Ep 38 - AI, Content & ChatGPT with Stew Fortier (Co-Founder & CEO, type.ai) and Kirby Winfield (GP, Ascend.VC

Foreign.

Hi everyone. I'm recording this intro on Valentine's Day. This discussion is not very romantic though. Fair warning, it's incredibly interesting on the subject of AI ChatGPT, generative content and much more. I have two guests today. Kirby Winfield. He's an investor in AI based companies. He has his own fund and a residency at the Allen Institute for AI here in Seattle. And then also Stu Fortier. He's an entrepreneur who has focused on content related businesses such as Foster and wellsaid Labs. And Stu currently has a new company. He's the co founder and CEO of Type, founded Type AI. I think you're going to really like this episode.

Welcome to the HI EQ Marketer. I'm John Busbee. This show is about learning from the best and most interesting marketers in the world, which are, not a coincidence, highly emotionally intelligent. If you'd like inspiration lessons, war stories, stories from startups or Fortune 500 companies, that's why we're here. So let's go.

Hi Kirby. Hi Stu. How you doing?

Hey John.

Yeah, Happy Valentine's Day.

I know there is nothing more romantic than recording a podcast about AI on Valentine's Day, but I'm excited, I'm thrilled that both of you are here. Thanks a bunch of. So Kirby, I'm going to start with you. You've been investing in AI based startups for several years now at the Allen Institute for AI and things have exploded, at least in the, in the popular media. How are you seeing the types of startups and companies evolving as you're receiving pitches and thinking about doing new investments? And what excites you most as a supporter and investor in new ideas at this time?

Yeah, I mean, I think the biggest difference between now and even, you know, four years ago, let's say, is just that, the time to value. For us, for a startup tackling a new problem in AI, it's just shortened tremendously. So I think in 2018 you had to pretty much bake in two years almost of really deep technical work because the models were not as sophisticated, were not as widely available. The compute was more expensive. So as a consequence, what you had was either deeply, deeply technical teams that may or may not have been optimized for go to market, or you had, you know, strong go to market teams that were kind of sprinkling AI. We used to call it sprinkling AI on top of just a kind of traditional vertical SaaS play, for example. And what's exciting right now is, you know, the time to value is shortened. The you don't need to roll your own, you know, you don't need to roll your own model, you don't need to roll your own stack. And I think that's really exciting from a standpoint of just like the reinvention of a lot of existing incumbent businesses that were not built with the DNA of AI on two fronts. One, you know, within, within the products that they're offering to the market. But two, increasingly these new companies can, you know, incorporate AI into their own business processes, which is going to make them highly, highly efficient. And so yeah, we're really excited about where the market's headed.

You talk about sprinkling AI and I kind of was chuckling a little bit because when I saw many companies talking about AI and then I might not necessarily startups, but larger companies, then you dig into it, I'm like, was this an Excel calculation? Is this really AI? And now, now fast forward to some of the stuff that I'm exposed to today and Stu, I'll, I'll direct my next question to you. You're an entrepreneur and you focused on a number of content related businesses. Foster well said Labs. Has the launch of Chat GPT been a surprise to you in terms of its achievement? I mean it's been kind of mind blowing for me, but I haven't been in it. And also in terms of its reception, getting to 100 million users and however many days shattering all types of records. What's your react been to that?

