It’s the week after HubSpot’s annual INBOUND event, and big announcements on AI were made. Paul and Mike also talk about our experiences using Google Duet AI, and they talk people: From Elon Musk’s quest to Time’s Top 100 People in AI, it’s another great episode!
Listen or watch below—and see below for show notes and the transcript.
This episode is brought to you by Piloting AI for Marketers. Complete eight hours of AI education to build a foundation and a framework for your AI journey. Individual and group rates are available. A code to save $50 is mentioned in the podcast!
00:02:48 — Elon Musk and the future of AI
00:21:04 — Our team’s experiences with Google Duet AI
00:42:59 — Microsoft’s offer for copyright infringement cases
00:46:17 — Mustafa Suleyman’s take on how to tame AI
00:51:58 — Time’s 100 Most Influential People in AI
00:54:20 — Falcon 180B
00:55:56 — Anthropic introduces Claude Pro
00:57:32 — HubSpot announces new developments at INBOUND
01:01:37 — How Elon Musk set Tesla on a new course for self-driving
Inside Elon Musk’s struggle for the future of AI
We just got a never-before-seen look at how—and why—Elon Musk decided to go all-in on artificial intelligence. This comes from an article by Walter Isaacson in Time, and is adapted from his upcoming book Elon Musk, which publishes today! Isaacson’s name may ring a bell, as he’s also the author of the Steve Jobs biography.) In the article, Isaacson gives new details on the actions Elon Musk has taken to get highly involved in the future of AI. It turns out that Musk has become increasingly worried about the development of advanced AI—and considers it probable that we develop superintelligent AI that poses an existential risk to humanity if not properly shepherded into existence.
Much of his discontent seems to have come from rocky relationships with Google over its acquisition of DeepMind and displeasure that OpenAI, which he co-founded, pivoted away from being a non-profit lab releasing AI advancements for everyone to use and build upon. As Isaacson details, Musk has spent years developing dedicated AI capabilities across his companies, including Neuralink, Tesla, and SpaceX. He’s also actively considering how to use Twitter’s data to fuel AI systems.
Now, he’s founded an overarching AI company called xAI to tie together all these AI efforts—and tapped a former AI expert at DeepMind, Igor Babuschkin, to join the company. His goal? To ensure AI develops in a way that benefits humanity and guarantees that superintelligent AI doesn’t cause existential risks to the species at large. Musk has made some bold public statements before; it will be interesting to see what develops.
We tested out Google Duet AI
Google recently released Duet AI for Google Workspace, an AI copilot across popular Google apps like Docs, Sheets, Slides, Gmail, and Meet—and Marketing AI Institute took a deep dive into its capabilities. Over the last week, we’ve spent hours kicking the tires of different Duet AI capabilities across the main apps…and we definitely have some thoughts on how marketers and business leaders can take advantage of these new AI capabilities.
Microsoft offers legal protection for AI copyright infringement challenges
Microsoft just announced Copilot Copyright Commitment, a policy that provides legal protection for customers sued for copyright infringement when using Microsoft's AI systems like GitHub Copilot and Bing Chat.
This comes as the explosion of generative AI tools has raised concerns about reproducing copyrighted material without attribution, and Microsoft aims to give customers confidence in deploying AI without worrying about copyright issues by covering any legal damages. The policy covers Microsoft AI products that use built-in guardrails, as the company faces ongoing litigation over Copilot's alleged copyright violations from scraping code.
"As customers ask whether they can use Microsoft’s Copilot services and the output they generate without worrying about copyright claims, we are providing a straightforward answer: yes, you can, and if you are challenged on copyright grounds, we will assume responsibility for the potential legal risks involved," writes Microsoft. A big statement. What does that mean for businesses?
Enjoy the episode…and stick around for the rapid-fire topics, including announcements at INBOUND, Time’s Top 100 People in AI, and much more.
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
[00:00:00] Paul Roetzer: this is very, very critical to understand why we're having this battle today between proprietary closed models and open models
[00:00:06] 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.
[00:00:27] My name is Paul Roetzer. I'm the founder of Marketing AI Institute, and I'm your host.
[00:00:36] Paul Roetzer: Welcome to episode 63 of the Marketing AI Show. I'm your host, Paul Roetzer, along with my co host, Mike Kaput. Good morning, Mike. Morning, Paul. How's it going? Good. It is. Monday. Well, we are recording Monday morning, September 11th. So this will be airing on Tuesday, September 12th, give you a little context of what's going on in the world around us.
[00:00:59] Paul Roetzer: We have some interesting stuff today. There's some new books that are coming out that are creating a whole lot of interesting fuel for content and AI. And I think the fall is going to be crazy with, some big conferences coming up where we can expect some announcements. We just had inbound, last week with HubSpot making some AI announcements.
[00:01:19] Paul Roetzer: So I, I just feel like that conversation you and I keep having about, do we need two a week? I feel like that's going to resurface us going into this fall. I think. As busy as the summer was, I think the fall is going to be crazy in terms of AI news. So here we go. Let's buckle in for another episode. This episode is brought to us by, the Piloting AI for Marketers series through Marketing Institute's Online Academy.
[00:01:44] Paul Roetzer: It's designed as a step by step learning journey for marketers to guide them through adopting AI to advance their companies and careers. Mike and I recorded these in December, 2022, is that right? Yeah. So after ChatGPT came out. The series includes 17 on demand courses. There's dozens of AI use cases and technologies featured collection of templates and frameworks, and it's about eight hours of learning.
[00:02:10] Paul Roetzer: So you can do it in a day. If you want to cram, there's quizzes, there's a final exam with. Professional certificate at the end, if you get 80 percent or above. So it is a great way to, if you're asking yourself, how do I get started with AI? How do I take the lead in my organization? That's what we built this series for is to really give you an in depth understanding of AI and give you the tools and frameworks to help you get started in your company and your career.
[00:02:35] Paul Roetzer: You can go to pilotingai. com to learn more. And use AI pod 50 for 50 off registration. All right, Mike, let's get into it. We got a lot of interesting topics today. All right, Paul.
[00:02:48] Mike Kaput: So first up, very big topic. We're going inside Elon Musk's struggle for the future of AI. So we just got a never before seen look at how and why Elon Musk has decided to go all in on artificial intelligence.
[00:03:04] Mike Kaput: This comes from an article by Walter Isaacson. You may know that name as the author of the Steve Jobs biography. And this article came out in time and it's basically adapted from a chapter from his upcoming book, Elon Musk, which actually will publish the day this podcast episode drops on Tuesday, September 12th.
[00:03:25] Mike Kaput: In the article, Isaacson gives a bunch of new details on Musk's actions and how he has
[00:03:37] Mike Kaput: It turns out that Musk has become very increasingly worried about the development of advanced AI, and he actually considers it probable that we develop some type of super intelligent AI that could pose an existential risk to humanity if it's not properly shepherded into existence. Now, According to this article, much of his discontent with the state of AI seems to have come from some rocky relationships with Google over its acquisition of DeepMind, and some displeasure he has with OpenAI, which he co founded, and he's Unhappy that they've pivoted away from being a non profit lab to start that was supposed to be releasing AI advancements for everyone to use and build upon, a future that has not come to pass.
[00:04:25] Mike Kaput: So Isaacson goes deep on how Musk has spent years now developing dedicated AI capabilities across all of his companies, and that includes Neuralink, Tesla, and SpaceX. He's also now actively considering how do you use Twitter's data to actually fuel AI systems. Very recently, he announced founding an overarching AI company.
[00:04:51] Mike Kaput: To tie together all these AI efforts. It's called X AI, and he actually tapped a former DeepMind employee, Igor Babushkin, to join the company. So his goal here is to essentially assure that AI develops in a way that benefits humanity and also guarantees that super intelligent AI doesn't cause this existential risk to our species at large, which he seems to believe is impossible.
[00:05:20] Mike Kaput: Risk or danger moving forward. So there's a lot here to unpack, Paul. I want you to maybe start out by contextualizing this for the audience. If you don't follow Elon Musk as much as I would say we do, or you're only following Elon Musk recently, he's known for making some pretty big public statements, some big PR efforts, controversial statements about big topics.
