It was another week of exciting news in the world of AI! Paul and Mike talk about recent ChatGPT announcements from OpenAI’s inaugural DevDay, xAI’s announcement of their new conversational AI agent Grok, and Gavin Baker’s analysis of enduring foundational models. There is lots to catch up on and understand, tune in to learn more!
Listen or watch below—and see below for show notes and the transcript.
This episode is brought to you by our sponsor:
Meet Akkio, the generative business intelligence platform that lets agencies add AI-powered analytics and predictive modeling to their service offering. Akkio lets your customers chat with their data, create real-time visualizations, and make predictions. Just connect your data, add your logo, and embed an AI analytics service to your site or Slack. Get your free trial at akkio.com/aipod.
00:02:26 — OpenAI made big ChatGPT announcements at DevDay and unveiled “GPTs”
00:19:33 — xAI and Elon Musk announced Grok, xAI’s new conversational AI Agent
00:30:28 — AI watcher, Gavin Baker, breaks down the future of foundational models
00:37:17 — Microsoft 365 CoPilot is now available for select enterprise customers
00:40:53 — Hubspot acquires Clearbit
00:43:01 — AI godfather Yann LeCun warns against AI one-percenters monopolizing power
00:47:01 — AI leaders submit letter of concern to Biden over open-source AI development
00:48:50 — AI company creates chatbot of AI journalist and commentator without her consent
OpenAI makes huge ChatGPT announcements at Dev Day
Just a couple hours before recording this podcast, OpenAI wrapped up its first-ever Dev Day, a conference for developers where the company made some big announcements. The biggest announcement was the unveiling of “GPTs.”
GPTs are custom versions of ChatGPT you can create yourself to do, well, anything you can think of. GPTs function like a version of ChatGPT that’s tailored for a specific purpose or task.
For example, you could build a GPT that functions as a creative writing coach, a negotiation assistant, or even tailor one that provides helpful information when you are doing your laundry.
Absolutely no coding is required to build GPTs. You simply carry on a conversation with ChatGPT to build these, telling it what you want to see in the tool, giving it instructions to follow, and uploading information the GPT can use to produce outcomes.
Altman demonstrated the functionality on-stage, building a prototype of a GPT called Startup Mentor in seconds that gave advice to startup founders—and informed its advice with insights from past on-stage talks that Altman himself has given containing advice for startups.
You can keep the GPTs you build private or share them with the wider community.
OpenAI announced that it is rolling out the GPT Store later this month, which will feature GPTs from verified builders. Enterprise customers can also deploy their own internal-only GPTs that safely use their data and information for specific purposes or tasks.
The event included a slew of other announcements as well, with notable ones including:
GPT-4 Turbo, is a new and improved version of GPT-4 that is more powerful and less expensive. The model also draws on a more recent knowledge base, including information up until April 2023.
GPT-4 Turbo also has an expanded context window of up to 128,000 tokens, a 4X jump from the previous version of GPT-4 with the biggest context window. This is also bigger than Anthropic Claude’s 100,000 token window. (This amounts to about 300 pages of content.)
The Assistants API. An API that makes it easier to build assistive experiences, including those powered by voice.
Custom Models program. OpenAI says a team of its researchers will start working with a small group of select enterprises to train GPT-4 on their custom domain.
Copyright Shield. A program where OpenAI will cover legal costs for customers caught on the wrong side of copyright lawsuits.
Elon Musk has announced his own version of ChatGPT, Grok
On Saturday, Musk announced Grok, an AI agent designed to answer any question conversationally.
According to an announcement from xAI, Musk’s AI company:
“Grok is designed to answer questions with a bit of wit and has a rebellious streak, so please don’t use it if you hate humor. A unique and fundamental advantage of Grok is that it has real-time knowledge of the world via the X platform. It will also answer spicy questions that are rejected by most other AI systems.”
xAI says the model is still in very early beta, based right now on only two months of training.
Musk said that Grok has “real-time access” to info on X. Musk has also indicated that all subscribers to X’s Premium Plus plan, which is $16 per month, will get access to Grok once it is out of early beta.
A quick note on the name: Grok (spelled G-R-O-K) is a popular term among science fiction fans that was coined in Robert Heinlein’s classic sci-fi novel A Stranger in a Strange Land.
In the context of the book, “from” is a Martian word that basically means you have established a profound understanding of something or someone at a very deep level.
The name/word “grok” is often used in geek circles to communicate that someone has a comprehensive understanding of a subject bordering on the intuitive.
Will only four foundation models have lasting value?
An AI watcher just dropped a brilliant analysis of where AI foundation models are going—and it’s been publicly endorsed by Elon Musk.
Gavin Baker, the managing partner and CIO at Atreides Management, an investment firm, broke down how he sees the future of foundational models like ChatGPT/GPT-4 and Grok.
Baker says:
“Foundation Models without significant RLHF—reinforcement learning from human feedback and access to high-quality proprietary datasets are likely the fastest depreciating assets in human history.”
He thinks only four of the foundation models out there are likely to have lasting value and transition to becoming true AI agents over the next few years.
Those models are: ChatGPT, Google’s upcoming Gemini model, Grok, and Meta’s open-source LLaMA.
He basically argued that these four models are the ones that have both robust RLHF mechanisms and access to proprietary data—for example, ChatGPT has access to Microsoft’s data through their partnership and closed data within enterprises, Grok has access to X, and Gemini has access to all of Google’s data.
LLaMA, by virtue of being open source, is included because it can be applied to any proprietary data anyone chooses.
He even makes the claim that, if this assessment is correct, ChatGPT basically goes to zero without Microsoft’s data.
After Baker posted this on X, Elon Musk himself weighed in saying it was an “Extremely insightful analysis.”
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
Paul Roetzer: [00:00:00] I'm glad I don't own a startup that probably got obsoleted by this, but I think that those people who did. Are going to have a world of new possibilities ahead of them to build even cooler, I think it's just going to be an explosion of entrepreneurship with this kind of capability.
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.
Paul Roetzer: My name is Paul Roetzer I'm the founder of Marketing AI Institute, and I'm your host.
Paul Roetzer: Welcome to episode 71 of the marketing AI show. I'm your host, Paul Roetzer, along with my cohost, Mike Kaput.
