Handley. We have an amazing panel, talking about the future of AI and writing. So that’s all coming up, Thursday, March 30th from 12 to 4:00 PM If you can’t join us, there’s going to be an on-demand option, so it’s free if you, attend live. There’s a paid on-demand option, so if you, if you can’t make it for whatever reason, you know, just, you can grab information from the site about how to get the information.
[00:03:58] Paul Roetzer: And the class on demand, it goes from 12 to 4:00 PM Eastern time. So check that out. If you’re interested, if you’re a writer, editor, publisher, you’re managing a content team, we’re going to try and get into, we’re not going to have all the answers, like I will tell you right now. We do not have answers for all of this, but we’re going to try and at least ask a lot of the hard questions and provide some guidance on.
[00:04:21] Paul Roetzer: Impact on individuals and teams and use cases and technologies and try and just have an open discussion on where this is all going. It’s going to end with a one hour q and a with some of the speakers. And I think we’re all going to be learning together, honestly, like just every day that comes, every new technology or new advancement, it’s just trying to process it in real time and it’s a lot to to take in.
[00:04:43] Paul Roetzer: So yeah, join us for that aiwritersummit.com and learn more about that event. A quick thank you to all of our sponsors. Our presenting sponsor is Writer at writer.com. GoCharlie, Visla, HyperWrite rasa.io, Demandwell, GlossAI and copymatic. We really appreciate them stepping up, stepping up for an inaugural event and supporting it and making it all possible.
[00:05:05] Paul Roetzer: All right. With that I’m going to turn over to Mike. If you’re new to the show, we do three main topics and then we end with a rapid fire. We were remixing right before we got got on it. We are recording this on Monday morning, March 27th. . It’s just everything is changing so quickly. It’s like, what do you even talk about sometimes?
[00:05:23] Paul Roetzer: Which of these do we choose? So, Kick us off, let’s
[00:05:27] Mike Kaput: go. Will do. And the good thing is we always have the audience covered to the point where sometimes we will rewrite the script, you know, minutes before the show. So it is up to date,
[00:05:36] Paul Roetzer: which we may or may not have done for this one. .
[00:05:40] Mike Kaput: So first up, get ready chat.
[00:05:43] Mike Kaput: GPT now has eyes, ears, and internet access. So OpenAI just announced a plugin system for ChatGPT, and this plugin system enables it to interact with the wider world through the internet. The plug-ins are developed in large part by third party companies like Expedia, Instacart, and Slack. And they allow users to perform a variety of tasks that use these sites from right within ChatGPT.
[00:06:12] Mike Kaput: So you can be instructing ChatGPT to accomplish tasks and it will rely on plugins from these sites to do so. So OpenAI itself is actually hosting three of these plug-ins. One is a plug-in that gives ChatGPT access to up-to-date information On the internet, there’s a Python code interpreter and there’s also a retrieval plugin that allows users to ask questions of your own documents, files, notes, emails.
[00:06:43] Mike Kaput: Also for public documentation. So what’s also really important to note here, I think, is that of the plugins initially announced, one of them is a plugin with Zapier, which itself integrates with thousands of other tools. So acting as kind of a force multiplier and how many tools you can start using Chachi p t with right out of the box.
[00:07:03] Mike Kaput: Now as of today, there’s a wait list to access the plugins for developers and ChatGPT plus users. So we are eagerly awaiting the ability to connect ChatGPT to elements of the internet and also other really popular services that we use every day. So I wanted to kick things off Paul, and ask you, did we just open up a whole new world of AI use cases for marketers, business people, individuals?
[00:07:30] Mike Kaput: Like which, what should we be thinking about
[00:07:32] Paul Roetzer: here first? Yeah, it’s a really big deal. So the main things, if you look at the announcement page, there’s actually like, they have their documentation, kind of the explanations on what you were just going through. I think the key thing to focus on is, as you said to, to right now, GPT-4, it’s training data basically ends around September, 2021.
[00:07:57] Paul Roetzer: So anything that has occurred, any new information, new context, events, none of that is in its knowledge base. And so first obviously connecting to the internet, the browsing capability is key. Now I actually do have access to that. So if you’re a ChatGPT plus user, you can join the wait list. I got access pretty quickly.
[00:08:16] Paul Roetzer: I want to say it was like three hours after they announced this, I was able to get the browser access. I do not have any of the other plugins, so I can’t speak specifically to them. But the browser one works really well. And it’s interesting because it’ll actually tell you like it what page it’s looking at to find the question.
[00:08:31] Paul Roetzer: So let’s say like right now we’re in what, the final four for men’s basketball? College basketball. So if I were to say which teams are in the final four, it may actually show like querying espn.com. Like it’ll tell you where it’s looking for the information, summarizing the information, and then it actually spits it out.
[00:08:49] Paul Roetzer: And so you can actually see now where did it get this from? How did it summarize this information? And you can actually start to see how it might actually drive traffic back to the source a little bit more than before where you just like weren’t really seeing where it was getting the information from.
[00:09:03] Paul Roetzer: So just that on its own is a really practical use case. But previously all it could do was emit text. . And so you had alluded to this, but this is moving into that like world of bit stuff we talked about in a previous episode where the AI agents can actually start taking actions on your behalf. So whether it’s booking trips or you know, getting you a reservation at a restaurant, like as you start plugging in all of these different applications and data sources, you can actually not only access that proprietary information, you can take action or the agent can take action on your behalf.
[00:09:41] Paul Roetzer: So that’s the thing that’s like really key. So on the browser aspect where it’s getting real-time information, it’s can start to solve, you know, we’ve talked about this hallucination issue, or just make stuff up that enables you to start solving for that. But that access to proprietary information becomes a really critical component.
[00:09:59] Paul Roetzer: And so like when it first came out, you know, you, you kind of watch for what are other people saying? So a couple, there’s this, guy Dr. Jim Fann, who’s aI scientist at Nvidia and he tweeted Open, I just announced ChatGPT plugins. If Che GT’s debut was the iPhone event. Today is the iOS app store event.
[00:10:19] Paul Roetzer: And that was kind of a universal thing. I saw that a number of times. There was another one, Dharmesh, said the GPT-4 launch was big. The ChatGPT plugins launch is even bigger. OpenAI took the chat interface and turned it into a chat ecosystem. It’s the app store for chat. So I think for people are trying to like understand, well what actually is it?
[00:10:41] Paul Roetzer: Imagine your iPhone before it had all the apps and the utility you have in your iPhone with all of these apps. But rather than you having to keep like going into all these apps, you’re just going to do everything through text communication. So you’ll be in the ChatGPT interface. And anything you want, you can just ask it of.
[00:11:00] Paul Roetzer: So maybe you plug in your analytics data or your c r m data or your sales pipeline information and you can just ask questions of it without having to like leave that interface. And the Zapier integration you mentioned is a critical one because they have, what, 5,000 or something app integrations. So on their site, like there’s a page we’ll link to, they say we’re excited to share that.