[00:05:48] Mike Kaput: How serious is he about this?
[00:05:51] Paul Roetzer: This is such a fascinating article. If you're like you're saying, like, if you have followed AI closely over the last decade, plus a lot of this isn't new information, but realistically for most of our listeners, this article would just be mind numbing. Like it is really, there's a lot to unpack and probably about a 2000 word article.
[00:06:16] Paul Roetzer: So I'm anxious for the book. I have it on pre order. I'm, I'm, I'm waiting for it to arrive, as well as coming wave from Mustafa. And I like, I have like five AI books lining up right now. Okay. So let's try to unpack this as best I can. And I, I do think that this is a really important topic that the history of AI, especially going back to 2011 is so essential to understand what is going on right now and all of the different key players.
[00:06:45] Paul Roetzer: So we're going to, mention some different names in here and mentioned some different companies. But again, I really think it helps to understand why we're in the moment we're in in business and in society. If we understand the interrelations between all these different companies and AI players. So let's start to unpack this.
[00:07:05] Paul Roetzer: So I started studying AI. Around, summer 2011, right after I finished the manuscript for my first book, which is basically around the same time that this article starts. So they start with the story of Elon Musk meeting Demis Hassabis, who had created DeepMind and at the time was trying to raise money for DeepMind.
[00:07:24] Paul Roetzer: DeepMind eventually sells to Google. We'll get to that in a minute. So they meet in 2012, hit it off. Asaba comes to the SpaceX, facility. So keep in mind again, Elon Musk started SpaceX in the early two thousands, before he got involved with Tesla. So he convinces Musk during that tour that AI is an existential threat to humanity.
[00:07:46] Paul Roetzer: And Musk starts becoming extremely concerned and decides on the spot to put $5 million into DeepMind so he can stay close to ai. And these, th this threat to humanity. Simultaneously, Musk is good friends with Larry Page, one of the co founders of Google. He used to sleep on his couch when he would stay at Silicon Valley.
[00:08:06] Paul Roetzer: So, Musk and Page, good friends. Musk brings up his concerns around AI, knowing that Google, from its founding, has been building AI. and expresses to Page his concerns around this. Page dismisses him, calls him speciest, meaning you're for humans. Like the machines should be allowed to do what they do and develop their intelligence and you should basically leave them alone.
[00:08:30] Paul Roetzer: So Musk becomes extremely concerned about Larry Page's lack of interest in the threat of AI. So, when DeepMind sells to Google then in January 2014, Musk becomes even more concerned. And in fact, he tried a last minute effort to stop Demis from selling to Google and tried to buy DeepMind himself. And Demis chose to, to sell to Google anyway.
[00:08:55] Paul Roetzer: There's an amazing story about that in Genius Makers by Cade Metz, a book we've talked about many times on the show. If, if this is fascinating to you, go read Genius Makers. It tells the story of how Demis sold DeepMind and Y and all that stuff. So then fast forward, DeepMind sells, Musk becomes more concerned.
[00:09:14] Paul Roetzer: So he turns to Sam Altman, who's running Y Combinator at the time. And they start through a series of dinners, decide let's create OpenAI, which Musk supposedly came up with the name for, and let's make the most advanced AI in the world available to everyone so that Google and Microsoft and these select few companies don't control it.
[00:09:34] Paul Roetzer: They go recruit Ilya Sutskova from Google. Ilya had in 2011, 2012, created a company with Jeff Hinton, who we've talked about before, who recently left Google. Ilya is one of the top AI researchers in the world, was involved in a major breakthrough in computer vision, which sort of set us on this deep learning path we're on today.
[00:09:55] Paul Roetzer: And then they also recruit Greg Brockman, who is the CTO at Stripe at the time. So Sam Altman, Greg Brockman, Ilya Sutskeva, Elon Musk start OpenAI. In the article, it says. As they're kind of going through this formation, one question they discussed at dinner was what would be safer, a small number of AI systems that were controlled by big corporations or a large number of independent systems.
[00:10:20] Paul Roetzer: So
[00:10:20] Paul Roetzer: this is very, very critical to understand why we're having this battle today between proprietary closed models and open models
[00:10:27] Paul Roetzer: went on to say. OpenAI, this founding team, concluded that a large number of competing systems providing checks and balances on one another was better. For Musk, this was the reason to make OpenAI truly open so that lots of people could build systems based on its source code.
[00:10:45] Paul Roetzer: So this becomes the friction point. Fast forward now a few years. So, 2018, Musk and Altman disagree on the future of OpenAI. Musk proposes folding OpenAI into Tesla. Altman disagrees. And that creates a friction point. So it says in here, Musk's determination to develop artificial intelligence capabilities of his own companies caused a break with OpenAI in 2018.
[00:11:12] Paul Roetzer: He tried to convince Altman that OpenAI should be folded into Tesla. The OpenAI team rejected that idea, and Altman stepped in as president of the lab, starting a for profit arm that was able to raise equity funding, including a major investment from Microsoft. So Musk exits OpenAI, leaves the board, Takes his, whatever he, I don't know how much he claims he put in a hundred million or whatever he's put into OpenAI.
[00:11:36] Paul Roetzer: So he leaves and Chart decides to chart his own path. So now we have OpenAI for profit arm that can now go raise money and build GPT and Musk splits off and goes his way. So it says, must decide to forge ahead with building rival AI teams to work on an array of related projects, including Neuralink.
[00:11:54] Paul Roetzer: Which implants microchips into human brains, Optimus, which is their human like robot, and Dojo, which we're going to be hearing a lot about in the fall, as a matter of fact, as of this morning, pre market, Tesla's shares were up over 6%, in large part because Morgan Stanley Sees dojo as a major driver of enterprise value moving forward.
[00:12:15] Paul Roetzer: So again, all of this is connected. All of it is driving all these different elements right now. And then he also started obviously focusing on Tesla and self driving cars. He said, it says he was resentful. Now, keep in mind he, when I'm saying this, he is the source for this article. So this is coming from.
[00:12:33] Paul Roetzer: His personal interviews with Walter Isaacson writing this biography. So when I'm giving these quotes, this is coming from Elon basically to Walter. He was resentful that he had founded and funded OpenAI, but was now left out of the fray. AI was the biggest storm brewing and there was no one more attracted to that storm than Musk.
[00:12:52] Paul Roetzer: So this kind of leads us to the conclusion here of what is his view moving forward and what are his motivations and what does that mean to you and your company and society. So we'll kind of wrap here with my thoughts, Mike. So in the article, it says what can be done to make must ask. So he called Isaacson to this like Meeting at his house.
[00:13:13] Paul Roetzer: And he's like, we have to leave our phones, in the house. So no one can like spy on us basically. So they walk out and they have this like private conversation, which Musk eventually agrees to allow Isaacson to publish. So he says to Isaacson, what can be done to make eyes safe? I keep wrestling with that.
[00:13:29] Paul Roetzer: What actions can we take to minimize AI danger? And this is the one that like, if you're not, if you're new to this is the port just starts getting sci fi, minimize AI danger and assure that human consciousness survives. This is why he wants to build a colony on Mars. That's why SpaceX exists to continue human consciousness in the event.
[00:13:50] Paul Roetzer: That it ceases to exist on earth. Okay. So then he noticed the amount of human intelligence was leveling off because people were not having enough children. Again, you hear all these stories about Elon Musk having lots of kids. It's because he believes population is collapsing that we're not having enough babies as a society.
[00:14:09] Paul Roetzer: So he is doing his part to make as many babies as possible. This is no joke. This is like. Legitimately why he's doing this. So meanwhile, the amount of computer intelligence is going up exponentially like Moore's Law and steroids. At some point, biological brain power will be dwarfed by digital brain power, meaning the AIs are just going to get way smarter than the collective human consciousness.