Paul Roetzer: And we are coming to you with a special edition sort of this week because We are recording this at [00:01:00] 3 p.m. Eastern time on Monday. Usually, we do this at what, 9 a.m. Eastern time on Monday. I think nine or 10. Yeah. Okay. So OpenAI had their DevDay today, their inaugural DevDay. We knew it was coming, but as the weekend progressed, it started to become pretty apparent that some significant news was going to come out of this DevDay.
Paul Roetzer: So Mike and I chat over the weekend, all right, let's bump the recording of this to the end of the day, Monday, so we can get. OpenAI's DevDay news in and we are very happy we made that decision because Mike and I both watched the opening keynote from Sam Altman and that is going to lead off our topics today.
Paul Roetzer: But first, today's episode is brought to you by Akkio. The generative business intelligence platform that lets agencies add AI-powered analytics and predictive modeling to their service offering. That would've been cool back in the day when we were working for my agency. Akkiolets your customers chat with their data, create real-time visual visualizations and make [00:02:00] predictions.
Paul Roetzer: Just connect your data, add your logo, and embed an AI analytics service to your site or Slack. Get your free trial at Akio. That's akkio.com/pod So yeah, if you're in the agency world, absolutely check that out. We would have been doing that ourselves a couple of years ago. All right, Mike, lots to cover starting with OpenAI DevDay, so let's dive in.
Paul Roetzer: Awesome.
Mike Kaput: So, just a couple hours before we started recording, OpenAI started, kicked off its first-ever DevDay, which is their conference for developers, where they made some very, very big announcements. And the biggest announcement was around the unveiling of something called GPTs. GPTs are custom versions of ChatGPT that you can actually create yourself.
Mike Kaput: And they can do basically anything you can think of, it sounds like. So GPTs function like a version of ChatGPT [00:03:00] that is tailored. For a specific purpose or task. So for example, in some of the examples they've shared, you could build a GPT that functions say as a creative writing coach or a negotiation assistant, even something that gives you a bunch of information that's helpful when you're doing your laundry on different stains and what clothes to put together.
Mike Kaput: And the cool part is there's absolutely no coding required to build. GPTs, you can simply carry on a conversation with ChatGPT to build these. And you can just tell it what you want to see in a given GPT tool, give instructions to follow and upload information that your GPT can use to produce outcomes.
Mike Kaput: So Altman actually, Sam Altman, CEO of OpenAI demonstrated this functionality right on stage. By building a prototype of a GPT, he called startup mentor and he did it basically in seconds just by typing in, build [00:04:00] me a GPT. That's going to essentially give advice to startup founders. And what's cool is he also uploaded a bunch of his transcripts from say, past talks he's done on stage about startups that had advice.
Mike Kaput: And this assistant was then able to inform its advice using what Altman himself had said in the past. Now you can actually keep these GPTs private or actually share them with the wider community. And OpenAI announced that it's rolling out something called the GPT store later this month. And that'll feature GPTs from verified builders.
Mike Kaput: Now, on top of that. ChatGPT Enterprise customers can also deploy their own internal-only GPTs that can safely, you know, use their own information for specific purposes or tasks. Now that was just one of many announcements. Some other notable ones that came out of this event include the launch of GPT-4 Turbo.
Mike Kaput: [00:05:00] Which is a new and improved version of GPT four, that's more powerful and less expensive. What's really cool is it also draws on more recent knowledge base. So it actually includes info up until April of 2023. So soon ChatGPT will have a much better updated memory.It also expands the context window. Of ChatGPT up to 128,000 tokens.
Mike Kaput: That's a four x jump from the biggest previous version of GPT four and much, much bigger than the typical version we are often using as ChatGPT users. Significantly. It's also bigger than something we've talked about quite a bit on this podcast and throw at Claude's a hundred thousand. Token context window
Paul Roetzer: and just for context that is about 100,000 words.
Paul Roetzer: So when you have 128, 000 tokens, it's basically like you could, our last book was like 55, 000 words. You could basically feed it twice.
Mike Kaput: A couple of other [00:06:00] things they launched that are really important in our opinion are the assistance. API, and this API makes it easier to build assistive experiences, including those powered by voice.
Mike Kaput: They showed off some really cool voice assistants they were able to build using that API. OpenAI also said it is launching what it's calling a custom models program. So it says a team of its researchers will start working with a very small group of select enterprises. To actually embed with them and train GPT-4 on their own custom domains of expertise.
Mike Kaput: And then one more that jumped out is something they call copyright shield, which is a program where OpenAI will start covering customers that are caught on the wrong side of copyright lawsuits. So Paul, suffice to say, my head is spinning. Um. First up, let's talk about GPTs. How big a deal are these?
Paul Roetzer: Yeah, it was a lot. You and I were both feverishly taking notes and trying to process all of this. [00:07:00] The GPTs in particular, you know, obviously jumped out to both of us. I think it sort of was the big news of the day to the nondeveloper crowd. I think it's a big deal no matter what.
Paul Roetzer: Obviously this event is for developers. So there were other things like the assistant API that I saw some developer friends were geeking out about to us. It's not as huge of a thing because we're probably not going to go in and use that per se. But the GPTs is for everyone, like literally giving us all the ability to program.
Paul Roetzer: So I know you and I were both doing the same thing. Like I'm just. As Sam's still talking, my mind is spinning with things we could build. And I think the real key here, as you mentioned, is zero coding ability needed. Massive disruption to the startup ecosystem. There's a lot of startup founders today who probably don't have a business model anymore.
Paul Roetzer: Like it just. It became so easy to build things. This is going to have an impact on existing companies, may make a bunch of, you know, tools that have been creating the last six months [00:08:00] obsolete, but opens up an entire new world of startups that can emerge out of this. And you don't have to be technical to build them now.
Paul Roetzer: So, you know, posted on LinkedIn right after it came out. And my first thought was wow, this is just democratizing programming. We've talked about Replit is going this direction. I want to build, you know, make a billion programmers. This is the idea that a programmer in the future just talks to the machine, just tells it what you want.