Yeah, I think Kirby's kind of thesis around time to value being shortened dramatically is probably illustrated best in ChatGPT. And granted, of course the OpenAI team's been around for quite a while. They've been doing a lot of the hard technical work. But the model behind ChatGPT had really roughly been out for a couple years. They did some reinforcement, learning to kind of improve it for chat. They did some refinements, but essentially the raw DNA was there. It just hadn't been packaged up and really optimized for the Chat experience until ChatGPT came out. And I think the one thing that kind of didn't surprise me is that if you reduce kind of a little bit of friction to be able to leverage some of these models, I'm not surprised that adoption rapidly increases. Meaning I think people really underestimate what great UI and just reducing the friction to access things does to adoption. It just, it's kind of unbelievable how much value gets created from what appears to be relatively trivial details. So that didn't surprise me. What did surprise me though is just the magnitude of adoption. You cited the 100 million user number, John, which is like even shocking to me. But that's just, I think kind of the new world we're entering. The second thing is specifically with machine learning, you know, if you had been tinkering around with GPT3, you know, a couple years ago when it came out, you could kind of get it to do what ChatGPT can do and. But it just took a lot of time to really like, whatever, find the right cues, become kind of a prompt engineer to figure it out. What I think is really astounding about ChatGPT is just the bit of fine tuning they did to make it optimized for chat. Drastically improved the quality and consistency in kind of the output. And so just making it much more like human accessible, just unlock these kind of latent capabilities that were already there. And so I think just the extent to which that was true, just this bit of optimization for making it accessible for most people to be, converse, interact with it in a conversational way. It's kind of shocking to me just how much quality went up and consistency. So, yeah, I think that's just like a taste of kind of what's to come.

Yeah, you're so right about the, the UI and the experience. You know, the previous product that got to 100 million really fast, TikTok, it's not like it was very. It's not like you couldn't record a video or you couldn't put text and video, or you couldn't edit a video, like all of those things you could do on your iPhone. TikTok made it so easy to do, which is one of the many reasons why it took off. Same thing here. And I want to shift into talking about with respect to content generation, accuracy, quality and creativity. One of the things that, that I did over the weekend was I asked Chat GPT to create a script, a Will Ferrell script about tennis. And I was, I was amazed that very quickly, like in 30 seconds, you know, it had a full script with a fade in, the fade out. You know, Will Ferrell was a tennis player, John C. Reilly was his opponent. You know, there was a, there was a setup where, you know, Will Ferrell was kind of a clumsy tennis player and then became a good one and won at the end. And I was like, the, the accuracy of what I asked for was like crazy, crazy good. And there was some elements which I thought were kind of creative too. You know, it knew to have a secondary character and so on, and at the same time, like it Wasn't funny, you know what I mean? Like, like it didn't, it didn't actually, didn't actually make me laugh. So there was something there about the creativity that was kind of, kind of missing. Let me start with accuracy first because I've seen Chat GPT, I've seen some on Twitter, I guess it's passed the bar exam, it's passed other difficult tests. Kirby. I know that, that I've, I've seen stuff that's happened at the Allen Institute where it's, it's done some really great things in, in deep science. Like where do you feel like things are happening in terms of accuracy and how far it can go and like what problems it can, it can tackle in the world of like calculation, passing tests, et cetera.

I think in talking to folks in the legal profession and the accounting profession and you know, and in the creative world as well, I think that sort of. The feeling right now is the technology can get you to some, some meaningful percentage of the work being done that you want done, but you need expert humans to finish the job. So an example would be, you know, I have a friend who was former general counsel at Facebook. It's been an angel investor for a long time and he was playing with Chat GPT and he said he asked it to, he gave it a kind of give it a term sheet and asked it to write a contract and it wrote a contract. And he said basically this is as good or better than a first year associate would write. Now when you have a first year associate write a contract, the partner needs to redline it, edit it, correct it. But the difference here is that you don't need the first year associate to do the first part of the work. So I think in the knowledge economy it's going to be hugely disruptive that way.