[00:14:32] Paul Roetzer: In addition, these systems will be able to learn the singularity, the moment where AI becomes superhuman at all cognitive tasks and Musk says this could happen sooner than expected. So in April, again, kind of driving to the end here, he assigns a team led by what was the guy's name that took over the XAI?
[00:14:55] Mike Kaput: His name is
[00:14:57] Paul Roetzer: Igor, Babushkin. Okay, that was a name, I didn't know that name. So this is a guy from DeepMind. Okay. So in April, he assigns Igor and the team three goals. The first is to make an AI bot that could write computer code, which we already have. Llama2 we know can write code. We know GPT 4 can write code.
[00:15:15] Paul Roetzer: So this isn't new. He's trying to compete here. The second is a chatbot competitor to OpenAI's GPT series that uses algorithms and trained data sets to ensure political neutrality. This is the Elon Musk version of this. This is what we're seeing with Twitter and X. And the third, and most important, Is he gave the team an even grander overriding mission, which is to ensure AI developed in a way that helped guarantee that human consciousness endures.
[00:15:45] Paul Roetzer: That would be best achieved. He thought by creating a form of artificial general intelligence or AGI. that could reason and think and pursue truth as its guiding principles. You could give it a task like build a better rocket and it could do it. So, this is, this is his play. This is everything he's doing.
[00:16:04] Paul Roetzer: So what does this mean to us? Why should all of us care about this? The reason is, Is because they're all focused on building next generation, large language models and AGI. And to do that, it's all about the data. So what does Musk have that the other people don't have? So they address this in the article.
[00:16:30] Paul Roetzer: It says the fuel for AI is data. The new chatbots being trained on massive amounts of information, such as billions of pages of the internet and other documents. Google and Microsoft have these, like that's what they have over Musk. But what can Musk do that's different? One, he now owns Twitter. So it includes more than a trillion tweets posted over the years and 500 million added each day, what he considers humanity's hive mind.
[00:16:55] Paul Roetzer: It's the world's most timely data set of real life conversations, news, interest trends, arguments, and lingo. So it becomes a massive training ground. for chatbots. The other thing he has, and this is the part that I've always said is overlooked, that people think of Tesla as a car company. That is not what Tesla is.
[00:17:14] Paul Roetzer: Tesla is a data company. They have 160 billion frames per day of video that are received and processed from the cameras on the Tesla cars. Once they start Moving like there's, there's word now that they're working on a 25, 000 version of a Tesla and robo taxis that, that would be cars you could just order and they'll come pick you up fully autonomous.
[00:17:37] Paul Roetzer: Once they have 10 million plus cars on the road, which is what they're working toward, they have probably a trillion frames of video per day. To train AI on a total worldview, everything happening around them. So to conclude here, think about this, what we're, what we're going to see in the next 12 months, we are going to see some early forms of whatever XAI is, whatever these models are that they're working on from this dojo mode.
[00:18:05] Paul Roetzer: So we're going to see that. There is increasing word that Apple is working on something called AJAX, which will include their own GPT, their own language model within Apple, but also multimodal, trained on all kinds of things, including Surrey data and everything on your iPhone. We know that Google DeepMind is working on Gemini.
[00:18:24] Paul Roetzer: It's the rumor is that will come out this fall and be more advanced than GPT 4. We know OpenAI is doing something. They're not going to sit still. GPT 4. 5, GPT 5, we don't know, but something will come. We do know they have a developer conference, November 6th. So that's a date to mark. We know inflection is working on something.
[00:18:44] Paul Roetzer: We're going to talk about Mustafa Solomon and inflection in a moment. And the latest rumors that Meta is building a GPT 4 competitor. Likely multimodal. What does Meta have that nobody else has? They have Facebook, they have Instagram, they have WhatsApp, they have Oculus. So again, when we think about the future and you're trying to figure out what does your career look like, what does your business look like, what does your industry look like?
[00:19:08] Paul Roetzer: This is why we say it's almost impossible to solve for that. Like it is. Like no one has a clue what's going to happen because just the variables of this story are so significant and the implications of what it means and what people are going to build. It's so hard to envision that. and so that to me is at the end, why this story leads off today and why it's so relevant is the history of all of this matters.
[00:19:35] Paul Roetzer: The interrelationships between all these key players in AI matter. Because they're all affecting the decisions that are going to be made that are going to drive the future of technology and business and society and education. And all we know to be true is the rate of change is accelerating. And every signal we see means it shows it's going to keep getting faster.
[00:19:57] Paul Roetzer: I need a drink
[00:19:58] Mike Kaput: of water. Wow, that's fantastic. I feel like just given your unique context into the last 10 plus years of AI plus Elon Musk, that's a really unique perspective. I think people aren't going to be able to get everywhere else, which is really, really cool. But yeah, it's We're in for a weird future, it sounds like.
[00:20:20] Mike Kaput: Maybe
[00:20:20] Paul Roetzer: sooner than we think. I really think this fall is going to be crazy. Like, if you think about last November 30th was when ChatGPT came out. Like, we're not even a year into this yet. Right. And I, I, I don't know that we're going to have a ChatGPT level event this fall, but I think collectively you're going to see a series of major milestones in AI in the next three months.
[00:20:42] Paul Roetzer: and collectively by the time the holiday season rolls around, we're going to be re imagining businesses across every industry. Like I really think we're going to see some massive innovations in the next like three to four months. Man,
[00:20:56] Mike Kaput: I can't wait.
[00:20:59] Paul Roetzer: We're going to certainly have plenty to talk about.
[00:21:02] Mike Kaput: So in our next topic today, we wanted to cover a little further Google's, Duet AI for Google Workspace. Now, in a previous podcast, we talked about how Google had released Duet AI, which is an AI copilot. That works across different Google Workspace apps. So think things like Docs, Sheets, Slides, Gmail, and Google Meet.
[00:21:26] Mike Kaput: And we actually at Marketing AI Institute took a pretty deep dive into the capabilities in Duet AI across these different Apps in our, Google business account. And over the last week, especially we've spent hours kind of kicking the tires of different duet AI capabilities across all the main apps.
[00:21:46] Mike Kaput: And, at least I definitely have some thoughts on how marketers and business leaders can take advantage of these new AI capabilities and what they should know. So we wanted to. Do something a little different here and Paul, you can maybe kind of set the stage of where we're at when it comes to us as an organization trying to experiment with this technology and then I can dive into some of the takeaways, some of the things we tried, some of the things we found as we explored these apps.
[00:22:16] Paul Roetzer: Yeah, I think it was last week's episode. We talked about one. You need a Google admin access workspace. So if you're a listener and you don't have admin access, you're going to need someone in your organization that does. So this again is assuming you're using Google workspace in your company. These same type of capabilities will be available through Microsoft copilot as well.
[00:22:36] Paul Roetzer: But right now, Google sort of first to market with wide scale availability. So you need someone in your organization to turn on admin access and then assign logging The duet AI licenses to users. So in our case, we selected a small group of people within the organization to test it. And what we did, and this is what I kind of recommend is again, have someone be the point.
[00:22:56] Paul Roetzer: So we made Mike the point person for testing throughout the week, just so we could talk about it today. We're not done with our testing, but get initial reactions. We created a single. Doc in Google Docs as a sandbox for experiences and what we'll often do is put like subheads for Paul, Mike, Kathy, Tracy, like whatever.
[00:23:14] Paul Roetzer: We'll put the people in and then you'll have those people go through and make their notes. And then the tertiary like heading would be, Gmail experience, Docs experience, Sheets experience. So it kind of created an organized way. To document what you're going through, however, your organization does it.
[00:23:31] Paul Roetzer: So for us, that's how we'll often do it when we're just throwing something together quick. And so that's what we did here was I turned on the access from admin. I assigned licenses out. We created the doc and said, okay, when you have a chance to go through and do this, organize your thoughts here, your experiments, your questions, anything about that.
[00:23:49] Paul Roetzer: And then we can go back in. So we're sort of short. Sharing on the on the fly here, like we're still in the midst of testing, but we wanted to be able to report back some tangible thoughts so far. And again, for us, we threw Mike in the lead here. And so we're going to kind of turn over and let Mike walk through some of his thoughts related to do at AI and Google workspace.