Paul Roetzer: And so I initially wrote like your imagination is the only limitation. And then I actually went back when I was editing my own post before I published it. And I realized Oh, wait, I could just go into GPT for turbo and say, I'm an entrepreneur, you know, specialize in marketing. We help clients, you know, with AI literacy and driving strategy.
Paul Roetzer: Like what are some tools I could build that could help us help these companies become, you know,use responsible AI and.Take a human centered approach, this is whatever, and like let GPT-4 tell me what GPTs [00:09:00] to build. And then I'm Oh, okay. Like those sound amazing. And I was okay, give me the prompts to build number four.
Paul Roetzer: Let's, let's go ahead and do that one. So like the idea of like experimentation and ideation around business development and tool development is it's just different. It's just one of those moments where you're okay, that just changes the way. You conceive of these ideas and the way you build prototypes.
Paul Roetzer: And yeah, I just think it, you could be anybody, you could be a marketer, a sales rep, a customer service rep, an entrepreneur, you, you basically have the, the ability to now build tools around knowledge. And that's a really, really powerful thing that I think it's going to take us a little while to process what that actually means and the impact it could have, but I know you and I, before we jumped down, we were saying like we used to runHackathons, like idea hackathons, basically marketing strategy hackathons for our clients.
Paul Roetzer: When Mike and I worked at an agency and then, now we do these applied AI workshops, [00:10:00] we're identifying use cases for people and, you know, how to pilot AI. And I was we should just run one of those. For GPTs, like just get all these ideas flowing of things we could build for our event business, for our media business, for our online education business, for advisory services.
Paul Roetzer: Like we do all of these things now. And so much of it is knowledge based and this to me, my head's just like swarming with ideas of things to build. And I was trying to like. Bring myself back in because I'm doing a talk tonight at like six o'clock and I went in and changed the deck for that talk. I was I gotta, I gotta update the presentation for tonight.
Paul Roetzer: So yeah, I just, I think that they're likely going to be a huge deal and I'm not sure we'll comprehend fully yet how big of a deal for another couple months here until we start seeing the flood of these things emerging.
Mike Kaput: Yeah, tell me if this resonates for you at all, but watching it, I was just so immediately lit up and excited because I feel like we have such a background [00:11:00] in being consultants and essentially coaches, trainers, educators, and we've created a fair amount of over the years.
Mike Kaput: Intellectual property around certain frameworks, hackathons, you mentioned, and I just immediately thought wow, the limiting factor for us hasn't necessarily been the ideas. It's been how do we build scalable tools and frameworks based and processes based on those. And this just made me. Feel extremely excited because the hard work of arriving at those frameworks is done, it's wow, we can bring it to so many more people with these types of tools.
Paul Roetzer: Yeah. I won't get into like the specific examples I have on my mind already. So we don't go build them, but I'm with you 100%. Like there's,I don't know. I'll, I'll give one example like that anybody could do, but if you do a lot of public speaking or interviews or podcasts, things like that, just to build.
Paul Roetzer: Like a GPT based on that knowledge [00:12:00] base. So if someone to be able to like have a conversation with you, like they could do it. And it's just trained on and has access to all of that knowledge.Yeah, it's just, it unlocks knowledge in entirely new ways. It makes that knowledge interactive and it is exciting.
Paul Roetzer: Like I'm, again, I'm glad I don't own a startup that probably got obsoleted by this, but I think that those people who did. Are going to have a world of new possibilities ahead of them to build even cooler stuff.And, yeah, it's going to be so fascinating. I think it's just going to be an explosion of entrepreneurship with.
Paul Roetzer: This kind of capability, because it was one of the limiting factors previously, is you needed a technical co-founder if you wanted to build a technology company. And I don't know that that's going to be true anymore.
Mike Kaput: Now, I want to talk about that point. You've hinted at a couple of times about the effect on startups, people building AI tools, SaaS solutions out there.
Mike Kaput: I think I, you know, tongue in cheek [00:13:00] assumed you that some of those founders are probably taking their first drink right about now, when some of these announcements came out. It does seem with. Despite all the opportunity that we're all excited about that overnight, this might change the game for some of the vendors and third party, tool providers out there.
Mike Kaput: Could you walk us through how you're looking at that startup ecosystem in light of GPTs?
Paul Roetzer: Well, I think we saw it as being pretty risky anyway. Like we've talked about for, you know, the last almost 12 months now, this idea that know, what is the mode of these SAS companies? If they're just like thin wrappers for OpenAI technology, like what was stopping OpenAI from just building those features themselves or the models themselves becoming capable of doing these things.
Paul Roetzer: And I think this is just the next step. Like it's just so obvious that. You know, there's very few knowledge work related things [00:14:00] that GPT-4 and its next iteration than the other foundation models aren't going to be able to do, especially since this is just a prelude to the AI agent thing. You, you referenced this, but I made a note of, let me see what it was.
Paul Roetzer: They said it's just a step toward AI agents. So we've talked about the idea of interactive agents on, on this podcast. Many times. We even had a whole,topic dedicated to Andrej Karpathy, who went back to OpenAI after heading up AI at Tesla for five years. In, in the spring of this year to work on this exact concept of these agents that can take actions.
Paul Roetzer: And so it's pretty apparent to me that OpenAI is further along than they're, they're telling the public.Because Sam specifically said, I quote, I wrote it down, gradual and iterative deployment. So that society can get prepared, like they are moving very quickly toward a day where not only will you be able to build these things, they're going to be able to take actions on [00:15:00] your behalf.
Paul Roetzer: They're going to start doing the work of knowledge workers in many ways. And so it's, it's very apparent that that's where they're going. And I think this is one of the really important things that came out of today is like to, to future proof your company, to future proof your job. You really have to start looking at the tea leaves here of like what they're telling you they're going to be able to do.
Paul Roetzer: The other one that I made note of was Sam said, I think you, you mentioned this. Was what we launched today is going to look very quaint at the developer conference next year. So when we're back together again next year, this is going to look quaint, which I use a quote from Greg Brockman, the co founder of OpenAI in my talks all the time, that like Greg says, the most amazing fact about AI is that even though it's starting to feel impressive, a year from now, we'll look back fondly on the AI that exists today as quaint and antiquated.