Yeah. And I would kind of extend that idea and say legal profession medicine, there's accounting, there's a kind of low margin for error in those fields. If ChatGPT writes your employment contract and screws up something important, the consequence of that's Fairly large. If ChatGPT is your writing assistant and yeah, you're riffing on a movie script with Will Ferrell, you can accept quite a bit of error and hallucination and kind of misfirings because ultimately it's more of a, know, low stakes creative piece that you're putting together. And I think in both of those cases though, it's like this spectrum, kind of what Kirby's talking about. There's this spectrum of work from total Novice through deep career expert on a topic. And over time, what's just going to happen is the starting point for AI is going to get more and more elevated. So the next iteration of GPT, that script might be a little bit funnier or more creative or more delightful and surprising. But I think for the foreseeable future there will be this really important merger with humans and there is going to be this very important, the crucial last mile of any job. And I think folks can become much more productive creatively, professionally. But there's going to be this necessary marriage between ever better AI and humans. And I think the conversation tends to drift towards humans being replaced. But I think at least for the foreseeable future, I think it's going to be this incredibly powerful merger between the two. And I think that's what's really kind of exciting to me. It's like not, not what can AI do, not what can ChatGPT to do. It's what can I do with it? What can humans do with these tools? I think that's really exciting.

Yeah.

And I think that's a huge sea change as well. I mean, I think for the last decade or so it's been, you know, when you're, when you're introducing new technology to customers, the last thing you want to do is sell the technology. You want to sell the benefit of the technology. And I think for the last 10 years we've just been selling the technology of AI and I think the big, the big sea change here is you're actually really able to bring forward the benefit and people see the benefit very quickly when they interact with the technology. And that's, that's, that's, I think that's going to unlock this entire new wave of companies like Stu's.

Yeah, and I think in the world of marketing it's really, it's really interesting. There's a lot of debate about it in some ways. Like I could see all these advantages in marketing because I can, I can ask ChatGPT or something to create like 10 or 20 variations of a text adjustment. Something that would take a long time and a lot of brain power and is mind numbing and it could do it almost, almost immediately. But then I see some concerns that every corporate blog is just going to be written by Chat GPT and so, so there's got to be, I think the ones that the marketing companies that win are going to have a dose of a, have a healthy dose of inspiration about what's your true, what's your true story? How does it intersect with the user story, etc. And then ChatGPT can embellish that or come up with variations, etc. So it's really, really interesting to me. I want to talk more about creativity and quality because we're kind of hitting on this. But I've seen AI generated content make me laugh, like in an unintentional way, but I haven't seen it, at least for me, have the nuance or intelligence to make me laugh as intentional comedy. Kirby knows this. I'm a big fan of Norm MacDonald. I asked ChatGPT to write a gambling story in the voice of Norm MacDonald, which was. I had no idea whether it would actually be successful. But just like Norm, he talked in a very folksy way. The output was very folksy. It was set up in the way that he would set it up, but it, but it wasn't funny. So like, like, is that just something that AI is never going to be able to do? Should we not ask AI to do it? What's your take on this?

Yeah, I'm happy to my bifurcation right now. I make a bifurcation between almost like substance and style, where styles presentation style is knowing that folksy way that Norm MacDonald delivers a joke or speaks style is write this in the style of Shakespeare or what have you. And then substance is obviously like, yeah, what's the joke? Is there actually meat underneath here? And I think I certainly don't know what the kind of mathematical equivalent of both is in the AI field. I think that this stuff will probably continue to get better at both. But I view substance very much as the human element of AI, where the potential is most exciting. You mentioned earlier, brands leveraging AI to create a lot of content for them.

Right.

One thing I notice with artists that use something like stable diffusion, people who are already artists or photographers even who use stable diffusion produce drastically more striking images than average Joes like me. I can't get. I can't prompt the thing. I don't have the like, intuition and sense of what a great image looks like, what a great shot looks like. And I notice folks who have that skill set without AI tend to come into these tools with a much better sense of how to leverage them. And so they get more kind of substance and just infuse the thing with a perspective that I think is ultimately what people kind of buy or drift towards. So I think, like, I think my guess is that this stuff will continue to get rapidly improve on kind of the style dimension, but that the most exciting point of leverage and like where human creativity comes in is substance. What's the, what's the thesis of this Will Ferrell movie? That's like really hilarious. What's like the, the setup to this joke that AI can just go and knock down? I think that's going to be like where humans spend a lot of their time. So that's kind of my sense. It's like a boring brand with AI is just going to be a boring brand with a ton more content. But like a really exciting brand with AI is going to like reach new heights that's going to be like really thrilling for everyone. But it's all about developing that intuition of how to apply the technology.