[00:24:10] Mike Kaput: So I'll try to keep this brief, but just to kind of contextualize this for people, because not only is this, I think, instructive for what can you actually find and do at AI features, but also just kind of a look at kind of maybe a V1 process of how I'm kind of thinking about these tools or any type of AI technology.
[00:24:27] Mike Kaput: So I actually started off not even by using the technology, I started off by actually Trying to dive into anything that would give me an overview of what's even possible here. And so Google actually had a really helpful handbook that we can link to in the show notes that came with, one of the blog posts on do at AI.
[00:24:46] Mike Kaput: So it was about 15 pages or so. So I kind of skimmed through that and really just tried to get a sense of, okay, where do I start? What use cases might I want to explore? And there. Is this really good section on five ways to use do at AI to work smarter and they break it down in the following way. First, it can help you write.
[00:25:03] Mike Kaput: So across different apps, you're getting some writing assistance. It can help you organize. You can organize data. It can help you visualize. There are ways to create art, and images from simple prompts. It can help you connect. It says you can use AI to have better calls with enhanced video, audio, and custom backgrounds.
[00:25:23] Mike Kaput: And last but not least, it can help you create an app. You can actually use certain features across some of Google's tools to build business apps using language prompts. Now I will say full disclosure, I did not test that feature. Maybe we'll do that moving forward, but I wanted to really get started with tools that we were super familiar with.
[00:25:41] Mike Kaput: And that we used every day. So I'll jump into each tool and kind of what I found with each one, and then maybe zoom out, overall kind of how I'm thinking about this. So first up with Google slides. So really the bulk of the features as of today, and they're releasing new features all the time is really a help me visualize.
[00:26:01] Mike Kaput: Feature which is kind of an image generator. So basically it will provide you with some inspiration We'll say hey, here's a prompt and here's a photo that goes with it. They could just drop into a slide They seem to be just random. They're pretty cool. But like have no real rhyme or reason And then you can actually just prompt Google Slides to create images right in the slide.
[00:26:23] Mike Kaput: So for instance, I typed in a cute robot wearing a backpack in a cartoon like style. Then, outside of the prompt, you can actually select from a drop down several other types of style prompts. Photography, vector art, sketches, watercolors, and a couple of others. So once you generate this image, you get eight results, right in Google slides.
[00:26:46] Mike Kaput: I'll be honest. I found like the images pretty appealing, even from pretty simple prompts for kind of cartoonish stuff. I tried some more photo realistic ones. It was not nearly the level of quality I've seen in a tool like a mid journey. But I found it really, really helpful actually to just have it right in the slide.
[00:27:04] Mike Kaput: I often struggle to get fresh art, use stock images. I'm not often always just jumping into mid journey and spending hours trying to prompt a really cool image. So for a really quick and easy image, This was a quite a useful feature. Now it actually, as of today, cannot actually read data or information on the actual slide.
[00:27:26] Mike Kaput: So I actually was trying to test out if it could do that, try to break it, try to see what put it through its paces. I think that functionality is coming. It seems to allude to the fact that is, but as of right now, you can't be like, Hey, turn these three bullets into. A fancy visual. So that would be really cool.
[00:27:44] Mike Kaput: I think to have moving forward
[00:27:45] or
[00:27:45] Paul Roetzer: it's like recommending visuals as you're putting content in, it would be really cool. Yeah. Cause right
[00:27:50] Mike Kaput: now those kinds of recommendations are just random images where they're like, Hey, that might look cool. And you're like, well, that has nothing to do with my presentation.
[00:27:58] Mike Kaput: Now, when I jumped into Gmail, this basically consists of a help me write. button so I actually started off trying to see if it understood well the context of previous emails so I asked it to summarize questions that were in a previous email from someone I was doing a podcast interview they said hey here's the questions in advance we're going to talk about and it seemed like it did a pretty okay job of saying okay I can go read these past emails and Write to you what's in them verbatim.
[00:28:28] Mike Kaput: This isn't really like the intended use of this, but I did want to see like, can I just jump in? And instead of having it write an email for me, can it just summarize sometimes? Like if there's a thread, someone forwards me, I tested that. It actually worked pretty well. It's really, really light though. When you ask it to summarize something, I mean, It did a great job of summarizing all the details from an event email from an event I'm speaking at in a couple weeks, but it was not at all remotely like in depth details.
[00:28:57] Mike Kaput: Again, not really what it's meant to be used for, but I found that was interesting. I started then trying to generate emails. Now, this might just be my personal preferences. I tend to honestly get a lot of value out of when I send an email, putting a lot of thought into it in order to like, if I can send one email that solves a lot of issues down the line, it's well worth spending time on versus an ill thought out response.
[00:29:24] Mike Kaput: So I don't know personally. How much I would be using this to generate emails. I gave it a prompt of, Hey, I've been asked to speak at an event, but I can't attend. Someone asked me if I could do a virtual event instead, I still can't make it, but I'd like to really write an email politely declining. So not something super, super important, but just seeing what it sounded like.
[00:29:44] Mike Kaput: And it wrote a great response. However, it doesn't sound anything like me. So like, that's also an issue. I mean, I would know in a heartbeat, probably at least with colleagues. If somebody used this because it wouldn't sound a single thing like them at all. So I found that to be relatively limited use.
[00:30:03] Mike Kaput: And again, I don't know if I'm the best person for it just because I'm today not super comfortable with outsourcing really core communications. And for something that's not a core communication, it doesn't take long enough for me to be automating this, but it could be helpful for other people moving forward.
[00:30:19] Mike Kaput: Especially if you're someone that has a hard time shortening your emails, there's features where you can. Highlight the text and have it rewrite it for you. That was quite helpful. I could see that if you are long winded in email, that would be a useful
[00:30:32] Paul Roetzer: feature. I could, that one's, that's gotta be something they're working on where it just automatically pulls your last 20 emails as context for the prompt in essence, and like automatically writes in your tone and style.
[00:30:46] Paul Roetzer: Like that just seems inevitable, right? I know other
[00:30:49] Mike Kaput: tools that. Through plugins do that. Yeah, learn the
[00:30:52] Paul Roetzer: writing style. No one's going to use this otherwise, right? Yeah,
[00:30:56] Mike Kaput: yeah because you would know in a heartbeat that you know given how long we've worked together if I use this to write an email and Didn't edit it.
[00:31:02] Mike Kaput: You'd be like this
[00:31:02] Paul Roetzer: sounds really it's kind of like when you see LinkedIn messages that have 40 emojis in them And it's like what you obviously use GPT-4 that loves emojis, right? Like right, which tip to people. If you're putting LinkedIn posts up and it's like, it has a slew of emojis, please take them out.
[00:31:16] Paul Roetzer: Like it just, it is so obvious. Either you love emojis more than GPT-4 does, or you're just using ChatGPT to write your LinkedIn posts, but emojis like, please stop. Like it's just too much and I love emoji as much as anybody, but get them out of your LinkedIn and Twitter feeds please.
[00:31:33] Mike Kaput: So, then I tried out Google Sheets, and I'll say there wasn't a ton here to do yet.
[00:31:37] Mike Kaput: Now, I'm super excited for what's possible here, but right now, all it will do is generate sample templates for different types of tables or things you want to do in a spreadsheet. Now, this wasn't bad. I actually said, I'm promoting an ebook on my website across all common marketing channels. Can you create for me a project plan?
[00:31:56] Mike Kaput: with space for like activity owners and all the tasks we should do. It did a great job of actually producing that, with decent suggestions. So there's a little bit of an ideation thing here that might be helpful. If you need a little help saying like, I'm trying to track certain metrics, how should I structure a table?
[00:32:13] Mike Kaput: It can do pretty decent templates. Once you get more complex though, it's not very good at following prompts as of yet. It doesn't have formulas or anything in those templates, unfortunately, which would be really helpful in my opinion. I was trying to say like, hey, I'm trying to build a spreadsheet to track against a certain goal.