Paul Roetzer: So this is obviously like a frame of mind within OpenAI. Greg tweeted that in February of this [00:16:00] year, a month before they released GPT-4. So when they talk about their current tech, which we all look and say, Oh my gosh, GPTs, we can, as a non developer, I can build something. They're Oh my God, just wait.
Paul Roetzer: Like they know they've got something way more powerful than this and that it will be ready to unleash on the world a year from now, six months from now. SoI think it's just really important to not just look at what they're announcing, but to look at what they're, this is leading to, this is, it's still very early innings of AI development.
Paul Roetzer: And so I think this is just a sign of more significant things to come. And so in the startup world, like really trying to think is the thing we're building future proofed or is OpenAI just going to launch a tool in three months that obsolete our entire idea.
Mike Kaput: Yeah, the AI agents conversation is continually fascinating because it really is a game changer if we end up having tools that can take actions on our behalf and that really changes the [00:17:00] value equation of frankly a lot of startups I would say out there if that comes to pass.
Paul Roetzer: Yeah. So I, you and I obviously both follow Ethan Mollick, the Wharton professor, and we've talked about it on the show, but he tweeted right after the event.Overall, I think a lot of AI startups and company initiatives just got folded into the core functions of GPT-4.The system is now faster and cheaper, works better with proprietary data, including using PDFs and documents when appropriate, and is more customizable by non coders.
Paul Roetzer: I was giving a talk at MIT early last week, and half the AI startups had ideas that were great, and would also clearly be mostly eliminated a week later, parenthesis, I couldn't talk about that. About what I knew about GPTs at the time, this is a tough field to be a founder, a good field for users that pretty much sums it up.
Mike Kaput: So as we wrap this topic up, did any of the other announcements stand out to you as [00:18:00] particularly significant here?
Paul Roetzer: Yeah. I think the GPT for turbo thing's a big deal, but no more like just from a user experience that the knowledge base is now from April, 2023, instead of September, 2021, no more picking your models, which was quite annoying.
Paul Roetzer: Like. My browsing, my doing DALL-E 3, am I using advanced data analysis? Am I doing a traditional, like you don't have to do that anymore. It just knows what you're doing. So I do think just the overall ChatGPT user experience jumped. And then the a hundred million weekly active users is the first time I think we're hearing that number.
Paul Roetzer: The OpenAI has been pretty guarded about the number of users for ChatGPT. And the number I kept hearing was still the one from January of a hundred million active users that month. And now we're at a hundred million active users per week.That's a pretty big number.
Mike Kaput: And just one more argument for essentially why you should be paying for a ChatGPT plus at this point.
Paul Roetzer: Yeah. They got to raise the price at some point, right?
Mike Kaput: Yeah. [00:19:00] I've seen people online and I start to agree with them that they'd pay five to 10 times the amount that they're paying now for access to the tool.
Paul Roetzer: I think like if I was if I was being questioned about it,I think I would probably pay.
Paul Roetzer: At least a hundred bucks per month per user right now. I don't know. But I'm in a small business environment. I'd be willing to do that. I think like if it was you and I, and it was well, we need access. Like I would, I don't think I would think twice about paying a hundred bucks a month for each of us.
Paul Roetzer: Yeah. Yeah,it's a, it's a great value.
Mike Kaput: All right. Next up, another huge announcement. Elon Musk has announced his own version of ChatGPT on Saturday. Musk announced Grok, G R O K. We'll get to that in a second. An AI agent designed to answer any question conversationally.According to an announcement from X.
Mike Kaput: A. I., which is Musk's A. I. company, quote, Grok is designed to answer questions with a bit of wit and has a [00:20:00] rebellious streak, so please don't use it if you hate humor. A unique and fundamental advantage of Grok is that it has real time knowledge of the world via the X platform. It will also answer spicy questions that are rejected.
Mike Kaput: By most other AI systems. Now X.AI says the model is still in very early beta. It's based right now only on two months of training. Musk says that Grok basically has real time access to info on X. And he's indicated that all subscribers to X's premium 16 a month will get access to Grok once it's out of early beta.
Mike Kaput: Now, a quick note on the name, Grok, which is, again, spelled G R O K, is a popular term among science fiction fans that was coined in Robert Heinlein's classic sci fi novel, A Stranger in a Strange Land. Now, in the context of that book, Grok is a Martian word that basically means you have established a [00:21:00] profound understanding of something or someone at a really, really deep level.
Mike Kaput: So it's actually often used, you know, in like geek circles or among tech circles to communicate that someone has a comprehensive understanding of a subject so much so that it borders on the intuitive. So Paul, let's first talk about why Grok is important. Is this a legitimate competitor to. OpenAI, Google, Microsoft, and the other major players in the model space.
Paul Roetzer: SoI, no, not yet. You know, I don't, I don't think it's reasonable to say it's competitive. All we have right now is this was announced over the weekend. It was quite random, which we'll talk about in a minute. They've been training it for very little amount of time. They've only even been pursuing it for a very short amount of time.
Paul Roetzer: So there's nothing about it that tells you it's actually competitive other than their own data, which says it's basically at. GPT-3. 5 level based on [00:22:00] their internal evaluations, which no one externally has assessed it yet. So that being said, it does have Twitter data, X data, and it has Elon Musk behind it.
Paul Roetzer: So,we've talked a lot about. The, you know, I think, I think the final topic we're going to, big topic we'll cover today is these foundation models and what separates them. But one of the key things is the proprietary data they have access to. So the fact that they have a pipeline to real time Twitter data is unique because Elon Musk turned the API off to everyone else.
Paul Roetzer: So no one else has that data at the moment.You could. Question whether the Twitter data is really that valuable in the current state of Twitter. But, if we go under the assumption that Twitter's real time data is extremely valuable as a representation of the pulse of society, which is what they're going for, [00:23:00] then yes, it, it has the potential to very quickly become a major player in the space.
Mike Kaput: So I guess.If it does become a major player, like why do we need another foundational model? How is this different or better or needed more than what's already out there?
Paul Roetzer: Well, I think for that answer, we may have to go into a little bit of history. So we've touched on this before on the, on the podcast, but Elon Musk was a co founder and the original investor in OpenAI.