Yeah, I mean, I think when I think about it, Stu says it really well. The substance piece is the piece that's going to be really hard to replicate. And I think given that humans are born with sort of some pre programming and DNA and then we're just kind of a bunch of meat and electricity that reacts to, to every single input the minute we're born, till the day that we write our novel like that. Human experience is so varied and unique and, and ephemeral. I wouldn't prognosticate, but it seems, it seems like that will be the last frontier for the technology because we are just innately so, so unique and our experiences inform the content that we create or the stories that we tell in a way that, that even a really sophisticated technical facsimile, you know, can't do.

What do you all think about if AI content should have certain types of standards or disclaimers? You know, I operate in the world of review and comparison content and writing about products and writing about something that someone would use. And I think I was thinking going into this conversation about the first iterations that I saw, and you probably can't call this AI, but I think about a word cloud. Remember, like a word cloud was a way of taking all of this content and seeing what people were talking about the most about. And it was kind of obvious that it had been programmatically generated, whereas AI content is not always obvious. And there was some controversy about a big news site writing financial information, financial advice that that where AI was kind of splashed into it, but it wasn't totally obvious that that had happened. I, I wonder, you know, should stuff written by humans be certified organic? Or what type of disclaimers do you think are appropriate with the production of content with AI?

Yeah, I'm kind of curious what both of you think, but my sense is that in the same way, none of us care if the other used spellcheck before sending an email. I certainly wasn't like, oh, did John use Grammarly to add any missing commas to this email? I think something analogous or something similar will eventually be true of AI generated content. We won't necessarily care how it was generated, like how the metaphorical sausage got made or whatever. We'll care if it's true, if it's useful, if it's accurate, and kind of how it got there is almost less relevant to some extent. So that's like my intuition is over time these things will just get infused in everything that we do, into all of our creative tools, into the very way that we work and that the next, the reason right now we care if it was human generated is because we know that this technology is still kind of unreliable with truthfulness. We know it can be inaccurate, we know it can just make stuff up where, I mean, obviously humans have that same problem, but hopefully to a slightly lesser extent. That's one reason we certainly care if it was human written or not. So I think as the technology improves, accuracy gets better, that'll be less of a concern, but a very much an open question in my mind. So I'm curious if that's shared.

I mean, I think my, my brain immediately just goes to, because I, you know, consume a lot of sports content. There's been, you know, automate automated article generation for, you know, sports recaps for a decade. There's been automated SEO content for a long time. What, what I think consumers care about in those instances is the accuracy, like you said, Stu, of the information. And where is that input coming from? And that leads me to kind of a question that I've been asking around. How do we attribute value to the sources that are being ingested and manipulated to create these, you know, these new outputs, whether they're image, text, sound, whatever they are. Because historically you've kind of said, okay, so I shifted to more of a business model discussion, right? Historically you've said, sure, you can, you can automatically generate a write up of the, you know, local high school, you know, basketball championship, but you need to make sure that you attribute where you got the scores from, right? And in a world where, you know, these massive models are ingesting millions of images and then creating art that hypothetically could be worth millions of dollars, how does that value get passed back to the source content? And I think that's one of the more interesting frontiers that's opening up from a business perspective. But yeah, I mean, I think at Some point source will always matter and it's how do you confirm and attribute value to the source?