[00:11:24] Paul Roetzer: Zap your chat. GPT plugin is part of the first batch of providers in this new ecosystem. This new product now in beta can pull in thousands of apps from your tech stack and allows you to automate tasks directly with chat GT’s interface. So again, you’re in ChatGPT and your browser, and you can now have access to anything Zapier is connected to in your tech stack.
[00:11:45] Paul Roetzer: It’s powered by their natural language actions. A p i, again, actions is a word you’re just going to keep hearing over and over again, that you can use simple natural language to complete actions in other apps. You can ask, ask it to execute any of Zapier’s 50,000 actions, a search, update, write whatever it is.
[00:12:04] Paul Roetzer: Turns your chat into action. It can write emails, then send it for you. Find context in your c r m, update them directly. Add rose to a spreadsheet, you know, slack message, anything you can imagine you can do. And then they kind of end with that. The chat GP plugin for Zapier is just the beginning, but it can already extend the power of AI chatbots, allowing users to go beyond simple conversations and instead perform business critical tasks with thousands of apps.
[00:12:29] Paul Roetzer: After all, if chatbots can perform actions in the real world, the sky is your limit when power powering your work. So I think that that to me is the biggest takeaway here is like not only. All of a sudden ChatGPT is a platform like now we’re going to start doing more and more things. We’re not just going to be writing content or doing searches.
[00:12:49] Paul Roetzer: You’re going to be able to interact with all these plugins and it’s going to be just a flood of plugins. Like it’s going to be wild how quickly these things come in. And the other thing I’ll say is it kind of reminds me of like what Alexa wanted to be with all those skills and like how you were just supposed to be able to ask Alexa for stuff, but then you never had a clue what it actually knew how to do or what it was connected to.
[00:13:11] Paul Roetzer: And so for me at least, Alexa just never delivered on the utility. It was promising. And I just stopped using it really early on. And it’s kind of like what Siri could have been or maybe still might try and be, but I don’t know. I mean, it’s a major move by them to become a true utility and all through text, you know, you know, language and chat.
[00:13:35] Paul Roetzer: It’s kind of fascinating.
[00:13:37] Mike Kaput: Talk to us a little bit more about the App store, i i, the OpenAI app store. I mean, it feels like this is a game-changing ecosystem. I mean, like you said, it turns this into a platform. What are the implications here? Are we going to see every single popular service and company have a ChatGPT plugin?
[00:13:58] Paul Roetzer: Sure. Seems that way. I mean, just the, again, this is all, what is this, like 72 hours old ? Like it’s, it’s pretty fresh. But I mean, my immediate thing was I started thinking about. Analytics data. , like the first obvious use case to me is like, let’s say you, you use QuickBooks or, or Stripe or even Google Analytics, I thought, I don’t know if it would work with Google, but think about all the places where you have data living and that you have to go in and look for reporting.
[00:14:26] Paul Roetzer: So like, I gotta go into HubSpot, I gotta go into QuickBooks for this and HubSpot for that and Salesforce for this and my social media platform for that. And like, just imagine it all is just connected to ChatGPT instead. And rather than having all these places you go log into, you can literally just ask anything of the ChatGPT interface and it’ll just pull from the appropriate plugin.
[00:14:50] Paul Roetzer: So now I don’t necessarily have to go log in to find that information. And you know, we’ve talked about it on the show before. How many times are we zooming each other, like sending a quick message in Slack or whatever and saying, Hey, like, where are we at with AI for writer Summit registration? How many people did you know, registered yesterday after the newsletter went out?
[00:15:08] Paul Roetzer: What was the open rate of that email? And I just gave you three different tasks. And now it’s like, oh man, okay. And now we gotta go in, someone on the team’s gotta go in and like pull three different reports and you spend like seven minutes. But minimum just, just to like grab three data points right now.
[00:15:21] Paul Roetzer: Imagine that same scenario. I’m just in Chad, GPT at 12 o’clock at night. Like whatever, I’m the middle doing something. I think, oh, where are we at with that? And I just ask chef GPT instead. And it’s connected to HubSpot and it’s like, Hey, where are we at with this registration total? Spits it out. How many people you know registered today spits it out?
[00:15:37] Paul Roetzer: Did anybody register at companies with more than a billion in revenue? Spits out a list of those people. . It’s just like, so now we truly get to that world of Bits concept where not only am I have access to realtime information, but I could say, Hey, you know, send the team a reminder tomorrow morning to follow up with anybody that’s over a billion in revenue.
[00:15:55] Paul Roetzer: Like those are I c P, whatever. So now you can start to actually do strategy and action on top of this data set. And so that’s, you know, it’s really, I haven’t had time to honestly sit down and really think about this deeply about all the implications of this. Yeah. But I do think it’s just going to explode in terms of how many plug-ins there’s going to be, and the utility of those plug-ins.
[00:16:18] Paul Roetzer: It’s going to create massive complexity again, because nobody knows what to do about this. Like, again, we’re still trying to explain what language models are to people and what ChatGPT is like, I’m like, when we do our intro to AI for Marketers class, I think last time we had 81% of people said they had experimented with ChatGPT.
[00:16:36] Paul Roetzer: We asked that question up front, and that was last week. When I was at the pavilion event, the CMO summit, I asked and I don’t know, like 20, 30% of the people in the room had done it. So these are the, this is the c. So, I mean, we’re still at a moment. While as you know, people listening to this podcast are probably racing ahead.
[00:16:56] Paul Roetzer: You’ve been experiment with ChatGPT since the day came out. You probably got DALL-E and Mid Journey and Stable Diffusion and like, you’re probably a le, you know, early adopter of all this tech. And this isn’t new to you, but I’m telling you right now, the agencies Mike is talking to, the CMOs I’m talking to, they don’t know all this stuff.
[00:17:17] Paul Roetzer: They’re, it’s abstract to them. They’re still afraid of it in a lot of ways, and they’re not racing forward and testing it like you may be. And so I think sometimes we even get caught in our own little bubble because we talk about the stuff all the time and we see it and we experiment with it. That is not the norm.
[00:17:34] Paul Roetzer: Like, don’t assume that people in the marketing and business world have even remotely figured this stuff out yet because they have not. The amount of change we’ve seen since December 1st of last year is. It is literally like five years of innovation packed into four months . And a lot of people have not caught up yet.
[00:17:55] Paul Roetzer: I sometimes I feel like I’m not caught up. It’s like you turn the internet off for like four hours during the day and you’re like, well, what the hell just happened? ? Like, yeah, I missed how many announcements in four hours. Like, okay. So yeah, I think it’s just going to be, it’s, it’s going to be a really big deal and I’m not sure we really comprehend how big of a deal yet.