Paul Roetzer: So Sam Altman. Elon Musk, Greg Brockman, Ilya Sutskevo, who we talked about recently on the podcast, they created OpenAI to be an open alternative to what Google was building. So back when they created OpenAI, they thought that Google was basically going to corner the market on an AI and they didn't care about doing it in a responsible and safe way.
Paul Roetzer: That was Elon's interpretation. So he teamed up with Sam, put, I think, 100 million in, creates [00:24:00] OpenAI. 2018, a few years later. They have a disagreement about the future of OpenAI. Elon decides to unceremoniously split or was pushed out one or the other. OpenAI launches a for profit component to the company and goes on their way.
Paul Roetzer: And Elon has, since that time apparently, been very unhappy with SAM and OpenAI and has continuously stated the need to build a competing company in essence. That He felt that, he needed to get back in the game. So while he is definitely one of the people trumpeting the dangers of OpenAI, the irony is all the people trumpeting those dangers are themselves building the most powerful AI in the world.
Paul Roetzer: So part of this is not actually that we need another foundation model. It's Elon Musk's desire to build a foundational model in his own,making of his own making of the things he finds to be essential to these models.So that's the OpenAI part. He also obviously has Tesla, [00:25:00] which is a major AI company, around, you know, trying to create autonomous driving.
Paul Roetzer: He has Optimus robots that he's trying to build intelligence into. He turned off the API to Twitter months ago so that these AI companies could stop scraping Twitter data. And he created X.AI, which is the AI company that is building this model. The other factor isHe bought Twitter for 44 billion. Two weeks valuation put Twitter at 19 billion.
Paul Roetzer: Thefastest way to make Twitter worth more is to make it a true AI company or to make the data of a Twitter power, a foundation model. So there's lots of factors here, but more Elon Musk's own desire. To build an alternative to OpenAI. The timing of the thing was certainly interesting. I think I noted that to you over the weekend was like.
Paul Roetzer: You know, he knew the DevDay for OpenAI was today. So they released this and said, it's a very, very [00:26:00] early beta. Almost no one has access to this thing. They've only been training for two months. And as I was cutting my lawn on Sunday, I was wait a second. There was no motivation to launch this thing today.
Paul Roetzer: Other than the fact that the opening day was right now. And he was just trying to like. I think mess with Sam and OpenAI. So I don't know. I don't, again, does the world need another foundation model? I don't think so. Does Elon need a foundation model for all of his initiatives? Absolutely. Like he's not going to build optimists as robots or his cars on Sam's.
Paul Roetzer: GPT-4 because he's has a beef with Sam right now. You know, Elon's approach is we'll build it and by, you know, switching Twitter data to becoming the fuel to power a real time AI engine. Now the sudden Twitter is potentially worth a lot more money.
Mike Kaput: So it sounds like really time is going to tell how valuable that data from [00:27:00] Twitter will be essentially in real time.
Paul Roetzer: Yeah. And obviously people have opinions right now of Twitter, the experience overall, you know, whether it's gotten better or worse since Elon took over,it's certainly become moreopen, I guess, in terms of what it allows on the platform. I will say that there's a couple of things that jumped out at me at the release of this.
Paul Roetzer: One is,there's no way they had time for red teaming this thing. So if you're on Twitter at all these days, you can tell that Elon is much more about let people communicate in whatever way they choose to communicate and the community will determine whether or not that information is viable or valid.
Paul Roetzer: So OpenAI red team, meaning like safety trust and safety, testing of GPT four for like six months before they released it on the world. This model has only been in development for two. So there's no way they've done any [00:28:00] real level of red teaming to ensure the trust and safety is built into this model.
Paul Roetzer: So I have a lot of concerns around how this model could be used and what it'll do and what it'll say, but my concerns are irrelevant here. They're going to release this thing no matter what.So I was paying for the premium on Twitter only because I was trying to see what it would do and what.
Paul Roetzer: You know, the functionality would offer and things like that. So the premium plus for 22 bucks a month, which when you compare it to ChatGPT for 20 months, seems like you're getting ripped off. I will pay it to see what it is. And I can tell you as a Tesla,driver for the last five years, I've had a Tesla.
Paul Roetzer: They did, they released this thing a few years back where it was like 10 bucks a month for basically live streaming data into your car and then all these other features baked in. And it became this enormous value. Like the 10 bucks a month is an absolute steal for what Tesla gives you. It is, it's like a lot of [00:29:00] value packed into the 10 bucks a month.
Paul Roetzer: So there's a part of me that's really curious to see what else they build into this premium plus plan, which I honestly didn't even know was a thing until I saw that the premium plus users would get early access. So I don't know, there's. You can't just look at this as another foundational model.
Paul Roetzer: There's too much other things that come along with Elon Musk and his, all of his businesses and all of his personality traits, I guess, for lack of a better way to say it, that fit into why he's doing this and whether or not it's going to work. I do think it'll be a, a foundation model. Like I think people who don't like Twitter and the experience there and Elon's way of communicating will hate this tool.
Paul Roetzer: But. Time will tell, I guess.
Mike Kaput: Yeah. Someone that read and very much enjoyed the Elon Musk biography. I can't say that I feel confident red teaming is something he. Would like to be doing.
Paul Roetzer: I'm guessing they don't have those people [00:30:00] onstaff.Yeah. Yeah. He doesn't hire communications people. Like he fired the whole communications team.
Paul Roetzer: He fired the ethics and safety team at Twitter. Likeit's just not his, his style. Soyeah, I don't know. I worry about this a little bit, to be honest with you.
Mike Kaput: So what's really cool is with these two big announcements, we also had this third topic that really, I think, ties together everything we're talking about here.
Mike Kaput: And basically, we just saw this AI watcher drop a really smart analysis of where AI foundation models are going. And interestingly, it's a take that's been publicly endorsed by Elon Musk on X. So this comes from Gavin Baker, who is the managing partner. CIO at a tradies management, an investment firm. And he broke down how he sees the future of foundational models like ChatGPT slash GPT four and Grok playing out.
Mike Kaput: So he [00:31:00] says, quote, foundation models without significant RLHF reinforcement learning from human feedback and. Access to high quality proprietary data sets are likely the fastest depreciating assets in human history by that. He seems to be saying that he thinks only four of the foundation models out there are likely to have lasting value and transition to becoming true AI agents over the next few years, both because they have mechanisms for our LHF.