Stu, I love the analogy you had about using spell check or Grammarly. I almost go one step further where if someone doesn't use spell check, I'm like, do you even care about what you sent me? And, and yeah, I agree like it with factual Wikipedia style content even going, even going beyond that, like whether like some of the examples that you provided are important. Now think about like a TV ad though. Any TV about investing has a 30 second disclaimer. All this white text at the bottom. But, but where, where do we kind of draw the line if, or should a line be drawn with AI? You know, some of the stuff that I've, I've seen are, are answering questions like is such, is so and so financial instrument safe? And imagine that that someone was writing an article in 2008 about whether collateralized debt obligations were safe. And, and this, this is just an AI project done by an editor somewhere. And, and then you know, they're, they're, they're sued. Like what, what happens in this kind of situation? So yeah, that's, that's sort of where, where my head is around like where, what are the, what are the fringes? I can almost imagine like you know, like the class action lawyer companies that I see asking about accidents and something sort of focusing on, on AI too. There's been another type of debate around whether AI generated content is inclusive and culturally sensitive. And how do you react to that question? I mean, where my head is, it's all in the data set that the language model is pulling from. Is it that simple or is there more to it than that?

I know I'd be curious to get both your takes there as well, because I think, I don't know if it's too legitimate of a debate, but there is some debate on Twitter at least around if these AI, quote unquote safety efforts are really what they claim they are, or are they really just an attempt to kind of control the thing completely and just have it do what the entity developing it wants it to do. What I think is kind of a cynical view because the reality is if you just train these models on what is on the publicly available Internet, it's going to see some of the worst, vile, darkest parts of humanity. And I think that's the last thing. I think the vast majority of people don't want a super intelligent machine to take on certain negative and clearly harmful traits of its creators. Right. I do think we want to have an opinion about what good is and what's a desirable outcome here. And I do think there have to be guardrails. And John, to your point, I do think a huge percentage of it is like the training data that you show it, it can't learn something it's kind of never seen and like it's a terribly nuanced question but my kind of thinking here is like one, you have to actually care about it and make it a priority because if you don't then you won't put the guardrails in place with your training data with the feedback loops that you create to specifically mitigate the more problematic aspects of these models left kind of unconstrained. So one is I think literally just caring about it and like making it kind of a stated priority. And the other one for me is like having a diverse set of perspectives on what falls in the bounds of problematic behavior. Right. It can't just be a small number of people making that decision. You do need to have the impacted have a real voice in that conversation as well. So those tend to be a little abstract or somewhat esoteric but I think it's just absolutely something that we have to care about. Um, and if you don't, I just don't think these things behave well inherently on their own. You have to have an opinion. So I don't know. That's my somewhat high level take.

Yeah, it's, it's challenging. I mean you've seen it that historically, you know, search is basically AI, right? It was just like unsophisticated AI for a long time and you've seen historically big challenges for search companies to not, not have their algorithms presume, you know, pre baked cultural bias. And you know, aside from having, I think the key is having an open forum and having having a sort of ombudsman on budsperson or people. I think to Stu's point it's a big part of it is just building in, building in the recognition that, that there can be biases inherent in the data or inherent in the model. And I think you know, from the limited sort of poking I've done on, on chat GPT, they seem to have, they seem to be addressing this serious with, with, they seem to be addressing this non trivially and that's very encouraging.

Yeah. And I would actually add something that didn't click for me for a while but is core to OpenAI's philosophy is this idea of like actually getting this thing, getting early iterations of what could become AGI into people's hands Quickly and sooner in the development process will actually accelerate a lot of the safety efforts, because the moment, if you have a team of five brilliant PhDs focused on the safety or alignment problems, you'll certainly come up with a lot of solutions. But if you have 100 million people trying to get the thing to do something problematic and figuring out all the different failure modes, your feedback loop will just be much more rapid in discovering all the ways this thing can misfire. I don't know if you've been on Reddit at all. They're on their 12th iteration of this prompt called Dan, which is like, it's an acronym for something, but it gets ChatGPT to break all of its rules and, you know, come up, share how to build a nuclear weapon and whatnot. And like, I just truly don't think 5 PhDs in the room would have come up with the DAN prompt. I think, you know, a couple randos on Reddit who are trying to break the thing, come up with it, and now OpenAI has visibility and now they can work to solve it. So I think that's actually been a really. I don't know, that certainly was kind of counterintuitive to me, but it clicks for me now, like, actually get this thing out, have people break it, and let's rapidly fix what comes.