[00:18:17] Paul Roetzer: So
[00:18:17] Mike Kaput: before we move on to our second topic, I wanted to really quickly hit on this one point which, and you can talk as much or as little about this as you see fit, but I felt like when I read this announcement in one fell swoop, OpenAI just caused an existential threat for thousands of existing startups, plug-ins, services, features that people are building.
[00:18:39] Mike Kaput: How does this impact builders, founders, or
[00:18:43] Paul Roetzer: investors? Yeah, I think. That’s probably right. There was David Sachs, who, you know, is one of the early PayPal guys and, early investors in Facebook. And you know, we’ve, we’ve talked about him before. They have the All-In podcast, which is like the number one business podcast in the world.
[00:19:00] Paul Roetzer: He had tweeted plug-ins could create an interesting network effect for OpenAI developers provide the AI with more info in order to access capabilities. AI gets smarter. Developers want to use the smartest ai. AI sucks in all the world’s knowledge. So it’s like, oh, just, just this little thing. . So I think there’s going to be a race to build on top of this.
[00:19:19] Paul Roetzer: Cause I think it’s where people are going to go. But it does then create these challenges for software companies. It’s like, well, are people going to stop coming to our platform then? Like, is it going to change the way people use the software? We. . Is it going to change how quickly existing software companies could be obsoleted?
[00:19:38] Paul Roetzer: Like does it, does the threat matrix actually did it just move on these software companies? Because there was a, there was one other tweet I saw. And this is a little technical, but like, just bear me with it for a second. It says, I’ve developed a lot of plug-in systems and the OpenAI jet GPT plug-in interface might be the damn craziest and most oppressive approach I’ve ever seen in computing in my entire life.
[00:20:03] Paul Roetzer: For those who aren’t aware, you write an OpenAI manifest for your api, which again, I don’t do this stuff, but that’s not very hard to do if this is what you do. Use human language descriptions for everything and that’s it. You let the model figure out how to do everything else. So basically you just say, here, OpenAI.
[00:20:21] Paul Roetzer: We have, we have all this datall this information living up here in this c r m or whatever it is. , here’s our open A, our p i, you connect the things, and then you basically just tell it what you want it to have access to, and then it just does everything else. So basically the, my takeaway is it’s insanely simple to build plug-ins is pretty much what this says.
[00:20:42] Paul Roetzer: So as chat chip, plug-ins are super simple to implement, basically just document your api, but for a language model rather than human. And let me just, I want to make sure I give proper attribution here. This is, Mitchell Hashimoto, who is founder of Hash Corp. So, and that was retweeted by Greg Brockman, the c t o of OpenAI.
[00:21:04] Paul Roetzer: So I know he is legitimate source. I’m just picking some random person here. So yeah, I think it’s going to be very disruptive. I don’t, I don’t know yet how, or how exactly it’s going to play out or how quickly, but it sure seems like, again, if you’re a software company, like every week, you’ve just gotta be like, Head on a swivel.
[00:21:22] Paul Roetzer: Like, what, what is going on? I gotta imagine SaaS companies saw this coming, like the, nobody’s talking about this that I knew of right before this happened, but it sure seems like an obvious play in retrospect that they would do this. So
[00:21:39] Mike Kaput: if ChatGPT plugins weren’t enough for your software company or individual to worry about, our second topic today also concerns OpenAI and GPT-4.
[00:21:50] Mike Kaput: So to set this up, back in episode 36, we talked about OpenAI publishing a blog post, essentially warning about the risks and considerations around artificial general intelligence or agi. Now there are a lot of different definitions and a lot of lack of clarity around what AGI actually means. Massive arguments within the industry.
[00:22:16] Mike Kaput: If something like agi In whatever definitions we talk about, it’s even possible. But OpenAI defines this concept of agi I as either highly autonomous systems that outperform humans at most economically valuable work or in other places on their website they’ve said AGI is AI systems that are generally smarter than humans and benefit all of humanity.
[00:22:40] Mike Kaput: Now, in both cases, the reason they’re publishing these definitions is OpenAI has a stated mission to attempt to bring AGI into existence. And the reason we’re talking about this is because, this past week, a team of Microsoft AI scientists claim that GPT-4, the latest version of OpenAI’s language model, which came out, I think 10, 12 days ago, that it exhibits what they say are quote sparks of human level intelligence or a g I.
[00:23:12] Mike Kaput: Now as we’ll discuss a little bit, they don’t really get into a one or two line definition appears exactly what AGI is. But they argue that GPT-4 S impressive performance in a wide range of tasks like math, coding, and legal exams indicate its potential as an early version of some type of AGI I system.
[00:23:33] Mike Kaput: Now, again, while some argue that even this entire concept of a agi I as a pipe dream, others like Sam Altman, CEO at OpenAI believe that if it is doable, that it would usher in a new era eventually for humanity. And this research seems to indicate that at least some very smart people at some very big companies in AI believe we are.
[00:24:00] Mike Kaput: Developing the very beginning of such a system. Now, I just want to kick off kind of Paul, when you read this, how legitimate is this? Should we be taking it seriously at all?
[00:24:10] Paul Roetzer: Yeah, so I think I’ve shared on the podcast before, like some of the ways that I try and stay at sort of the forefront of what’s happening is you monitor the latest research papers.
[00:24:21] Paul Roetzer: So when this one sort of hit my , hit my radar on Friday. Now keep in mind, Friday I’m in San Francisco, taking a red eye back from this event. And so I’m like trying to think, okay, do I have the energy on this red eye to read 154 page research paper from Microsoft? I tried, I failed. I did not make it through the paper at as I was flying through the night across time zones, but, It’s from Microsoft.
[00:24:55] Paul Roetzer: So they had early access to this technology and the abstract, if you never read one of these research papers, they start with an abstract that’s usually, you know, between two and 400 words, maybe 500, and then the conclusion. And so oftentimes what I’ll do, because these things can get pretty dense, they’re the often highly technical, you can just read the abstract and the conclusion and then you can kind of follow even just the subheads.
[00:25:17] Paul Roetzer: Like you can kind of scan this thing and get a sense, is this a really important paper? But more importantly, often is, where is it coming from? Like who, who are the sources for this? And so this one did capture my attention immediately because it was from Microsoft research. And so I’m just going to, I’ll read a few quick excerpts from the abstract.
[00:25:35] Paul Roetzer: Now again, we’re, we’re not going to go deep on AGI today. We’re not going to like, you know, get kind of more than a surface level here. But I think what’s really important. We, we’ve talked before, this is why OpenAI exists. It’s why Google DeepMind exists. It is why many of the top researchers in the world work for these labs is because they believe AGI I is possible and they’re pursuing it.
[00:25:56] Paul Roetzer: Now, there are some that do think we’re actually getting quite close, and that was why when I saw this, I was like, well, wait a second. This is the first time I’ve actually heard someone say they think this is, this might be it. And that’s why I was like, oh, and that’s Microsoft saying this like okay with is, is OpenAI in agreement?
