GenAI and Digital Content Creation with Phil Libin

Hello and welcome to the latest edition of the Techron AI video series. I'm your host, Mike Bazar today with Phil Lipman, who is CEO of Mm-Hmm. Which, mm-Hmm.

Gonna explain how that came about in a minute. But we're gonna be talking about where are we on this fabulous AI journey. 'cause I think, um, a lot of folks have gone from irrational exuberance to a little bit of maybe a touch of the trth of disillusionment, but hey, we're just getting started here.

Philip, welcome the show. Thank you. Nice to be here.

So, what's your assessment of where we are in, in some ways, I think we went from this is gonna change the world to now we take it for granted already. So that was pretty quick. Sometimes we think about all these LMS have become commodities, and yet people are kinda still doing, I would say, pretty rudimentary stuff.

And, you know, the ROI might not be there just yet, but maybe there's more to come. What's your assessment? I think, uh, LLMs have been massively overhyped.

Like probably literally like nothing else has ever been overhyped before. Like I've never seen anything in my, you know, already many decades of working in tech. Uh, this is the, by far, the most ridiculous hype cycle that we've ever seen.

Uh, and that hype is obscuring that there is actually really cool stuff, you know, that, that that's happening. Uh, but it's kinda hard to see it because of, because of all the ridiculous hype, uh, both on the, on the positives and negatives, both on the, how it's going to, you know, solve physics and, and, and then do everything for us as well as on the negatives about it, you know, destroying humanity, both sides of that are just kind of ridiculous. Uh, but there is, there is cool stuff.

Uh, we're starting to see it. Uh, hopefully we are actually making some of it, but other companies are as well. Um, but yeah, it's tough.

It's tough. It's tough for the real stuff to be seen given all of the nonsense. Hmm.

And what is the real stuff in your opinion? 'cause we're starting to hear more about, well, agen ai people are talking about domain specific large language models, um, and other things where we're gonna have say, better reasoning capabilities. What's next?

Well, I think, uh, I think we have two things, right? We need to work on, um, actually getting AI to, to, to fulfill this promise to actually have any reasoning, which is, uh, hard to say how far we are from that. I think probably further than, than, than most people think.

Um, I don't think that LLMs are on the path to that. I think LLMs are a really elegant tool. Um, but they are, they're, they're really limited.

Uh, so you can, you can do great things with 'em if you're using it correctly, but I don't think we get us anywhere near a GI or a SI, uh, uh, or, or, or, or anything else. Uh, but I think we will get there eventually, but it's gonna require some, some new inventions that, that we haven't seen yet. At the same time, we should be designing real products using LLMs that are genuinely useful.

Uh, that passed the toothbrush test, which is, you know, I think it's Larry Page's, I think it was Larry Page, uh, that that first used that expression, you know, years and years and years ago. Toothbrush test is a, is a, technology passes it when you're using it multiple times a day without really, like, without thinking about it, without like, showing off like a toothbrush. Like, you know, use a toothbrush a few times a day, not because it's cool just because like, you just sort of do it, right.

It's useful. Uh, and I think we're just just starting to see the first of those in ai. I think the Apple intelligence stuff is a good example of like a big company doing things that I'm fairly optimistic about.

Uh, I think, you know, I've been playing around with, uh, with the developer betas of it and yeah, a lot of it passes the toothbrush test. It's just like stuff that is useful. It's like low stakes, high utility, use it all the time.

It's cool. I think it's great. Um, I think, you know how the use cases will, will, will follow as well.

And at the same time we should be developing the, the more sci-fi stuff, but we may be a long ways off from, from having that be real. So what's the backstory for a home and, uh, and what is your stake in this AI conversation? Well, hmm.

Is, is is an app for, uh, recording, presenting, uh, yourself, basically, it's a, it's a communication app. Uh, we started it during, uh, during Covid, right? Right in the kind of, in the midst of lockdown, just because all the video stuff just became so boring and dreadful and we're like, yeah, I think we can do, I think you can make it so that video is better than, than, than than what it was.

Uh, and we think that, you know, communicating standing out is super important. It's kind of a superpower. Successful people generally are good communicators and successful teams generally for communication culture, we think we can do a lot better.

So it's a, it's a video tool for, uh, explaining things. Uh, either live like, like you and I are on Live now or prerecorded or any of that kind of stuff. And the name is, uh, you know, is for explaining yourself and for communicating.

If you're doing a good job, you see a lot of people nodding and saying Mm-Hmm. Because they agree with you. So that's why it's called that least that's the, that's the revisionist version of the name that I'm gonna stick with.

The versions, The initial wave of Gen AI at least was aimed that, uh, text and code, and now people are talking about applying it to audio and video. What can be done, what's realistic, and um, you know, and what is worth doing. Very little, very little original stuff can be done.