Mike Kaput: And they have access to proprietary data. So the four models that he has identified as the longevity leaders here are ChatGPT slash GPT, four, five, et cetera. Google's upcoming Gemini model. Grok and Metas open source llama. So for example, he's saying things like ChatGPT have access [00:32:00] to Microsoft's proprietary data through their partnership, as well as proprietary data from within the enterprises they work with.
Mike Kaput: Grok has access to all of X's data and Gemini has access to all the proprietary data that Google has been collecting for decades now. Llama by virtue of being open source is included in this list because it can then be applied to basically any proprietary data set and tuned on that. So Baker actually makes the claim that if this assessment is correct, ChatGPT basically could go to zero if it didn't have access to Microsoft's data.
Mike Kaput: So after Baker posted all this on X, Elon Musk himself weighed in saying it was quote, extremely insightful analysis. So Paul, there's a lot going on here, but basically sounds like he's saying. Only a handful of these big models out there have both the proprietary data and the mechanisms for,human reinforcement feedback to [00:33:00] actually create unique defensible value in the marketplace.
Mike Kaput: What's your read on what he's saying here?
Paul Roetzer: Yeah. I think overall it's, it's a really smart take. I don't know about the going to zero. Like it,I think there's going to be so much verticalization of these models and so many like specific industry applications and being able to tie them to internal proprietary data.
Paul Roetzer: And you're going to have big enterprises that maybe don't want to work with any of those four options to build their models. So they're not going to necessarily. So I don't know that it's not like. Those other companies don't exist, but I do agree that there's going to be a centralization. Into a few key models.
Paul Roetzer: The one that I would throw in there that as of like last night at seven o'clock, so Kai-Fu Lee, who wrote AI super superpowers that you and I are both big fans of former president of Google China and now runs a venture firm in Google that his [00:34:00] firm announced, zero one dot AI.As a, what they said, proud to introduce the world's top open source model.
Paul Roetzer: Why I-34B as our first release, encouraging element projects that they say is more powerful than Lama.So, you know, I, the story isn't done yet. There, there are Baidu there's other players, but I think his basic concept of.The companies that have access to the data,likely have a massive advantage, like obviously Facebook slash meta has tons of proprietary data.
Paul Roetzer: Tesla X Grok, like they're going to have all kinds of proprietary data. If he starts merging Tesla data, we didn't even get into star, Starlink. Am I saying the right thing? Yeah. Yeah. Like their satellite data, SpaceX,like every, there's just, you can't count Elon Musk out. Like there's just so many proprietary data sources he's going to have and so much motivation.
Paul Roetzer: I [00:35:00] do think Gemini to me is. You know, has the opportunity to leapfrog like this fall. I think whatever Google is doing with Gemini and whenever that does come out, I think it's going to be the most powerful model. It's going to have the most proprietary data. It's going to have all the learnings of DeepMind baked into it.
Paul Roetzer: I could see Gemini sort of resetting the conversation, for a while, like until we see GPT five or whatever. I just feel like Google is going to come out with something that's going totake the lead. But yeah, I agree. I thought this analysis was just really interesting. And then that Chamath and Elon both replied to it.
Paul Roetzer: And I saw their replies and.Yeah, I think it's, it's something that's definitely a good thought to go thought process to go through.
Mike Kaput: So it seems like even if he's even directionally correct here, that.That has pretty big implications for people using AI models and building AI solutions. Could you maybe unpack [00:36:00] a few more of those implications for us?
Paul Roetzer: Yeah, I don't know that it changes anything in the real near term, though. Like if you're an enterprise or if you're a business, you know, whether you're a marketer, CEO, whatever it is. I don't know that this analysis changes the decisions you're making over the next, you know, three to six to 12 months.
Paul Roetzer: I think it's a little bit longer term. It's going to take a while for this to play out as to whether or not this ends up being true. I think directionally it's going to be true in the mid to longer term, but I would have probably more. This is like your CIO and your, you know, leadership are going to start digging more into this concept.
Paul Roetzer: And when you start to make You know, multi year bets, five to 10 year plays, we're trying to look out to the future. You're going to want to consider where do we think the market goes, but this is just one opinion. And I think to make those kinds of decisions where you're affecting the future of the company with who you're going to build with and partner with.
Paul Roetzer: You're going to take in [00:37:00] a lot more perspectives than this one, you know, tweet.But, you know, I think again, it's, it's directionally a really important perspective to consider as you're thinking about the long term strategy for your organization.
Mike Kaput: All right, let's dive into a few rapid fire topics this week.
Mike Kaput: So first up, Microsoft 365 CoPilot is now available for select enterprise customers. So remember CoPilot is Microsoft's AI assistant that works across its different apps. This AI product is now priced at 30 per user per month, but the Verge reports that enterprise customers will need to commit to at least 300 users.
Mike Kaput: In order to get access. So Microsoft right now says that 600 customers already have early access to co pilot and they're doing things like summarizing and generating docs in word, analyzing data in Excel, turning notes into plans, and much, much more [00:38:00] simply by prompting co pilot to go work across those apps, the verge also reports that Forrester.
Mike Kaput: Predicts up to 6. 9 million workers will be using Microsoft Copilot in 2024.So Paul, what's going on here with the pricing, the minimum users, is this any type of turnoff for enterprise customers with this launch?
Paul Roetzer: I don't think so. We've known it was going to be 30 bucks a month. For months, so that that's not new.
Paul Roetzer: The 300 minimum user threshold, I think is new. I don't believe I had heard that yet.I think the biggest thing is when I look at the adoption, I think about the impact that this is going to have on enterprises and knowledge work as a whole. There's three factors that come to mind, the education of the user.
Paul Roetzer: So do you know that they actually understand what these tools are capable of and the use cases, the onboarding with the technology, [00:39:00] and then. Downstream from that is the overall utilization rates of these different capabilities. So if you unleash like Copilot across all 365, and it's in emails, and it's in docs, and Excel, and PowerPoint, and all these places, and nobody's trained how to use it, I have actually talked with people who have access to 365 Copilot, and one, you know, one friend was yeah, you know, I use it to write emails sometimes.