[00:26:13] Paul Roetzer: And so I’ll come back to that in a second. But here’s the quick quick excerpts from the abstract. So artificial intelligence researchers have been developing and refining large language models that exhibit remarkable capabilities across a variety of domains and tasks, challenging our understanding of learning and cognition.
[00:26:31] Paul Roetzer: With the latest GPT-4 model that they were experiment with, we contend that. Early version of GTP four is part of a new cohort of large language models along with ChatGPT in Google’s palm, for example, that exhibit more general intelligence than previous models. We demonstrate that beyond its mastery of language, GPT-4 can solve novel and difficult tasks that span mathematics, coding, vision medicine, law, psychology, and more without needing any special prompting.
[00:27:02] Paul Roetzer: Moreover, in all of these tasks, GPT-4’s performance is strikingly close to human level performance and often vastly surpasses prior models such as ChatGPT. Given the breadth and depth of GPT-4 S capabilities, we believe it could be reasonably viewed as an early parenthesis, yet still incomplete version of an artificial general intelligence system.
[00:27:27] Paul Roetzer: We discuss the challenges ahead for advancing towards deeper and more cons, comprehensive versions of agi. I. Including the possible need for pursuing a new paradigm that moves beyond Next word prediction, which is what large language models do. So when you just read that abstract, they’re wording around it a little bit.
[00:27:47] Paul Roetzer: Like I think they’re trying to not be definitive that this is indeed Agi I, but the original title of this thing that was still embedded in the code was something to the effect of like first contact with agi and then they changed it apparently at the last minute. So it almost feels like they wanted to be more definitive, that they believe this actually is agi.
[00:28:11] Paul Roetzer: I, so again, I, we have said multiple times in this episode. I don’t even, or this, this show, I don’t even know that it truly matters if we ever achieve universally agreed upon agi because the definition is, is so vague. Just look at what is becoming possible with whatever it is we have currently landed on whatever G PT four is.
[00:28:32] Paul Roetzer: If, if there’s a spectrum of agi and some people think we’re 20% of the way there, or 80% of the way there, like it doesn’t really matter to me. Like, look at the power it already possesses and we know it’s early. And so, like, I’m still listening to the two and a half hour Sam Altman interview with Lex Fridman.
[00:28:50] Paul Roetzer: Yes. And he asks them about it. And Sam’s like, I, you know, if, if in retrospect someone tells me or if in, I think he said, if an Oracle told me that GPT 10 was an agi, like I’d believe it. Like he, he sees what they’re doing as a path to get there. And he thinks that there’s some things that might be needed to truly achieve it, but he also wouldn’t be surprised if the path they’re on actually does a achieve it.
[00:29:19] Paul Roetzer: So again, I think it’s just. You know, I think there’s listeners to this show who, who want to know what’s coming around the corner and want to be ready for it, not just solve for today’s use cases and technologies. And this is going to be an ongoing, really important topic to pay attention to and to monitor the progress of, because it’s going to be really hard to prepare for this if you’re not part of the process.
[00:29:46] Paul Roetzer: And that’s why G, that’s why Open Eyes is doing what they’re doing, where they’re releasing things in public before they’re really ready for primetime. Like GPT-4, they want to condition society for the the big thing. And so they’ve said before, like, we don’t want to just drop AGI in the world and say, here, figure it all out.
[00:30:04] Paul Roetzer: So they’re trying to release these virgins that may be perceived as moving very close to agi. I so that we have a chance as a society to figure out what this all means.
[00:30:16] Mike Kaput: And I think we’re doing a version of that or attempting to, in discussing this topic at all for the audience. Because at some point, if we’re talking about agI, as in some version or whatever you want to call this thing as a reality today, you’re going to see more developments around it.
[00:30:34] Mike Kaput: And now is the time to start understanding, at the very least, that we’re talking about a very real research paper looking at very real capabilities of a very real model that came out two weeks ago. We’re not talking about the Terminator or, you know, killer robots or anything. This is,
[00:30:51] Paul Roetzer: or some random Twitter thread from some random user.
[00:30:53] Paul Roetzer: Exactly. 200 followers. Like this is a legit stuff. It’s not peer reviewed yet because it’s published on archive, but it it’s legitimate technology with 154 pages worth of charts and experiments and, you know, it’s, it’s.
[00:31:07] Mike Kaput: It really is. It’s I didn’t realize we’d be having the AGI conversation seriously this early.
[00:31:13] Mike Kaput: I thought we had maybe a couple more decades, but who knows?
[00:31:17] Paul Roetzer: at least a couple more years. Yeah.
[00:31:19] Mike Kaput: Right. So this is actually really good segue into our third main topic, because
[00:31:24] Paul Roetzer: it does show, bring it back to Earth. Now, let’s come back to, we’re coming
[00:31:28] Mike Kaput: back to reality and the reality itself is pretty impressive and it shows what’s possible when these systems become really, really good at a lot of different things.
[00:31:37] Mike Kaput: So I want you to imagine for a second using AI to complete a massive business project. In just 30 minutes, I want you to imagine you could accomplish tasks that would take humans hours or even days. This is of course, something that actually happened in a remarkable experiment. A Wharton professor, Ethan Mo.
[00:31:59] Mike Kaput: Used a combination of AI tools to market the launch of an educational game. So in 30 minutes he did everything using just natural language prompts and inputs from, he conducted market research, he had AI create an email campaign. He had it designed a basic website. He had it craft a social media campaign, and he completed multiple other tasks in half an hour.
[00:32:24] Mike Kaput: Now the results of this, which will link to the full description of, in the show notes, really just showed the unprecedented potential of AI as your kind of co-pilot, as this multiplier of human effort, which has huge implications that we’ll talk about for the future of work productivity and creativity.
[00:32:45] Mike Kaput: So I think it’s really helpful and jaw-dropping to quantify the outputs that happened over the course of this 30 minutes. So over the course of these, this 30 minute, you know, Blitzing through an entire marketing program. Molik used no more than 20 inputs, actions, or prompts, and he as a result, generated 9,200 words of content.
[00:33:10] Mike Kaput: He got a working H T M L and c s s file. He generated 12 images. He generated a voice file, a movie file, and he used these assets in tandem with AI to create a marketing position, document, an entire email campaign, the website, logo, script, and video and social as he put it. AI would do all the work. I would just offer directions.
[00:33:33] Mike Kaput: So I kind of was pretty stunned reading this like he shares in the entire post, you know, the quality of the outputs, and what he was able to do within 30 minutes. I was pretty stunned personally. I mean, we’ve always known these tools can increase productivity. In an exponential fashion, but seeing it actually happen was another thing entirely.