That's good. Um, gen AI is a good way of automating work that didn't need to be done in the first place. It's a good way of making you not do things that you didn't wanna do, that you don't care about.

Um, so like, it makes it very, very easy to get to mediocre. You can go from nothing to like, something that's pretty mediocre really fast. Uh, and that's not super useful.

I think it's useful for some people some of the time, right? I think a lot of people find themselves in jobs or in positions where they wind up having to do like, b******t things that they don't really care about. Um, and I think Gene AI can help automate that.

I don't think that actually makes the world better. I think these like b******t things ought to just not be done, uh, rather than just having more of them done because we've just lowered the price of, of low quality stuff. Uh, but it's great for that, right?

It's great for, there's this book, aha. Uh, hold on, I think I have it somewhere. Oh yeah, here we go.

Okay. There's this book that came out several years ago, not about AI at all, uh, called B******t Jobs, this thing. Um, and uh, it's great.

It's, um, uh, it's what David Graber just, he just died, uh, sadly I think this year. Um, and, uh, this book is about how lots of industries and, um, you know, countries, industries, jobs that have these, these b******t jobs, which jobs that don't do anything when people are kind of in 'em and they're just there to like pass papers around or like boss each other around, but they don't actually produce anything. And we all kind of know these things.

And some industries it's like 90% of jobs are b******t jobs. And some industries it's only like 5%. It really depends.

Like there's very, there's relatively few blue collar b******t jobs. Like most plumbers, like, it's a real thing. They do something that's like real, and it like creates value, almost like builders, but a lot of the better paying kind of what used to be called white collar, you know, knowledge work, a lot of it doesn't do anything.

And there's like, this book is fascinating. It goes into like multiple types of, of b******t jobs and it's like, it's kinda cool. It's, it's a serious book.

It's like, it's got a kind of a funny title, but it's, it's actually like a serious, like, economic work. Um, and uh, this creates like, lots of problems. It creates lots of problems for the people that are in the jobs.

And, um, and ai, gen ai, the first wave of it was great at eliminating b******t jobs or automated, which is like kind of good, but kind of sad, right? So like, um, spam, like if your job is like writing cold call emails, you know, for the kind of trying to scam someone into opening, yeah, you could probably do that much faster with gen ai. The question is like, if that's your job, like, I dunno, like feel it's kind of bad and, and the world doesn't necessarily need more of that and a lot of like summarizing stuff, like summarizing emails that I don't wanna read.

Yeah, it does a great job of that. But then I don't even wanna read the summaries. I just don't wanna read 'em at all.

I just rather just structure my life so that I don't have to like, read AI summaries of stuff. Uh, and I don't for the most part because it's like, if I don't care if, if I don't care about what you wrote, why would I read it? Right?

I'd rather just like, not deal with you. Not everyone can do that. So sometimes it's useful.

Oh, so, so that kind stuff is like, it's just bad. Like the images it makes, they're bad, they're cool, right? But like, it got old real fast.

Like, I remember the first time I saw Dolly, I was like, wow. I was amazed. I was like blown away by the thing for like, you know, for a couple of weeks.

But now, like, you see it, you see an AI generated image and it like, it doesn't even register. It's just visual noise. It's just crap.

So like, it's a good way to make mediocre things quickly. And my general advice is if you find yourself in a position where you're making a lot of mediocre things as to like, I don't know, like I've tried to structure my life so I don't have to do that AI or, or, or, or not. It can also be used to help me make really good things, but that's a very different set of use cases that are, that are much later in coming.

All right. Well, there's a couple of things to unpack there, but the first is, a lot of those jobs you're describing are things that need to be done the same way over and over again. And a lot of the Gen AI models are probabilistic and rarely do the same thing the same way twice.

Yeah. So is there a misalignment here on, you know, these functions that we're trying to automate with N ai and versus the fact that they need to be, you know, the reason they're boring is 'cause they're the same way a hundred percent of the time? Yeah, I mean, I think like a lot of stuff, um, gen AI is effective at, um, that's scamming things, right?

Not, not only that, but it is very good at scamming at like low quality scams at low quality, like, uh, uh, you know, email campaigns, that kinda stuff. And like why? Well, because like, it, it can make things that are just different enough to be like barely believable and plausible.

And so you can, you can scam people at a bigger scale. I'm not saying that's why companies are making these things. I'm just saying that like what some, among the first uses that we've seen are these like low quality scams, uh, initially with text and then tweets and then replies, and then just like generating, you know, SEO farmed, you know, content.

Like what some huge percentage of the internet is currently AI generated b******t, right? Like, I saw something about this who's even wired or ARS technic a couple of months ago where it's like 60% of the internet is like AI generated crap or something. Uh, I think, I think we see a lot more of that in non-English stuff.