Paul Roetzer: Like that's the extent of how it was being used. And so I think what's going to happen is, and I don't know what Microsoft's go to market strategy is besides selling the licenses, how you get utilization of the features to where they're truly infused into businesses. That to me is like one of the biggest, if not the biggest question about the impact this has is how is it actually diffused through an organization and how our teams talk the use cases and.
Paul Roetzer: Best practices, because we know from firsthand experience talking to these big companies, there's no one internally trained to be the person leading onboarding of 365 co-pilot. It's [00:40:00] a role that isn't really even be imagined by most companies we talk to. Right. So I think it's a huge opportunity for outside advisors, for you, the listeners, like You know, to get into an AI ops role where you're helping with education and onboarding and training and integration of this technology, because this is going to be an extended run.
Paul Roetzer: We're talking three to five plus years. Where this is going to take a while to figure out the impact it has within companies and how to prepare for it. And again, as I'm,our experience has been, those people do not exist within companies yet. And I think people are going to need to raise their hand and say, Hey, I would love to be involved with the rollout of 365 co pilot, you know, I think these are some of the challenges we're going to run into, I've got deep experience doing this within the company, like.
Paul Roetzer: I'd love to create a role and be a part of this rollout and make sure we get the value out of it. So I think that 2024 will probably see those roles start emerging more.
Mike Kaput: Another existing technology update and from a company we're very familiar with on November [00:41:00] 1st HubSpot Announced that the company has acquired Clearbit, which is a leading company that collects B2B data on businesses.
Mike Kaput: And it actually now bills itself on its website as the first AI native data provider, this acquisition will eventually. Give HubSpot customers access to Clearbit's data on over 20 million companies, right within the HubSpot platform. So Paul, obviously we've said many times, we're very familiar with HubSpot from our agency days.
Mike Kaput: The Institute runs on it today. It sounds like this is clearly a data play and therefore in some respects, an AI one. So can you unpack what's going on with the Clearbit acquisition from an AI perspective?
Paul Roetzer: Yeah, I assume it's largely going to be centered around personalization and prediction. So the more data you have, the cleaner that data, the better you can drive personalization within a CRM like HubSpot and the better predictions you can make around [00:42:00] churn, can, you know, conversion rates, things like that, quality of leads.
Paul Roetzer: You know, and we can save personal experience like HubSpot insights is, is offered within HubSpot. I don't know where the data necessarily comes from. But last time I looked, there was I don't know, a couple dozen data points that are enriched with some HubSpot insights data, our, our experience has generally been, it's not very reliable, like it's, it's actually pretty crappy data, for the key things we look at it to the point where we can't even really use it to make any kind of predictions.
Paul Roetzer: And we don't rely on it for personalization. So I think this kind of play where you get more reliable data, more real-time data infused in could be a really valuable play for a CRM company that starts to look at how do we drive smarter marketing, smarter sales, smarter customer service, smarter operations.
Paul Roetzer: Data is going to be core to all that. So it seems like a smart move. We've never used Clearbit ourselves. I can't speak specifically to Clearbit, but conceptually it makes a lot of sense why they would do this. I'm excited to see it.[00:43:00]
Mike Kaput: So in another news item, Yann LeCun, who is widely seen as one of the godfathers of modern AI, actually just put some AI leaders on blast over what he calls, quote, corporate lobbying with the U.
Mike Kaput: S. government in order to create regulatory capture. By influencing AI regulations. Now, LeCun mentioned people specifically like Sam Altman of OpenAI, Demis Hasabis of Google DeepMind and Dario Amodai of Anthropic. He claims basically they're attempting to make it so that a small number of companies functionally control the development of AI and that they're doing that by essentially lobbying the U S government to influence AI regulations.
Mike Kaput: LeCun has oftentimes and again said that he thinks AI doomsday scenarios, you know, this existential risk of AI running amok and out of control is just totally overblown. And that the real near term risk is that the future of AI ends up being dominated by a [00:44:00] handful of closed for profit Companies so Yann LeCun is one of the huge person in artificial intelligence like what is going on here with him coming out against some of these other companies, especially given the fact that he works, you know, at meta.
Paul Roetzer: It's the open versus closed source like this. This is what this is all about. So there are there are definitely people in the camp that think. Open source is key to everything. We touched on this a little bit with the executive order conversation last week in our podcast. Yeah. The big concern with regulatory capture is you have three, four, five companies who are all, stand to benefit the most from this, the big tech persona, as we talked about last week,who, you know, want the regulation because it's going to make it a lot harder for competition to show up.
Paul Roetzer: In particular, it may make it extremely difficult on open source companies and models. So Yann is a [00:45:00] major proponent at like. He, so the history with Yannis when he agreed to go and run the research lab at Facebook, when Zuckerberg recruited him, it was under the stipulation that Zuckerberg commit to keeping everything they did open.
Paul Roetzer: So all of their research, all of the, everything they built, because in Jan's opinion, That was what was, would attract the best AI researchers. They don't want their work hidden behind, you know, closed models and no one being able to talk about the work they were doing. So to Zuckerberg's credit, which I don't often give,he has stayed true largely to his commitment to LeCun through the years and they have continued to push out open source research and models.
Paul Roetzer: Now I I struggle with this one a lot because I see a lot of the justified fear in opening these models up. And so I wouldn't say I can confidently see the open source side of this. Like I think I actually tend to lean toward [00:46:00] there needs to be a point where there, we don't just open source everything.
Paul Roetzer: Like it is going to get very dangerous, but Yann is much smarter than me about this stuff. And he has a very specific point of view. But like we talked about on the executive order one last week, you have to consider the perspective of the people championing the different.So I would always listen to Yann's point of view.
Paul Roetzer: I read everything Yann writes. I follow any conversation he does. Because there's something to be learned from him.Always. And I think he's very... even keel about this. Like I think he's, he's very systematic and why he feels this way. And he's very good at diffusing the arguments for the opposite direction.
Paul Roetzer: But I just, it basically comes down to open source versus closed. And there are a lot of people who really think open source is critical to the future safety of AI.And there's a lot of people who think it is The largest existential risk to our future is allowing these open models out into the world.[00:47:00]
Mike Kaput: So as a follow on to that, LeCun actually also joined people like VC firm, Andreessen Horowitz, and dozens of other leading AI voices in submitting a letter recently to the Biden administration. And it basically cautioned that parts of the executive order on AI that came out last week could actually.