[00:33:56] Mike Kaput: So I just wanted to kind of start off by addressing the elephant in the room. Is this the new normal for
[00:34:01] Paul Roetzer: marketers? Yeah, I mean, the tangible takeaway for me is obviously it was an experiment, but I put this on LinkedIn on a Sunday morning. So like, not everybody’s checking LinkedIn on Sunday mornings,
[00:34:16] Paul Roetzer: So I shared this with a little bit of context, 20, not even 24 hours later when we’re recording this, that LinkedIn post has 48,500 impressions and 270 engagements and 49 reposts. So this resonated with people, is what I’m saying here, like it is that back to earth tangible example of how this can be used.
[00:34:41] Paul Roetzer: And he did it with a collection of tools that combine costs less than $60 a month. You and I worked together at my agency for a really long time. I owned a marketing agency for 16 years. If I had to break this down and scope this in a proposal, I can’t fathom this list of things wouldn’t have cost between 10 and $20,000 less than all those things you just explained.
[00:35:04] Paul Roetzer: Minimum 10 to $20,000 and probably a month of work. He did it in 30 minutes. Now is it perfect? Would you actually launch the campaign with this? Probably not. You probably need to do some editing and the website might need some adjustments and you gotta do Q, you know, quality testing and all this stuff.
[00:35:21] Paul Roetzer: But I mean, what maybe like another five hours of work, you’re going to knock this thing out. So I think like if you pay an agency to do this work, if you are an agency that gets paid to do this work, or if you’re just a corporate marketing team or maybe an individual contractor, think about how much time it would take you to do this in a traditional vehicle.
[00:35:37] Paul Roetzer: And he used four AI tools that cost 60 bucks. Yeah. To do this entire experiment. So I think that to me is like, It’s just representative of the fundamental transformation we are going through at the moment. And some of the comments on my LinkedIn thing, like there’s people who are like, why are people Harding this?
[00:35:55] Paul Roetzer: Like, why are they cheering this on? Like, this is terrifying. Like this is what we do for a living. And like you have people in these comments like, oh, this is amazing. It’s so great. And I just, I didn’t take a stand on it one way or the other. I was just sharing the information and say, Hey look, this is really interesting.
[00:36:08] Paul Roetzer: And you do, you have some people in the comments like, this is incredible. Like this changes everything. It’s so great. And other people are like, oh my God, what’s, what are humans going to do when the machines do all the work? Like, do we need universal basic income? and you have like this, like the fear factor emerges, but these are real concerns and questions that we don’t have answers to.
[00:36:25] Paul Roetzer: And so it’s also representative of how unprepared the business world and society and the economy are for what’s about to happen. So that’s why I actually love the experiment. I thought it was just, Thought provoking. Yeah. And I do think it’s helpful to like really see how tangible this is. I’m going to read you a couple of the prompts that he used.
[00:36:45] Paul Roetzer: So he has AI as marketing strategist. Now they had created this Saturn Parable game, so I was able to like know this, but he said, look up the business simulation market. Look up Wharton Interactive Saturn parable. So it’s like, okay, cool. And it went and found it. And he says, pretend you are a marketing genius, we are going to launch the Saturn parable.
[00:37:05] Paul Roetzer: You should give me a document that outlines an email marketing campaign and a single webpage to promote the game. It does it. Then he goes on to give some additional prompts. It creates four additional emails. Then he goes, AI is a site developer. Outline the webpage and what text and graphics it uses.
[00:37:23] Paul Roetzer: You do not need to create the webpage, but do give me the text so it writes a webpage for him, . Then he goes on and says, you are an expert site designer. You are creating the launch announcement page for Saturn Parable outline below. And then he gives like the outline, create the following webpage, make it an HTML page that I can run on my computer list any additional assets I will need to make it work and where to put them.
[00:37:46] Paul Roetzer: So again, if, if you are listening to this and you’ve never done the prompting of an AI agent, this is how it works. Like he’s telling it the system, you are this system. And then he’s saying, here’s what I want you to do system. And then you can kind of keep iterating on that information. So then he needs a hero image.
[00:38:03] Paul Roetzer: So he asks it to create an image. It does it Mid journey shows the completed website. He’s like, oh, this doesn’t look great. So I tell the designer, the hero image looks weird because it’s magnified in the top and bottom. Can you fix it? So it fixes it in the c s s file. Then he goes, and as his AI is a social media manager, can you write me the social media campaign?
[00:38:22] Paul Roetzer: I need to promote this thing using, or this using the Wharton accounts on social. So it creates it and it does goal, audience strategy, content plan. Like it builds a plan. Like it doesn’t just give the, and then it goes, show me some example posts. So it writes sample posts for Facebook and Twitter, and then finally it uses 11 labs and d i d d dash, I don’t know if it’s dead or whatever that is.
[00:38:45] Paul Roetzer: But they actually, he creates a script and then he creates a synthetic actor to record the video. And then it goes into, like, as you outlined all of this in 30 minutes, like. It’s wild. Like it’s really hard to comprehend, but I think it makes it very tangible. Again, we don’t need AGI to change everything GPT-4 in its current environment without the plugin ecosystem
[00:39:12] Paul Roetzer: can enable stuff like this. Like where are we going to be in three months, six months, 12 months? I have no idea. Like as much as we live this stuff every day, it’s really hard for me to wrap my head around how much this is really going to change the future of work, the future of business. Like it’s hard to think about
[00:39:33] Mike Kaput: So all that being said, I don’t want to put you on this pop. Do you have any kind of first draft thoughts about what you’re thinking about when it comes to employment or productivity or work as a whole? I mean, as someone who has owned a business, owns a business, and employs people and has those issues
[00:39:51] Paul Roetzer: top of mind.
[00:39:53] Paul Roetzer: Well, we’ve talked about it a little bit before. My biggest fear at the moment is a lack of understanding of this technology, what it’s capable of, what its limitations are, and that the people at the top of companies and agencies and media, you know, outlets and publishers, whatever, the ones that are going to make the decisions about the impact on staffing probably don’t understand the technology driving those decisions.
[00:40:14] Paul Roetzer: So I think, and given the economy at the moment, I think there’s going to be way more pressure on reduction of workforces and AI is going to believe to be an ability to continue to maintain productivity levels while reducing workforce. Hasn’t that been what the
[00:40:32] Mike Kaput: tech companies themselves have been saying that Microsofts and Facebooks?
[00:40:37] Paul Roetzer: Well, I think Meta was the first one to. Not so overtly say we’re going to use AI instead of the 10,000 people. Everybody else. I think a lot of the layoffs were basically just like, if you look at Twitter, I mean that was like the, you know, the prime example for everybody that was just like, this is just a bloated organization.
[00:40:54] Paul Roetzer: We don’t need all these people. I don’t know that like Elon Musk had some grand plan to replace them with AI agents or anything like that. And then I think a lot of tech companies were under massive pressure to follow that lead because then their investors in the stock market, you know, wall Street, were pressuring them to make cuts because, well, if Twitter can function with 80% less people, like you can certainly function with 20% less people.