[00:32:32] Mike Kaput: Here's what I need to hit. Can you build me a week by week tracker with like formulas and stuff? And it just can't do it super well. The issue today is that it does not have the ability to reference any data that's actually in the table. Now, Google expressly says this is coming. So it has a message in the help me organize function that you're using here.
[00:32:55] Mike Kaput: And it says that you're going to be able to eventually Literally just ask questions of and organize your data via prompt. Now I find that to be extremely exciting because we have so much data that we want to learn more about. We use Google sheets already, and if it can add a high level of competence, help us organize and interpret data.
[00:33:14] Mike Kaput: That starts to get really fascinating from a use case perspective.
[00:33:17] Paul Roetzer: I feel like the more valuable assist here would be help me analyze. Like help me organize is an interesting choice. I just feel like based on code interpreter with chat GPT, which advanced data analysis is what it's called now, but I feel like the analysis of data is what we all.
[00:33:32] Paul Roetzer: Want out of sheets. So I wonder if there'll be that option too. So actually
[00:33:37] Mike Kaput: the exact text of their message where it says they will have this it says hey we can't really do this It says we're still learning, can't help with that, try another request. Soon, HelpMeOrganize will be able to accurately edit and analyze content in your spreadsheets.
[00:33:55] Mike Kaput: So that would be really cool if that ends up being as robust as something like advanced data
[00:33:59] Paul Roetzer: analysis. And that's something, like again, given how well received Code Interpreter has been and how valuable. I gotta guess they're racing to get that feature built into this because they're, they're going to have to have it, by the fall, I would think like, I don't, I don't see this as a year from now.
[00:34:16] Paul Roetzer: Like that's if they're already touting it's coming, they probably have it in full blown testing right now. And they're just working out the bugs.
[00:34:24] Mike Kaput: So. Quickly, in Google Meet, there are a few interesting features. I'll be honest, we don't use Google Meet that much. I use it when someone else is sending me a Google Meet link, but we use Zoom, for a lot of our purposes.
[00:34:36] Mike Kaput: So I'm not the world's most familiar person with the tool, but there were some cool features. I mean, they had fun, generative AI backgrounds. You can literally... use some really cool premade ones they have, or you can use a text prompt to actually just generate a custom background. That's kind of fun. They claim they've got AI powered features to adjust the video lighting, in the actual video while you're recording.
[00:34:59] Mike Kaput: I did a test and I, I noticed a really noticeable difference, but I didn't. I didn't see any huge difference over what we do in Zoom anyway. I actually did a side by side comparison where I recorded the same video twice, one in each tool. And I'll be honest, I, I, maybe it's just me, I did not see a ton of difference that would encourage me to jump in and start using Google Meet.
[00:35:25] Mike Kaput: Tried out the transcripts and things, I wasn't seeing a ton there that today was super advanced. So I don't, especially too, just with how often we're using things like zoom, what, what a lift it would be to switch over that tool. I think you'd need a really compelling use case to be switching here when it comes to video.
[00:35:45] Paul Roetzer: And we've talked about this before, like zoom announced some recent recently, I think last week, even that they're making some major updates, like so many of these features are going to be easily reproduced by any of the major players. Teams is going to have it in Microsoft zoom. So it's going to be. Like really cool stuff.
[00:35:58] Paul Roetzer: If you think about the advancements and deep fakes and the advancements, the ability to edit existing video, like, look at my, if you're watching this on YouTube, the lights I have behind me, like I'm sure that we'll be able to say, Hey, make like this light brighter. And like in production, like we're going to be able to go in and do things like that or fix this or remove the headphones from my head so I can record with the headphones.
[00:36:20] Paul Roetzer: But like now deep fake me and get rid of that. All of this is going to be do like anything you can imagine doing to video. Yeah. You're going to be able to do on the fly, like runway today, this isn't an art rapid fire. They just announced like a, I forget what they called it, like a producer mode or something where you're going to be able to tell the video to like, look up and down this like text to video generation.
[00:36:40] Paul Roetzer: You're going to be able to edit it as like a producer and just tell it what to do. And I just feel like all of those features are going to be so amazing. Just. Available anywhere. Yeah. So yeah, I agree. It's like you're, we're not switching to Google meet because of these. Some are just nice to have features, but zoom we'll have them in three months.
[00:36:59] Paul Roetzer: For sure.
[00:37:00] Mike Kaput: And so I'll quickly wrap up here and just talk quickly about Google Docs because this is the one I was most interested in testing. Given the amount of writing I do, we do in Google Docs, given how we're using other AI writing and content tools for our podcast process, for our blog posts, things like that.
[00:37:16] Mike Kaput: So I basically tested out on some existing workflows of trying to summarize content, reword content, write actual content for us. It's not that we would publish it, but as a first draft, help me outline stuff. Now I have to say here, to quickly sum up, and I only ran select tests, I was pretty disappointed.
[00:37:35] Mike Kaput: I didn't find this tool to be Anywhere near the Claude twos or the GPT fours of the world or writer or Jasper even it's just it Google says up front. It's a creative writing aid and that's about it So I get that but I was like you guys are building core language models It doesn't like get language wrong, but it's not great at summarization.
[00:37:57] Mike Kaput: That was probably the best thing it did, but it still took multiple prompts to try to get it to actually summarize things. Hopefully, even when prompted to write something based on a podcast transcript, it wrote a full post that was totally coherent. But it, even when instructed to, it didn't use a lot of the content of the transcript.
[00:38:17] Mike Kaput: It took the topic of the transcript, say we had a transcript on, the release of ChatGPT Enterprise. It got the topic right, but then it went ahead and just wrote a blog post as if you said, write me a blog post about ChatGPT Enterprise. It didn't do anything like Claude 2 will extract quotes, will actually just structure what you've already said in the podcast into a post, which is extremely valuable.
[00:38:40] Mike Kaput: For us, I didn't see any of that here. So I was, this one fell quite flat to me and there's, more to say about all these tools and we'll publish some more content, have some more conversations around them since this is just kind of stage one of testing, but overall there's definitely features in this suite I would use, but I was pretty disappointed, especially by Google docs, which I found to be.
[00:39:03] Mike Kaput: That this was going to be the main use case for us moving forward. It can't do anything like create social shares very well from a post. I dropped a post into Google docs and said, create me some social shares and tweets based on this. It literally just split up sentences. Didn't even do anything with like any amount of reason or creativity.
[00:39:22] Mike Kaput: The other tools today can do that at a very high level, I would argue. So I didn't find this to be very helpful at all. So that's kind of where we ended. Our existing experiments and we're going to kind of be doing more of that
[00:39:35] Paul Roetzer: moving forward. I got a couple of quick questions. So one, given our experience with Google Bart, I'm not surprised by your experience with Google Docs.
[00:39:44] Paul Roetzer: It's unfortunate, but that's kind of what I would have expected. Okay. So based on this one week of trial, does it replace any existing tech third party AI tools that we're using for you?
[00:39:55] Mike Kaput: Third party AI tools? I would say right out of the gate. No.
[00:40:00] Paul Roetzer: Okay. One of the big questions we had is, does it replace like an AI writing tool?
[00:40:04] Paul Roetzer: Like we use Jasper, we use Writer, we use ChatGPT, we use Clog. We like, is it complimentary to those or, or is it just not, you're just not
[00:40:14] Mike Kaput: even going to use it? Honestly, I'm, I'm struggling to even see how it's complimentary as of today, just because of the ways in which we use it. I think that could change pretty quickly.
[00:40:24] Mike Kaput: Like you were saying, if they roll out some more of these features and it gets a little more robust as an actual writing assistant and gets to a baseline, it would be simpler to use this tool. Natively in Google Docs than in third party
[00:40:36] Paul Roetzer: tools. Okay. Any clear use cases that would change your workflows as of right now?
[00:40:41] Paul Roetzer: Now I'm asking as the CEO of the company trying to pay 30 bucks a month starting next week for this
[00:40:46] Mike Kaput: technology. Yeah, as of today, I mean, honestly, Google Slides was the biggest one since we don't really use Google Slides, but there's nothing stopping us from doing it. Having image generation would be pretty helpful with all the decks we have to build for public speaking.