Yeah, it's the biggest red team in the world.

That's totally it.

And there are now dozens or hundreds or thousands or millions of business ideas, whether it's harnessing AI, whether it's protecting from the dark side of humanity. There are millions of businesses. I guess I ask my last question, Kirby. Where are you most excited? Kind of in the field of AI, whether it's content or medicine or contracts or comedy, where are you most excited?

Well, I think as an investor, I have certain areas that I kind of just go to because I'm experienced or have some prior knowledge or prior investment history in them. And so for me, I think I mentioned earlier, the, you know, the idea of value attribution is really compelling to me. You could think about it in the same context that you might have thought about advertising in 1996 when we were, you know, hard coding, you know, 5, 5K, you know, custom images of, you know, champion athletic logos on the main page of GoToNet. But you could see incredible, incredible value being created. I wonder about that value transfer. I think the, that opportunity space is going to create generational companies. So that's certainly a place I'm, I'm watching those, those, those solutions tend to need to Be pretty horizontal. Incumbents definitely have big advantages there, but as we've seen, that doesn't mean that new players can't come in and take that value. So I think that's one area that's really exciting. And then I think whether I'm excited about it or not, I always tell the story. I have a friend who's a partner at a Big 5 consulting firm and does a lot of, you know, people, people work with Fortune 1000 companies and this person said that for the last five years he's been on huge engagements with big companies to plan for what's happening now. So with regards to just the attrition at the line level for knowledge workers and I guess that could be exciting if you're in the business of creating targeted custom models and products built on AI that will take over that work. But this idea that all people are now going to be doing better, there's a better, higher use of their time that will be applied at those companies is just not true. And I think quietly, and I haven't heard anyone say this, but, but I think quietly, a lot of the layoffs that you might be seeing right now, this is a great time to be under the COVID of a macro recession, be reducing force, knowing that you may not have to rehire it based on, based on the rapid advance of AI. So again, not sure I'm excited for that culturally or socially, but there will be a lot of, a lot of opportunity for startups there as well.

Stu, where are you most excited about AI over the next few years?

Yeah, my mind drifts towards a lot of the kind of creative applications around content creation and just I think the Internet was a chance for everyone to have a voice, which obviously comes with a lot of good, bad and ugly, but is a thrilling and fundamental shift societally. I think the next phase, once you have a voice, once you have your Twitter account, very quickly it becomes super hard to just express interesting ideas, to actually contribute your knowledge to the collective pool of human knowledge. I think even today, despite all the perspectives being shared online, if I was going to go raise a seed round, I'd probably just hit up Kirby asked for his two cents before I googled something. So I think what's really exciting to me with this AI stuff is it's finally a chance to radically reduce the friction from the human expert to the well crafted artifact of knowledge that can be shared broadly. I think these language models specifically will help pull out quite a bit of insight, find the right words, find the way to flesh out the ideas that are currently sitting latent offline. So I think that's really exciting as a chance to help more people articulate their unique knowledge and to share that. So that that gets me fired up.

Kirby, Stu, thanks so much for joining me today. And, and let us all get back to our Valentine's Day human checklist and make sure, make sure we're doing what we need to do.

I wish, I wish there was an app for that.

That's right. Thank you, John. Appreciate it.

Good seeing you. Still.

Center Field delivers outcome based marketing solutions and customer experiences that convert to many of the world's largest brands in home services, insurance, e commerce, business to business and more. Simply stated, if you'd like to get more customers out of your digital marketing, we believe we're the best and we'd love to meet and show you what we can do. Check us out@centerfield.com p podcast and thanks for listening. You've been listening to the High EQ Marketer. Subscribe to us anywhere you listen to podcasts and if you enjoyed this episode, please drop us a like rate the show five stars. That's the High EQ thing to do. Thanks for listening.