[00:41:18] Paul Roetzer: And so the stock prices were getting hammered because these tech companies were now perceived to be bloated in terms of staffing. I don’t think it had anything to do with ai. And that’s my concern is like, that wasn’t even AI cut. Same thing with, I’ve always worried about media companies, like as tough as media models have been, it had nothing to do with AI the last three to five years.
[00:41:36] Paul Roetzer: You had all these layoffs of journalists and, you know, producers and all this, and it’s like, wasn’t even AI yet. So I think we’re now heading into the AI phase of workforce assessment. And I think it’s going to be a mix of pressure to drive efficiency and productivity with fewer people is going to cause job loss and an unprepared society that didn’t realize AI was coming for knowledge work as quickly as it is.
[00:42:04] Paul Roetzer: And so I do, I wish I had a more optimistic view in the near term, but I think it’s, it’s going to get ugly in the next year. I think there’s going to be a lot of disruption, to the workforce. I think a lot of people’s career paths are going to get shaken up a bit. And the best thing I can say is like, It’s scary, but I would just be really, really proactive understanding this stuff because I do think that the people who seek the knowledge and understand this technology and can be the ones that apply it, they’re going to have the best chance.
[00:42:41] Paul Roetzer: Like, I’m not telling you like it’s bulletproof, that if you go figure this all out, like your current employer is going to appreciate that knowledge and give you a promotion and like, you’re going to keep your job when everybody else is struggling. Not saying that, but I’m saying your chances of being employed and having enormous career potential ahead of you and benefiting from all this change is only going to be possible if you figure this out.
[00:43:07] Paul Roetzer: So it’s, the only thing I can say is like, I always, like, I even teach this to my kids. My kids are 11 and 10, like, life is messy. It’s imperfect, it’s hard. And the only way you can get through it is to control all the variables you can. So like when, when life is hard or business is hard, you just have to buckle down and do like, what can I control today?
[00:43:31] Paul Roetzer: And what you can control today is learn ai, experiment it, become confident in your abilities with it. Be the one in your organization that understands this stuff. That’s the only control variable you have at the moment. You can’t, you can’t control how quickly this happens and you know what it’s going to do in society and the workforce and whether AGI i’s going to come around.
[00:43:51] Paul Roetzer: You have no control over that, but you can control your own career path and what you choose to do with this knowledge. And I think that’s the best thing you can do. It’s, but that’s what it’s always been. It’s just always learned if you get stagnant as hard as it is to keep up. But if you get stagnant, then you’re, you’re not going to be putting yourself in the best position to kind of weather the next year or two as the economy is down and AI is coming for knowledge work.
[00:44:19] Paul Roetzer: Like that’s, it’s just the perfect storm right now. If the economy is humming along and. We weren’t having to lay people off for financial reasons, then maybe it’s not as bad. But I think the fact that it came for knowledge work so quickly and the economy sucks, it, I just, I don’t feel great about the next like 12 to 18 months.
[00:44:41] Paul Roetzer: And the impact it’s going to have on workers.
[00:44:45] Mike Kaput: I think that’s at least very good advice for us to be able to navigate kind of this new normal moving forward. And I think kind of in the interests of providing people with as much knowledge as possible. You know, we’ve got a ton of rapid fire topics. Might be a little bit of a fire hose realization , but there’s a bunch going on.
[00:45:04] Mike Kaput: I fully anticipate our rapid fire sections to get more and more populated as we see this accelerating rate of innovation. So let’s just dive into these real quick here at the end of the episode. So first step is a funding announcement from character.ai. So this is a rapidly growing conversational AI company.
[00:45:25] Mike Kaput: They just raised 150 million in a series A round led by Andreesen Horowitz. So they’re valued now at a billion dollars. They were founded by X Google engineers, and basically they allow you to create personalized AI companions and talk to them. So they have these pre-built companions that include, you know, ones that mimic famous people like Elon Musk.
[00:45:47] Mike Kaput: They mimic made up characters like Tony Stark, and some are just like overall AI helpers, like a programming assistant. And what’s really crazy here is how explosive their growth has. They launched in beta in September of 2022, and they have since become one of the top 400 websites worldwide in terms of traffic.
[00:46:10] Mike Kaput: And they have seen their user send 2 billion messages to these AI conversational companions. And what’s really crazy is a billion of those were in the last month alone. So Paul, I just wanted to get your thoughts on character.ai, just because the funding is significant, the growth is incredible, and it’s probably one of those companies where, you know, if you follow the space, you know of them.
[00:46:35] Mike Kaput: But to the wider world, we don’t really hear a lot about character.ai compared to the OpenAI of the world. So why is this company
[00:46:42] Paul Roetzer: significant? Yeah, so they, I mean, part of the reason is because they didn’t really officially launch. We talked about it on episode 25, it looks like on December 5th was when they tweeted like, Hey, introducing character.ai to the world now.
[00:46:56] Paul Roetzer: We’d been tracking them for probably close to a year at that point. Because one of their founders, Noam, he, he was one of the co-authors of attention is All you need, which is the paper we’ve talked about before from the Google team in 2017 that created the transformer architecture. That’s the basis for GPT.
[00:47:14] Paul Roetzer: So that was kind of like why they had been on my radar is once, you know, I knew that he had gone off to start that company co-found that company sort of been tracking the authors of that paper because, that’s Aiden Gomez. We’ve talked about Aiden before, the co-founder and CEO of CO here, which is another language model company that was also one of the authors of that paper.
[00:47:33] Paul Roetzer: Yeah. So that paper and that team, has sort of like gone on to become many of the influential leaders in the language model space because they all kind of played a role in not only that paper, but some of them worked on like Lambda, the language model at Google. So yeah, I think it’s, it, they’re a really big deal because.
[00:47:55] Paul Roetzer: Language models appear to be the foundation of all of the innovation we’re seeing in intelligence right now. Now, obviously image is going to play a role in video and code and all this stuff, but at the, at the core of it all is this transformer architecture. So the people who know how to build know what the capabilities and limitations are th they’re worth following.
[00:48:15] Paul Roetzer: And, while I have explored character that I haven’t actually like tested use cases and stuff with it yet, if I can’t talk specifically to here’s my experience and my personal thoughts on it, but there’s enough, traction around them and enough kind of pedigree of where this all came from that they absolutely weren’t paying attention to.
[00:48:37] Paul Roetzer: And even the subhead of that, you know, the article I believe was from our friend Cade Metz, at the New York Times. He said, it’s has the Silicon Valley companies among the few startups poised to compete with OpenAI. So again, if, if Cade’s telling us that, They’re on par with OpenAI or potentially coming at them, then you gotta pay attention.
[00:49:00] Paul Roetzer: Awesome.