[00:40:59] Mike Kaput: I think if the Google Sheets functionality they are teasing ends up being legit, I think that's worth the price of admission alone. However, as of today, like, should I say tomorrow we should be paying 30 bucks a user for this? I'm not sure, just because of our particular workflows, but also if the writing was just much more robust, I think you have a clear, that's a clear win, even if the other stuff isn't up to snuff.
[00:41:24] Paul Roetzer: So to, to wrap this, like the way I think about this stuff as a CEO, the way I would advise companies to think about this is I would probably pay 30 bucks a month for one license and I would put Mike in charge of ongoing testing and then as new updates are made to it. We would have Mike jump in and run the same analysis, take the same use case as you ran, run them again.
[00:41:45] Paul Roetzer: Did it make improvements? How's it going? And so I wouldn't turn away. We've talked too many times about, you cannot count Google out. Like this is an early version. We know they have more advanced capabilities than they're showing. So you, you can't just give up on Google, but it sounds like our assessment at the moment is we are not going to scale up a full blown pilot project with this.
[00:42:03] Paul Roetzer: We're not going to turn license on for everybody in the company. We'll, maybe do a 30 buck a month use for one person and then see how it goes from there. So yeah, I, hopefully that's really helpful to people. Cause again, this is one of the ones we've been waiting for. It's like, if it is what they say, it's going to be, it changes everything.
[00:42:19] Paul Roetzer: Well, in this case, it's not yet. So now you wait for co pilot and say, okay, well, if, if that's everything that demo video shows that we love so much of co pilot, then it changes everything. And hopefully they're a little further along than Google appears to be. For sure. And
[00:42:34] Mike Kaput: yeah, that timeline note is important.
[00:42:36] Mike Kaput: Like this is a matter of weeks and months before the analysis here could totally change, not years. Like you can't look back. You can't say, I'm giving up on Google because I heard them say on the podcast, Duet AI isn't good. I'll revisit it a year from now. It's like, you need to be evaluating it on a much more, rapid timeframe.
[00:42:56] Mike Kaput: All right, so, for our third big topic today, Microsoft just announced a new policy called the Co Pilot Copyright Commitment, and this actually provides legal protection for customers who are sued for copyright infringement when using Microsoft's AI systems like its GitHub Co Pilot and Bing Chat. This kind of comes as we have these broader generative AI tool concerns about things like copyrighted material or reproducing it without attribution.
[00:43:24] Mike Kaput: So Microsoft looks to be giving customers kind of confidence here by rolling out this policy that covers their AI products and They actually said, quote, as customers ask whether they can use Microsoft's co pilot services and the output they generate without worrying about copyright claims, we are providing a straightforward answer.
[00:43:43] Mike Kaput: Yes, you can. And if you are challenged on copyright grounds, we will assume responsibility for the potential legal risks involved. Now, Paul, that seems like a pretty bold strategy. Like what's going on here? Why is Microsoft doing this? Like, does this have teeth?
[00:43:59] Paul Roetzer: I'm going to kind of treat this one like a rapid fire topic because this is so above my pay grade to talk about this.
[00:44:06] Paul Roetzer: So my take is knowing sales, they're probably hearing this objection a lot from big enterprises. So if they're having to come out and do something like this, which obviously had to get through their legal internally to even Like think about doing this and then to actually announce it, they must be getting massive pushback on this issue is my general take on the go to market around these technologies and what they're hearing from their early users.
[00:44:35] Paul Roetzer: I, as a non lawyer, I could come up with 10 immediate questions and objections to just being like, Oh, okay, Microsoft said it's cool. Like, let's go. And it might be that I've been spending the last week talking with my IP attorneys about, like, trademark and copyright related things, so I'm, like, fresh minded, like, in the legal world, thinking about lots of challenges to this.
[00:44:59] Paul Roetzer: All I'm going to say, and I'm going to end with it, is, great, like, take this to your legal team and let them worry about this. Like, I can't even fathom... What the contract would look like for this agreement, like to make sure that this is true, like, okay, so we're going to use copilot to write code. That's going to create a product that we're going to raise 20 million on.
[00:45:22] Paul Roetzer: And then someone's going to sue us and we're not going to be able to use that code, but Microsoft said, it's okay. So we're like, let's not worry about it. I did that. It just. It seems like a really nice idea. I don't know how this is executed, but I am not an attorney. So if this, if you're a big enterprise and you're thinking this is relevant to you, take it to your legal team.
[00:45:42] Paul Roetzer: And, that's all I got.
[00:45:45] Mike Kaput: Yeah. And with the same caveats that I'm not a lawyer, it's interesting to me that. They're like, Oh, it's okay if you get sued, like nobody wants that outcome, like, even if we get to that outcome, we've already lost. Right.
[00:45:59] Paul Roetzer: Yeah. I mean, if that's the best answer we have for people right now is like, we have a few billion set aside to deal with these lawsuits.
[00:46:07] Paul Roetzer: Yeah. Oh, my God. It's going to be messier than I thought.
[00:46:11] Mike Kaput: All right. Let's jump quickly into some rapid fire as we We get into the end of the segment of the podcast here We have mustafa suleiman who's the co founder and ceo of inflection? Which you mentioned briefly before because he's also the author of a new book coming out this week titled the coming wave He actually just did an interview with rob wiblin for the 80 000 hours podcast and he gave some insights into some big key topics and given like how big a player he is in AI.
[00:46:40] Mike Kaput: I think his perspectives are really worth listening to. Paul, you had mentioned his openness about what they're building at Inflection and how he sees AI research and labs advancing moving forward in the next one to three years really stood out to you. Do you want to maybe walk us through the points that caught your attention in this interview?
[00:46:59] Paul Roetzer: Again, just to, to recap or the importance of these topics. So Mustafa Solomon founds inflection with Reid Hoffman, who started LinkedIn raises 1. 3 billion, has a new book coming out called the coming wave. Mustafa's co founder of DeepMind, which means he was there in the early days with Demis when Demis is raising money for Elon Musk.
[00:47:22] Paul Roetzer: And like all of these people are connected, they all come from the same tree. There's like three or four major players going back to 2011. All of these major AI researchers and founders today spun out of those people. Jeff Hinton, Andrew Ng, Demis Hassabis, like they're all working together at different points and now in some ways competing.
[00:47:43] Paul Roetzer: So it's just, again, I, I'm fascinated by this topic of how this is all happening. But I think it's really important for everyone to understand. So. The key things that jumped out to me here is, first of all, we've said this before, but the generative AI tools you're experiencing today, whether it's, duet AI or chat GPT or inflection pie, whatever is the least capable AI you're ever going to use.
[00:48:08] Paul Roetzer: It's only going to get smarter, faster, more powerful, more human, like more intelligent. So it's going to have these massive implications on business, on the educational systems, on society, on elections, on politics, like all this stuff. It's going to affect everything. So the challenge becomes, but how much better are they going to get and how fast is that going to happen?
[00:48:28] Paul Roetzer: So interviews like this... Give you a window into trying to get some semblance of what that is going to look like and how quickly it's going to happen. So the few quotes I pulled out of this, that, that really caught me, Mustafa said, we're going to be training models that are 1000 times larger than they currently are in the next three years, even at inflection, his company.
[00:48:51] Paul Roetzer: With the compute they have, they will be 100 times larger than the current frontier models, the most powerful models we have in 18 months. That means like if they don't buy any more NVIDIA chips, which they're buying by the tens of thousands, if they bought none, they could build something 100 times more powerful than what we have.
[00:49:10] Paul Roetzer: The other thing he said is, people think that ChadCPT going from 3. 5 to 4 was this like small event, no big deal, just 3. 5 to 4. What he's saying is no, it was 5x more powerful to go from 3. 5 to 4. So we assume these tiny increments, but these are actually massive increases in what these things are capable of doing.