Mike Kaput: Harm open source development and essentially this letter argues that some of the reporting and safety requirements that the executive order outlines for top foundation models could also end up sweeping in small open source models that don't really pose the same types of threats or security risks and.
Mike Kaput: If that happens, it would make it much harder or outright impossible for open source models to develop freely. So we've talked a bit about this tension between the few powerful foundation models and the open source ones.Is this just another signal that some of the [00:48:00] bigger players might be going for regulatory capture?
Mike Kaput: Or is this just a misguided part of the executive order in your opinion? Do you think they knew about this as they're writing it?
Paul Roetzer: Now, I think this is them realizing there is a great risk here that if the government decides to over regulate, That it could have some pretty serious, near term impact on the open source community and movement, and that they need to take action real fast and they need to get vocal. And I think they probably feel like the big tech regulatory capture group.
Paul Roetzer: has probably been more successful at this point, lobbying the government and they need to step up their game to have their voice heard and hopefully influence this before it gets too far in their opinion.
Mike Kaput: All right. For our last topic this week, a little bit of an unnerving one, unfortunately, a user on X named, Sinead [00:49:00] Bovell, and I hope I'm pronouncing that correctly.
Mike Kaput: She is an AI educator and journalist, and she posted.An incredible and horrible story about an AI company that just straight-up trained a chatbot to mimic her, used her data without her consent. You know, she's got public data out there of commentary, videos, education, and then it actually notified her that it was up and running.
Mike Kaput: And guess what? It will soon include a voice clone. So she posted a video about this. Obviously is quite upset, has sent a cease and desist last week. She said, and the company has since taken it down, but she notes that this isn't a scalable, practical solution. And also, you know, what happens if they don't take it down?
Mike Kaput: SoPaul, I'm curious, like we've talked about the fact this is going to increasingly happen. You know, with politicians and personalities, but this is happening to , [00:50:00] you know, a professional who has an online presence.Obviously it's someone who's in the eye of the public, but not exactly, you know, a Kardashian out here.
Mike Kaput: So isn't something like this illegal? Like what protections exist against this happening to your average person?
Paul Roetzer: I assume it's illegal, but you gotta... Have the, you know, the resources to pursue it from a legal perspective, which usually costs money and attorney bills. We had this happen, obviously earlier in the year, we talked about music where people are stealing, you know, musicians, you know, voices and creating lyrics and releasing songs under their name, Drake, I think comes to mind as one.
Paul Roetzer: So you have this like weird camp, like people are going to do this all the time. And you're going to have the people who are like, Oh, that's cool. I think the all in guys, didn't somebody do this to their podcasts or they did synthetic versions of them? And and they played it up Oh, that's great.
Paul Roetzer: Like whatever. And they didn't, as far as I know, they didn't tell them to cease and desist. They like just rolled with it. So I think there's something, [00:51:00] something to keep in mind. One,not all. Startup founders are ethical people. Like they may steal stuff just to steal stuff and make money.
Paul Roetzer: The other ones are, I was, I've always found humorous. They'll steal your stuff and then they don't know that it's illegal. And they'll tell you they stole it and think that you're going to be happy they stole your stuff. And so that's the group I've just never comprehended. They do it. This has happened to me countless times in my 20 plus years in business where.
Paul Roetzer: Someone will just blatantly steal something and then they'll just reach out to me and say, Hey, I stole your thing. Check it out. Here's the link. And it's you're kidding, right?So, it happens all the time. Sometimes they think they're doing you a favor. Sometimes they know they're stealing your stuff.
Paul Roetzer: So all I would say is. For personal branding purposes, for corporate branding purposes,not only, you know, have your IP attorney, you know, on call on speed dial,but [00:52:00] monitor your brand, like have alerts set up to know if a course you created, a book you wrote, a PDF you released under a paid thing, liketrack that stuff.
Paul Roetzer: Because people steal intellectual property all the time. AI is just going to make it easier and it's going to make it multi modal. So instead of just text and links and stuff, now they can steal your likeness and your voice and everything. And it's, it's going to be prevalent. So yeah, I think something like this is just.
Paul Roetzer: A reminder that we have entered a new age where anybody can steal and recreate anything and it's going to be really easy to do and fast someone's going to create the GPT, like knockoff, where you can build these agents to do bad things and like it.It's a whole, I hate it and do the shows like this, but welcome to the age of generative AI.
Paul Roetzer: When we can all create anything we can imagine, it's going to be awesome. And then sometimes it's [00:53:00] not,well,
Mike Kaput: I'm going to pivot and end the show on a more positive note because I do want to give a shout out to the fact that every week on this podcast, we cover the most important stories. In AI and hopefully make it easier for you to follow what's going on and unpack why it matters, but we have limited time on the podcast.
Mike Kaput: And as you might've noticed a ton happens every hour of the day in AI. So we have dozens of stories we usually don't get to. So things we can't get to on the podcast are also included in marketing AI institutes newsletter that goes out every single week. And you won't just get a recap of what we discussed today in our newsletter every week, you will get new news stories that you need to follow.
Mike Kaput: It's an incredible weekly digest that can get you up to speed very, very quickly and cut through all the noise out there. So if you haven't checked it out yet, I would encourage you to go to marketingaiinstitute.com/newsletter. [00:54:00] So Paul, with that. I appreciate you breaking everything down.
Mike Kaput: I'm glad we were able to cover, some of the breaking news happening this morning, this afternoon as well.
Paul Roetzer: It worked out well delaying this. So hopefully you all, you know, it helped you all unpack all this. It was a lot.Like Mike and I said, we're still processing the GPT thing.
Paul Roetzer: Ourselves. It's going to be exciting. It's going to be fun. It's never going to be boring moving forward. So Mike, thank you as always for curating everything and leading the conversation. And we appreciate all of you listening each week, or watching if you're watching on YouTube, like my dad, Hey dad, give him a shout out to my dad who watches this every week.
Paul Roetzer: And we'll talk to y'all next week.
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 [00:55:00] explore dozens of online courses and professional certifications.
Paul Roetzer: Until next time, stay curious and explore AI.