[00:49:01] Mike Kaput: So next up is a recent study by computer scientists at the University of Maryland revealed some research that indicates the existing methods for detecting texts that’s generated by these large language models may be very unreliable. So they basically argued that, you know, even subtle paraphrasing of the text that some of these models put out can significantly reduce our ability to detect the accuracy of whether or not it’s human or AI generated.
[00:49:32] Mike Kaput: So basically what this means is it sounds like many tools that claim to be able to detect AI generated content with a high degree of accuracy probably aren’t actually able to do so, or are very easily thwarted. So when you read this, is there were you. Surprised by anything here or is, I mean, is there any credible way to consistently detect AI generated
[00:49:55] Paul Roetzer: text?
[00:49:56] Paul Roetzer: I have yet to see anything that says this is possible. Like every re every study we’ve seen, every research paper, pretty much confirms that this is not a reliable way to figure this out. So whether you’re using it to find papers written at university by students, or trying to like, identify misinformation, what we’ve always assumed is it’s going to be AI versus ai.
[00:50:20] Paul Roetzer: Like somebody builds an AI detection tool and somebody else builds a AI masking tool. Like it’s, I I’ve yet to see any credible source say that they think this is a solvable problem. Like, so I just, I would continue under the assumption that we cannot rely on AI to save us from aI think is what I actually tweeted when I shared this, is like, AI isn’t going to save us from ai.
[00:50:41] Paul Roetzer: Like, let’s, let’s, try and find another way here.
[00:50:47] Mike Kaput: All right, so another big announcement, Adobe Firefly. Adobe is launching a family of creative generative AI models called collectively Adobe Firefly. So basically it’s giving you the ability to generate images and stylized text, and these tools are going to actually be integrated into all of Adobe’s creative apps.
[00:51:08] Mike Kaput: So you’re going to have generative AI in Photoshop, illustrator, et cetera. Now here’s what jumped out to me as the really interesting part of this. Adobe ensures that their models are trained on copyright free licensed or Adobe stock library data. So basically preventing Adi potential issues that we run into around ownership and copyright with artists and brands.
[00:51:33] Mike Kaput: And they also have some plans to compensate artists who contribute to the training data. So I wanted to get your take on their approach to the copyright licensing and kind of artistic ownership issue here. Is this kind of a viable way forward to address some of the challenges we’re seeing in that space?
[00:51:49] Paul Roetzer: it sure seems like it. And I think this is similar to like we’ve talked about with AI writing, in recent weeks where it’s only a matter of time till the big legacy tech companies show up. And do everything, all these startups are trying to do. So, you know, Grammarly shows up, they got 30 million people, you know, 30 million users.
[00:52:09] Paul Roetzer: So Grammarly enables a writing tool and it’s like, oh boy. Like if you were making a writing tool and you’re trying to compete with Grammarly, that that gets pretty difficult. Google, workspace has, you know, everything baked in. If Microsoft 365 co-pilot has everything baked in, and now you get into like the design side, it’s like, okay, well Adobe, no stranger to ai, but kind of asleep at the wheel on the generative AI thing.
[00:52:33] Paul Roetzer: For a few months. But as we’ve said before, if you have proprietary data and you have distribution, meaning you have a massive customer base, you can not only catch up really fast, you can basically obsolete all the startups real fast. And so I think it just, it makes it a very complex scenario for early stage features, because I mean, reality, what’s happening is a lot.
[00:52:56] Paul Roetzer: Generative AI right now is building what end up being features in a larger platform. And so if Adobe shows up and they figure this out and they do it well, I mean certainly the demos are impressive or what it appears to be able to do is impressive. If it actually delivers on that, then you got some problems if you’re building a one-off tool in this space.
[00:53:18] Paul Roetzer: And so, yeah, I mean, it’s a big deal because Adobe’s a big deal, massive customer base. they all, and if they can bake this stuff in, and I don’t need to go outside of it to do, you know, AI for, slides and AI for, you know, landing pages and a, like, if I can just do it all in Adobe, great. Like I’d rather not have seven licenses.
[00:53:38] Paul Roetzer: That’d rather one. And that’s where I think like we’re going to, over the next 12 months, you’re going to either see this, you’re going to see an explosion of generative tools, but they may not last very long. , because if the platforms figure out how to do this and they have the data and distribution to own the market, then you could see like a massive consolidation back where it’s like, ah, it’s just.
[00:53:56] Paul Roetzer: rather than adding 25 generative AI tools in 2023, maybe you just like, oh, well Salesforce has it and HubSpot’s got it and Adobe’s got it. Like, we don’t need all these tools anymore. Let’s just do it all within these, and maybe there’s some sacrifices, like it’s not as feature riches. Maybe some of the purpose-built tools, but it’s enough where you don’t want to have to pay the other licenses again, especially in this economy, like you don’t have budgets to go get 25 new tools, right?
[00:54:23] Paul Roetzer: If the one you’re already paying $7,000 a month for has all the capabilities you need, so
[00:54:28] Mike Kaput: another. Great example of what you just said is that Nvidia is also making a play here. They announced a suite of cloud services designed to accelerate enterprise adoption of generative ai. So basically they enable businesses to create large language models and generative AI applications using proprietary data.
[00:54:47] Mike Kaput: They’re calling this, broadly NVIDIA foundations. So there are companies like Getty Images, Morningstar Shutterstock, who are kind of the pilot customers of using NVIDIA AI foundation services. And these span all the kind of generative AI use cases we just talked about, you know, language, images, video, and 3d.
[00:55:08] Mike Kaput: So we’re basically giving enterprises the ability to bake generative AI right into their own business models, using their own data and using their own proprietary services and apps. So, How important is it for enterprises to have this ability to create custom models, use their own data? Why is something like an NVIDIA Foundations better than just like using gen generic tools or GPT
[00:55:36] Paul Roetzer: four or whatever?
[00:55:37] Paul Roetzer: I think, I mean, the key takeaway for me here is just there are so many marketers and business people who have no idea who NVIDIA is or who Jensen the CEO is. Like, it’s, it’s almost shocking how little people know about Nvidia and none of the stuff we’re talking about. AI happens without Nvidia.
[00:55:58] Paul Roetzer: They’re their G’S power ai. They’re originally built for gaming, like for, you know, online gaming and speed and everything. And then they realized, you know, sometime in the late, you know, early 2000 tens, like the teens that these chips could, could power artificial intelligence and all this deep learning movement in language and vision and stuff.
[00:56:16] Paul Roetzer: Jensen, the CEO delivered the first like supercomputer to OpenAI to build. GPT, like, so they are at the core of everything. So I have been insanely bullish on Nvidia as a company because they are the infrastructure to ai, they enable all of this. And so I just think that they’re making a plane now to become more and more relevant at the enterprise level.