[00:49:33] Paul Roetzer: And then his big call is that the industry needs to be more transparent about what they're building, how they're building it and what they're using to do it. So he said, for example, it's much better to be, much better that we're just transparent about it. We're training about, Oh, cause he was basically taking a hit at Sam Altman.
[00:49:49] Paul Roetzer: He's like. He, Sam Alvin claims they're not training GPT 5. He's like, give me a break. Like they're obviously training bigger things there. So he said, we're going to training models that are bigger than GPT 4. We have 6, 000 H100s, those are Nvidia chips, in operation today that are training models right now by December, 2023, a few months from now.
[00:50:10] Paul Roetzer: They will have 22, 000 H100s fully operational. And every month between now and then, we're adding 1 to 2, 000 NVIDIA chips, these H100s. So people can work out what that means in terms of our training runs by spring, by summer. So once you know what they're capable of training on, the AI researchers can project out the size of these models and how powerful they're going to be.
[00:50:34] Paul Roetzer: So, to me, this kind of data... Makes it extremely clear that the rate of change is accelerating, and it quantifies that rate of change in some ways. and everybody is doing the same stuff. Amazon, Anthropic, Cohere, Inflection, Google, Meta, Microsoft, OpenAI, Stability, they're all Buying tens of thousands of NVIDIA chips, which is why their stock price is soaring.
[00:50:57] Paul Roetzer: And they're all trying to build way more advanced AI than what we have. So what I will say here, and I'll end with this is if you are a CEO, or if you're a leader of a company, you have a moral responsibility to solve for this. It is going to change everything about your business, the way you do your tech stacks, the way you build your strategies, the way you hire and develop talent.
[00:51:19] Paul Roetzer: All of it, your products, your culture, everything is going to be redefined and re imagined in the next two to three years by this change. And it's going to continue on. So you have to figure this out. I understand as a CEO, it is daunting to try and solve for this stuff. Or as a leader, director level and above, like, it's really, really hard.
[00:51:41] Paul Roetzer: You have to do it like, and that's a, it's a really hard thing to think about, but you have to go into 2024, find ways to shift your time and responsibilities. So you can be intimately involved in solving AI in your company.
[00:51:58] Paul Roetzer: So we
[00:51:59] Mike Kaput: also saw a recent announcement from time magazine where they released their 100 most influential people in AI list. Now this is a list that they curated using their editors and reporters perspectives. They whittled it down from hundreds of nominations, and it includes people you might expect like Sam Altman, Demis Hassabis, Elon Musk, and other researchers and minds in AI like Fei Fei Li, Jeff Hinton, and ethicists Timnit Gebru and Margaret Mitchell, who we've talked about several other times on the podcast.
[00:52:30] Mike Kaput: So, Paul, as you were reviewing this list, Anything jump out to you in terms of people included or not included?
[00:52:38] Paul Roetzer: One, I wish I had the time. I would go build the tree for everyone. So you go back to the 1980s and you see Yasha Bengio, Jeff Hinton, Jan LeCun, Andrew Ng, like these kind of like the forefathers of modern AI, all starting in like the 80s and into the 90s.
[00:52:53] Paul Roetzer: And then you could just take this list of hundred and kind of branch them off of how they're all connected to the same, like four or five people. That would be fascinating. If someone has time to do that, I would love to see that visualization. The three people that immediately jumped out to me as missing, and I keyword searched, so I'm pretty sure they're not on this list.
[00:53:09] Paul Roetzer: Lex Friedman. So Lex's podcast, if you don't listen to it, some of the most, like, Important and influential interviews I've listened to about AI and my understanding of AI have come from Lex's interviews with many of the people on this list over the last two to three years. I just realized I don't think Zuckerberg was on there.
[00:53:27] Paul Roetzer: Zuckerberg on there? I don't think he's
[00:53:29] Mike Kaput: on there. I don't think
[00:53:30] Paul Roetzer: he is. Yeah, so that would be, I mean, Jan LeCoultre's on there, who runs AI there, but I, it's kind of interesting you leave Zuckerberg off the list. Ethan Malik, who we've talked about many times, Wharton Business School professor. Speaker at Macon, probably the most practical AI knowledge on the market today.
[00:53:46] Paul Roetzer: So I just feel like Ethan has to be on that list. And then Allie K. Miller was the other one that jumped out to me. I mean, Allie has a massive following, started AI practice for startups at Amazon. She was at IBM previously. Now she's doing her own thing, speaking, teaching classes like Allie's just. Brilliant at this stuff.
[00:54:02] Paul Roetzer: So Lex, Ethan, Allie, and then the bonus was Zuckerberg. If he's not on there, as the people that jumped out to me, but lists like this are great. I think it helps build awareness about the people you should be following. Like you follow these a hundred people or a curated list of these people. You're going to stay up to date on what's going on in the eye world for sure.
[00:54:20] Paul Roetzer: So we just
[00:54:20] Mike Kaput: got another release of a important foundational AI model, Falcon 180 B, and this is an open source AI model that is now the largest openly available model with 180 billion parameters and according to hugging face, which is a repository of a lot of these models. They say that in terms of capabilities, Falcon 180B achieves state of the art results across natural language tasks.
[00:54:46] Mike Kaput: It actually, is beating or topping the leaderboard that Hugging Face has for against models like Google's Palm 2, and they are considering it on par with Palm 2, that makes it in their words, one of the most. Capable LLMs out there. This model, like its predecessor, Falcon 50 B was created by the technology innovation Institute, which is a research lab funded by the government of Abu Dhabi.
[00:55:14] Mike Kaput: So Paul, how significant is Falcon one ADB?
[00:55:17] Paul Roetzer: Yeah. I mean, this goes back to the whole challenge of open versus closed frontier models, they're building. This and Lama two, I guess are probably considered largely the most advanced open models right now. It can potentially be very significant. I think there's massive resources, billions of dollars, from a government entity being poured into building an open model.
[00:55:39] Paul Roetzer: And, I think it's just going to be the story of the future. We're going to have these competing choices from as businesses of building on open models or building on closed models. But there's no apparent, slowdown in the rate of innovation when it comes to the open models. And I think we're going to keep seeing that.
[00:55:56] Paul Roetzer: So
[00:55:56] Mike Kaput: another big announcement, Anthropic, who we talk about quite often on the podcast, just introduced Clawd Pro, which is a paid plan for its Clawd. ai chat system. And with Clawd Pro, you now get five times the amount of usage of Clawd 2, their most advanced model we talk about. Quite often, and this is five X more than the Claude free tier.
[00:56:17] Mike Kaput: You also get priority access to claw. ai during high traffic periods and early access to new features. Now this plan right now is exactly the same as chat GPT plus in terms of price in the U S it's 20 bucks a month and it's 18 pounds a month in the UK. These are the two countries where it's currently available.
[00:56:36] Mike Kaput: Should business leaders be taking a look at this?
[00:56:40] Paul Roetzer: Yeah, I would do the same thing we were talking about before. Claude's obviously very powerful. We've had good experiences with it ourselves, especially for summarization of long documents like research papers. I think it can do up to 75, 000 words of context in a prompt, so you can give it tons of context up front.
[00:56:56] Paul Roetzer: So again, I, what I would be thinking about as a company is I would have someone or a team of people who are constantly testing the most advanced models available to you. You can't just make a bet on one model or one app company, one software as a service company. I would be having someone who, in a regular interval, whether it's monthly, quarterly, is running the same use cases through these different models and trying to see.
[00:57:19] Paul Roetzer: So pay your 20 bucks a month for Claude, pay your 30 bucks a month for Duet, pay your 20 bucks a month for ChatGPT Like, just absorb that as research and development costs, and you have to continually test this stuff. So
[00:57:32] Mike Kaput: in terms of other AI announcements, we have a big one from popular marketing automation and CRM platform HubSpot.
[00:57:39] Mike Kaput: They actually just unveiled their AI roadmap as part of their inbound conference. So this roadmap details the AI capabilities that they're launching throughout 2023. That's across their marketing sales service and CRM hubs. And these launches include things like. The existing capabilities they released earlier this year.