I think it's a lower percentage of English language. I think there's a lot of other languages where it's like literally the majority of like the text out there is just like AI vomited nonsense for the purposes of, you know, ad baiting and SEO optimization and kind of low quality scams like that. And there, the fact that it does something that's, that's non-deterministic is like, like that makes it more effective at it because yeah, you want things to be a little bit different, hard to detect, you know, that kind of stuff.

Um, it's, it's pretty good at like yeah, if you, if you wanna write, you know, um, if you want to write a book report about, you know, the Scarlet Letter for 10th grade and you're trying to kinda do a mediocre book report about, you know, ne the Nathaniel Hawthorne I think was the writer. I haven't read that book in a few decades. Uh, yeah, I'd probably do a decent job of putting together an average book report, but like, like, but why, like, you haven't learned anything, like why?

Um, so a lot of these things are yeah, are like, it's not the first wave of use cases is not living up to positive expectations. It's maybe living up to like negative expectations in terms of the tsunami of b******t that we're subjected to because of it, but it's not quite living up to like ROI and positive expectations other than in some, in some edge cases. But okay, you know, let's us, like people who design products, we need to start work.

We need to start making them so that, so that they are good and there are ways to use it, right? Like, like I said, like I think some of the Apple intelligence stuff uses LLMs in a really nice way. Um, and there there's a bunch of best practices and we've been working on stuff.

We've been, we've rolled out features in mm-Hmm. That use, that use LLMs in, I think ways that they're not trying to make it easier for you to do something. They're trying to make something you do better.

They're not trying to like save you time. They're trying to make the, your presentation or your video better. Like even if it takes you more time, it, like, it makes suggestions, it asks you questions.

There's like, there's ways to think about it that are beneficent, beneficent meaning like in, in, in the best interest of the user. Um, but that isn't the first wave. There was like, there was just a lot of, there's just a lot of garbage.

There's just a lot of b******t. And I don't think, I think that's made the world a little bit worse. Like, I think it, I think it's part of the, in ification of the internet, not all of it, but it's part of it.

So I think it's made like our, our general day-to-day life, slightly worse than it was before, but hopefully the benefits come a little bit later. Hopefully like the real positive use cases are just starting now and they're a bit more back loaded. So overall, I'm optimistic that I think AI will make, will make life better.

But the initial results are, yeah, just kind of mostly garbage with a little, with a few bright spots and hopefully the bright spots are getting more numerous over the next few months. What is the future of, say, video content production look like? And I'm asking the question because part of the hype cycle has been, well, there's gonna be all these people who either couldn't gain access to the studio or they didn't have the skills to write, but they had a great idea for a, a video or short film or whatever it is, and they're all gonna go create stuff using these AI tools.

And yet I've seen some studies that suggest that if we all use the same AI tools to come up with a a story, we're all gonna come up with the same boring story. Yeah. Or slightly different boring stories.

Um, look, um, It's not like Microsoft Word, like Microsoft Word now, made it possible for anyone to write the Great American novel or you know, the great worldwide piece of literature. Like Yeah, it does, it does. Like the fact that we all have things that we can type on and we no longer have to like hand write stuff or you use a typewriter, like yeah, it's definitely democratized writing.

Has it improved? Like, has it improved writing? Like, no.

Can anyone write an amazing novel? No. Everyone's got the same tools.

Can everyone write an amazing novel? No. Uh, is it faster to write an amazing novel than it used to be?

Probably not, right? Like real work, uh, real accomplishment. Like it still takes time.

Um, right? Like the difference between productivity, which is like how much you can get done in an hour and accomplishment, which is what is, what did you get done in your life is like very different and accomplishment takes, you know, it takes a while, it takes years often, and it could be slow. And the tools, like, it's good that more people have access to them.

Uh, and it's good for example, that like, okay, if I'm writing a novel, um, you know, Google probably helps me save time. I can probably do research, you know, better and fact checking. So yeah.

And we have some great novels, right? Probably more than we used to because of that. But it's, it's not like everyone that has access to a WordPress all of a sudden can write a great novel.

And the AI is like, it's, it's similar. It's a tool. Some people will use it to do really brilliant things and a lot of people will use it to make boring s**t.

And okay, like great. The video, the AI video stuff, um, I mean, look, it's not, it's a totally different art. It's a totally different media.

It's just like, it shouldn't even be classified as video. It's just a different kind of thing. Like, um, you know, this idea that like, oh, someone's got a brilliant idea, but they, but now, and now with the ai, they could just make a, a great film b******t.

No, they can't. And do they really have a great idea or is there a great idea that I'm just gonna make a great film by pushing a button and then make money on it? If, 'cause if that's a great idea, then yeah, that's not gonna work either.

That's, that's not a great idea. Um, and they use like, generated video as a tool in their video. Yeah, of course.