[00:56:41] Paul Roetzer: And this idea of custom models and these foundations, it’s going to be something you’re going to start becoming in contact with more. Again, there may be a lot of people who invest in Nvidia personally, like as a AI stock. I’m not giving stock advice, but I have been bullish on Avidia for a really long time.
[00:57:00] Paul Roetzer: But I think you’re going to start to see more real application or hearing the name Nvidia more in kind of actual business and marketing circles. Where previously it wasn’t really. A company that would be talked about, like as your tech stack. You weren’t thinking about them as a player. So I just pay attention to Nvidia.
[00:57:19] Paul Roetzer: They’re, they’re doing insane stuff. Like if you go watch his keynote from last week and look at some of the demos they’re doing on the things they’re building, I mean, it’s just one of the most advanced companies in the world, from an AI perspective. Gotcha.
[00:57:34] Mike Kaput: All right, so last but not least, we have an interesting article that
[00:57:38] Paul Roetzer: page six, this is like our page six ,
[00:57:42] Mike Kaput: right?
[00:57:43] Mike Kaput: Yeah. It’s super, super important, and yet it’s still buried on page six . So an article from S four dives in through some kind of exclusive interviews about what they’re calling this quote, secret History of Elon Musk, who we all know, and Sam Altman, who is the CEO of OpenAI. So I’m not sure everyone who follows us or follows the podcast knows that, you know, for quite a while, Elon Musk.
[00:58:09] Mike Kaput: Was an integral part of OpenAI, helping found it as a nonprofit, along with people like Reid Hoffman, kind of giving some initial funding for it and being heavily involved in its operations. So this article dives into the kind of rift that has happened between Musk and Altman Musk left OpenAI years ago, largely over butting heads with Altman about the direction of the company.
[00:58:33] Mike Kaput: And so why this matters is that we’re actually in recent weeks, increasingly seeing Elon Musk tweet quite a bit about the direction of artificial intelligence and come at OpenAI specifically. So Paul, you are intimately familiar with, you know, the work of Elon Musk and Sam Altman. What was your takeaway reading this article and on this overall conflict between the two and what that might mean for ai,
[00:59:00] Paul Roetzer: as I thought it was fascinating, kind of more from like a.
[00:59:03] Paul Roetzer: Like a curiosity perspective. And again, almost like rumor mill sort of stuff, like people just love that kind of stuff. But the, like the Cade Mets book Genius Makers, which we’ve referred to on the show before, tells the story of the founding of OpenAI and how it all came together, which is fascinating in its own right.
[00:59:19] Paul Roetzer: But when Elon Musk left and they became for-profit, and it was kind of like, it was, it wasn’t like super bitter breakup, it didn’t appear. But in recent months, there has definitely been a lot of like undertone in the tweets from everybody about, you know, Elon Musk. Like I, yeah, I don’t know how I give a hundred million to a nonprofit that all of a sudden becomes a for-profit and blah, blah, blah.
[00:59:41] Paul Roetzer: But when you go through this article, it’s like, okay, like this is starting to make a little bit more sense where the falling out occurred and why it occurred. And at the time they positioned it. Elon Musk was investing heavily in building AI at Tesla and was like starting to have some conflicts of interest as a board member in particular took Andre’s Kapai and moved him, like went from OpenAI to Tesla.
[01:00:04] Paul Roetzer: So now you’re starting to like steal talent while you’re on the board of the company. And so that was how they positioned it. But this article makes it appear like that maybe that was an issue. But no, it was, there was a very big difference in direction, strategically and it sounds like Musk wanted and control.
[01:00:20] Paul Roetzer: And so I think the what end up might end up being the most relevant part of this is there’s been some murmurs that Musk hasn’t really. But, quelled, I would say that he’s just going to build his own version of the true OpenAI. So the challenge has become open. AI has become closed. They’re not sharing anything.
[01:00:38] Paul Roetzer: And as we talked about, I think in the last episode, they’re basically saying, yeah, we made a mistake. We shouldn’t have been sharing everything we were building. We’re now going to keep everything in house. We’re not going to explain how we do this, how we train ’em, how big the models are, none of it. Anyon Musk is like, well, that’s the wrong approach.
[01:00:53] Paul Roetzer: And so either Musk’s just going to go fund the companies that are doing the mo open approach, or he’s just going to do it himself, knowing him, like find somebody else to run Twitter eventually. and then he can free up his time to have a 12th company to do this. Like, but they definitely have very diverging opinions of ai.
[01:01:12] Paul Roetzer: Musk is is sort of a, it’s going to end the world guy. Like, he, he really sees, the downside, and isn’t shy about talking about that. And he may be right, like, I don’t know. Sam Altman and Open Eye tend to prefer like, well, we gotta push forward and we’ll figure it out as we go. So I don’t know.
[01:01:33] Paul Roetzer: I think it’s a, it is just a fascinating because they’re the people that are going to influence the future of society for better or for worth. Worse. Like, we, we have no choice. These are, these are the people leading this tech. And one way or another, they’re going to have an impact not only on your jobs, but on your industries, on the business world and on society.
[01:01:49] Paul Roetzer: So it’s just, we gotta pay attention to these people, because their decisions are going to affect all of us moving forward. So, and it’s just fascinating. ,
[01:02:00] Mike Kaput: there’s no lack of fascination among all these stories. Paul. Paul, as always, thank you for staying on top of artificial intelligence so our audience can too.
[01:02:11] Mike Kaput: We so appreciate the insights and, You know, I’m sure we’ll have plenty to talk about next
[01:02:17] Paul Roetzer: week. Yeah. And like we said, I think the takeaway for me again is like, as hard as it is to keep up on this and as daunting as the task is, would be, it’s, it’s your best path forward. So like, try and focus on the positive and the opportunities ahead.
[01:02:30] Paul Roetzer: Keep learning, keep exploring, keep challenging yourself to, you know, find the threads about AI that you find fascinating. Explore those. And you’ll come out ahead. Like, It’ll, I do believe it’s all going to work out, but I think that people are proactive, have the best chance of, n near term and long term benefiting from artificial intelligence.
[01:02:53] Paul Roetzer: And we just need more smart, smart people asking the hard questions. And that’s what I love about the comments section of my LinkedIn post. It’s. Super smart people asking really good, hard questions that I don’t always have the answers to, but I love that people are asking ’em, and I know that they’re talking about ’em, you know, with their friends and in their, in their organizations.
[01:03:12] Paul Roetzer: And that’s what we need, just more conversation, more dialogue so we can all try and like move in a positive direction for this. So thank you as always. Hopefully we’ll see some of you at the AI for Writer’s Summit this week. Otherwise we’ll be back next week with another episode and I’m sure a bunch of new updates next week.
[01:03:28] Paul Roetzer: So thanks as always, Mike, talk to everybody next week. Thank you.
[01:03:31] 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.
[01:03:53] Paul Roetzer: Until next time, stay curious and explore AI.