[00:57:58] Mike Kaput: So like their generative AI content assistant, but it also includes new features coming like AI sales forecasts, advanced AI chat bots, and what they're teasing as an AI powered website builder. So Paul, I'll let you get into the context around you and HubSpot. Cause there's a lot of it there. You have a deep familiarity with the platform and the company's evolution.
[00:58:19] Mike Kaput: What did you make of the roadmap?
[00:58:22] Paul Roetzer: So my past life, I think most people are aware of this, but maybe not. And we have a lot of new listeners. So I started a marketing agency in 2005. We became HubSpot's first partner in 2007. And I built and scaled my agency on the back of HubSpot largely. Most of what we did was advising, consulting, and integration work for HubSpot clients to help them grow their businesses.
[00:58:42] Paul Roetzer: And then I sold that agency in 2021. But our institute is still powered by HubSpot. So we still use the platform today. So I've been a HubSpot. Advocate and user since 2007, basically, pretty intimate knowledge through the years of what they were, weren't doing with AI. I would say that they, through the years were maybe a little slower on the uptake in terms of the impact that I was going to have, so while Salesforce and Microsoft and Google and others were building massive research labs and buying up AI companies.
[00:59:13] Paul Roetzer: HubSpot wasn't necessarily playing in that game. They were doing some interesting like machine learning driven capabilities like lead, modeling and, things like that. They were doing some with language generation, but generally speaking, it wasn't an advanced AI platform, I would say. And I think for, for HubSpot, like many, software companies, chat GPT was a wake up call and generative AI certainly ushered in a new way of thinking about the future.
[00:59:41] Paul Roetzer: That being said, Dharmesh Shah, the co founder and CTO and a friend of mine has, I think, seen this vision for a long time. He's given talks at HubSpot's annual conference inbound about AI, going back to like 2018. So I think that they're all in on AI now. I think they're playing catch up in some ways in terms of their generative AI capabilities.
[01:00:02] Paul Roetzer: But it's so early right now that I don't see that as necessarily a disadvantage. And what HubSpot has that we talked about earlier is data. They have a lot of proprietary data based on millions of users of their platform. CRM data that can be very, very interesting for emails and webpages and marketing campaigns and sales efforts.
[01:00:21] Paul Roetzer: So, I, I would, I would pay attention closely, especially if you're a HubSpot customer. We talk about a lot of the fastest ways to get value from AI is to use the existing platforms you already have better, use smarter features. You don't have to go through procurement to get approval for these third party tools.
[01:00:39] Paul Roetzer: It's just like, if HubSpot can build these capabilities in and they're actually valuable. And very user friendly and take advantage of all the data HubSpot has, then it could be a game changer for HubSpot customers. I don't think it's there yet. My experience with these tools has been they're very raw and early.
[01:00:57] Paul Roetzer: But I also know based on inbound and conversations with HubSpot, what they've said publicly. I don't think there's any doubt that HubSpot, is all in. The other thing I will say is HubSpot's stock has crushed it historically. Like they have continually, exceeded expectations and none of that is because of AI.
[01:01:15] Paul Roetzer: So I think when I look out to the future of HubSpot, if they get AI right, this is a company that has been extremely strong and grown continually through the years. And if they can now mix in AI knowledge and capabilities, then it can be a really impactful, play to their enterprise value moving forward and the value they create to customers.
[01:01:37] Mike Kaput: All right, we're going to end this episode just like we began it. Another part of Walter Isaacson's Elon Musk biography has been released by CNBC, and it details Tesla's new FSD 12 self driving system for its vehicles. And this is actually powered by... a really innovative new AI approach in self driving.
[01:01:57] Mike Kaput: So according to CNBC, instead of being based on hundreds of thousands of lines of code like all previous versions of self driving software, this new system has taught itself how to drive by processing billions of frames of video on how humans do it. Just like the new large language model chatbots train themselves to generate answers by processing billions of words of human text.
[01:02:20] Mike Kaput: So Paul, you alluded to this. It sounds like it's a big deal. Can you unpack for us what's going on here?
[01:02:25] Paul Roetzer: Yeah, so we did talk about this on a previous episode briefly, when we saw, like, the Friday night video from Musk where he was just, like, randomly recording this driving experience. Here's the implications.
[01:02:37] Paul Roetzer: So, I honestly thought they were doing this already. Like, I was kind of surprised. We wrote about it in the book. They were doing parts of this, but this is a full change in the development of autonomy in cars. Previously, there was thousands of lines of code. There is thousands of lines of code. So, today, if I get my Tesla and tell it to drive me to the office...
[01:02:55] Paul Roetzer: There's all this code that tells it, these are stop signs, these are stop lights, these are right turns, left turns, and it's all coded of how to handle that situation. What they're saying now is they're going to take these 160 billion frames of video per day and they're just going to learn what a good human driver does.
[01:03:13] Paul Roetzer: And that's it. Like, there will be very minimal code that builds the software that drives the cars. The cars will learn to drive By human drivers who don't get in accidents, in essence, what he told his team was, what's a five star uber driver drive like that's what the full self driving should do. And so the metric they created it's in the article we're looking at some machine learning systems generally need a metric that guides them as they train like a goal.
[01:03:40] Paul Roetzer: So musk who like to manage by decreeing what metrics should be paramount gave them their load star. Um. The number of miles that cars with full self driving were able to travel without a human intervening. So intervention that you have to stop the car, you have to take the wheel back. That is now the driving metric for full self driving and Tesla's is how often do you have to take the wheel back?
[01:04:03] Paul Roetzer: I will tell you in an average trip from here to the office, from my house to the office, which is like eight miles. I will intervene five times. It is terrible at changing lanes. It like makes all, it just, there's all kinds of reasons I'll do it. What they're basically saying is I shouldn't have to take the wheel once.
[01:04:18] Paul Roetzer: And they want to get that number down. The reason this matters is because if this works, they're basically modeling this after how ChatGPT was trained. How DeepMind trained AlphaGo. Like, it just learns from behaviors and data. It could change the way software is built in the future, all software. So if you think about what goes into building CRM software, it's all kinds of code, hundreds or thousands of lines of code that tell it what to do.
[01:04:42] Paul Roetzer: What it's saying is just watch a great email market, learn what the email marketer does, how they do their job. And then develop automated systems based on what the great email marketer does. So if it works, one, it'll unlock another trillion dollars in value for Tesla stock, which is probably part of the reason it's up so much today.
[01:05:00] Paul Roetzer: Two, it could change the way software is built across all industries.
[01:05:05] Mike Kaput: Wow. So ending on a light, fuzzy topic here. I love it. Yeah. Tons to unpack this week. Paul, appreciate you taking the time and energy to help us understand what's going on in AI and in Elon Musk's brain this week.
[01:05:20] Paul Roetzer: And I hope people enjoy these bigger macro topics because to me, they're just, they're so fascinating to talk about.
[01:05:26] Paul Roetzer: They're so interesting to learn about. And I hope that by us explaining them the way we do. The context starts to matter to you that like, why do it sucks? It's actually understandable if you understand the history of how this stuff all came to be and why it might not suck three months from now. Again, you have to kind of understand the history.
[01:05:46] Paul Roetzer: So I feel like some of our role here is to try and unpack why all this is connected and why it matters. So that it's not just a bunch of product updates and Twitter threads of 10 ways to use chat GPT. Like you can get that anywhere. So hopefully this is really helpful to people because. I enjoy talking about it.
[01:06:03] Paul Roetzer: It's like, otherwise it was you and me standing around the coffee machine, talking about
[01:06:07] Mike Kaput: this stuff. Well, I definitely know it's helpful to the audience based on the amount of people that reach out every week, about the podcast saying how much they enjoy it. So please keep doing that.
[01:06:16] Paul Roetzer: Cool. All right.
[01:06:17] Paul Roetzer: Well, thanks. We will talk with everyone again next week. Thanks Mike. Thanks Paul.
[01:06:21]
[01:06:21] 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.
[01:06:43] Until next time, stay curious and explore AI.