It's probably pretty good at like putting in some, you know, doing some backgrounds, doing some interesting things like as part of a narrative storytelling technique. Sure, sure. But so, so is, you know, so so is every video tool that's like come out, like it just adds to the repertoire of what you can do.

But I, in no way do I believe that it's gonna somehow like, automate greatness or like make it make more of it scale. Like that doesn't make any sense. And of course, the vast majority of the stuff we see right now with, with AI video, it's just, you know, it is funny for a while, right?

Because it's so like, grotesque and like weird, but at some point it stops feeling grotesque and weird and you're just like, oh yeah, okay, yeah, you ai you AIed up some video like, congratulations. So will it become harder to find the quality in a sea of mediocrity, or will the quality stand out more because there is so much more mediocrity? I think, I think that like when we launched this new AI features in Mm-Hmm.

Uh, one of the features we launched is called Questions. Um, the idea there is that like the AI asks you questions, it doesn't give you answers, it asks you questions. Like if you wanna talk about something, you're like, I have something I wanna talk about.

And it asks you, it's like, okay, what do you wanna talk about? And you answer and then it like asks you follow up questions to try to like tease a good story out of you. Like the answer is already in your head.

It helps you bring them out. I think applied that way. Is it like, change the world?

Like no, it's not a world changing feature, but it helps you make better quality content and it, like, it, it helps you to do it. So in that sense, like better stuff can come out if you want to use it. And you, and your intention is to make higher quality stuff.

I generally don't think that it's harder to find greatness. Um, I think the internet is, is has done a good job of like making somebody make something excellent. Like there's all sorts of ways that it surfaces that it, that it, that, that you find it.

And, and typically we've, that, that trend is resisted in ification, like in spam. Like however it is that like mediocre things trying to spam of like other ways of finding something great tend to bubble up. I believe that'll continue.

I'm optimistic that like AI will not be the end of like great stuff and it won't necessarily make 'em harder to find. I do think that we're gonna, we are living through a tsunami of low quality stuff. Like we're in it already far more than it used to be.

Like, look, um, I, I live in, I live in Arkansas, Bentonville, Arkansas, uh, it's great here. Uh, and um, we had the, we had the, uh, eclipse a couple of months ago. We had the eclipse come through, uh, and I was like, oh, cool, we're gonna have, you know, almost full eclipse like in my backyard, I'm gonna sit there with my dog and watch it.

And I thought, oh, we should get, I should get some eclipse classes, right? Because everyone was like, oh yeah, you need eclipse glasses, don't go blind. So I went on Amazon, I'm like, eclipse glasses, can't buy eclipse glasses on Amazon because there's like 10,000 spammy, uh, entries.

There's like obviously fake eclipse glasses, NASA approved, you know, from like weird company names. They're just like a combination of like consonance for like 38 cents for like a dozen, right? Like literally thousands of listings.

And I was like, Jesus. Like what, what? Like what, first of all, why is anyone trying to like sell me scam eclipse classes?

Like how much money can you really make by like making a few people go blind? Like if you're selling 'em for like 39 cents anyway, like, is it really worth it? Like it kind of shocks me that there's like somebody somewhere that's like, you know, what I'm gonna do is I'm gonna list a whole bunch of fake lips, glasses on Amazon like that that's just, that's bad.

Like says something about people that's, that I would like, but it's also like, I buy so much stuff from Amazon and I couldn't, I couldn't buy lips glasses. I'm like, what? How did, how did Amazon let this happen?

Obviously three years ago if I did a search for Eclipse glasses of Amazon, I would've, I would've, you know, clicked buy now and it would've shown up. So like, this is a new problem. This is an AI generator problem.

I've wound up getting at Walmart. You know, Bentonville is like the home of Walmart, so it's easy. So I wound up getting 'em at Walmart because at Walmart, like they didn't have 10,000 listings, they just had like five.

And so I just trusted 'em more and I'm like, that's weird, right? Why do I trust Walmart for this kind of stuff more than Amazon? I used to trust Amazon all the time, but, and I still do, except like in this case, they've somehow enabled the, the b******t wave to like overwhelm them.

So yeah, I think we'll see that, but I think we'll see, we'll also see great stuff and there are genuinely amazing use cases that I think can be built just like they are far between and they are, they're like islands. They're like beautiful little islands in a sea of b******t. Um, and you know, we have to find those islands.

All right folks. Well, you heard it here. The whole point of AI is not to do stuff for us.

It's to help us to do much better stuff, maybe a little bit faster, but at the end of the day, quality rules. Hey Phil, thanks for being on the show. Thank you.

All right. And thank you all for watching the latest episode of the Text on AI series. You can find this episode and others on our website.

We invite you to check them all out. Until then, we'll see you next time. Bye.