ChatGPT. Deep fakes. Avatars. Google's Bard. In case you haven't noticed, artificial intelligence is changing the game. That's why we have Ryan Steelberg, Veritone's CEO and President, back on the show to check-in on the state of AI and how it impacts employment, as well as everything from communications to advertising and marketing. Veritone, who recently acquired PandoLogic, as well as Pando's Wade & Wendy, a conversational AI solutions, also powers Chad & Cheese in French, Spanish, German and Portuguese. ¡Bueno!
TRANSCRIPTION SPONSORED BY: Disability Solutions partners with our clients to build best-in-class inclusion programs and reach qualified, talented individuals with disabilities of every skill, education, and experience level.
Intro: Hide your kids. Lock the doors. You're listening to HR's most dangerous podcast. Chad Sowash and Joel Cheesman are here to punch the recruiting industry right where it hurts, complete with breaking news, brash opinion, and loads of snark. Buckle up, boys and girls, it's time for the Chad and Cheese podcast.
Joel: Oh yeah. It's your favorite Degenerates, aka the Chad and Cheese podcast. I'm your co-host, Joel Cheeseman. Joined as always, the Jennifer to my Ben Chad is in the house, and we are happy to welcome back [laughter] Ryan Steelberg to the show, everybody.
Ryan Steelberg: Thanks guys.
Joel: CEO and president...
Chad: There he is.
Joel: At Veritone.
Ryan Steelberg: Am I number 1000 who's been on this show twice, or where do, where do I rank? I mean how do I come back, like, and break SNL type of records here?
[laughter]
Chad: We haven't had that many repeat visitors, to be quite frank.
Joel: I've been on Xanax trying to manage my depression of being on a thousand shows, so I don't really remember exactly where we are at this point. Sorry.
Chad: Just, just to be fair, he's been taking Xanax pretty much as long as Xanax has been around. So, yeah. Today, Ryan, we brought you back because, well, first and foremost, we gotta thank you for this. Play it Cheeseman.
[foreign language]
Joel: The world thanks you for this.
Ryan Steelberg: Can't make German sound good.
Chad: This is all your fault.
Speaker 1: In Portuguese.
[foreign language]
Joel: Oh God. [laughter]
Chad: This is all your fault, Ryan. We are now in five languages. People can barely understand us in English.
Joel: So scary. So scary.
Chad: Thanks to Veritone.
Ryan Steelberg: Offensive in multiple languages. [laughter]
Chad: So that's one application we'll talk about. We'll talk about practical applications later, but John McCarthy coined the term artificial intelligence, AI, back in the '50s. But the promise and the practical application of AI really hasn't been demonstrated. I don't think really been incredibly powerful until here recently. I mean, we did see IBM with Deep Blue beat, you know, chess Grand Master, Gary Kasparov in '97, then Watson beat two jeopardy champs in 2011.
Joel: How dare you not mention Chat GPT in the first three minutes of this podcast.
Chad: We're getting there. We are getting there.
Ryan Steelberg: Yeah. That's like ubiquitous now, right? I mean, that's, like I say, throwing a pebble in the Pacific Ocean, you're gonna hit somebody that's doing something with Chat GPT.
Joel: Are you surprised by that?
Ryan Steelberg: Yeah well, the speed of how it came out is incredible. And what you're kind of seeing is, as an ex Googler, they've had a lot of this tech for years, right? So, you know, Lambda is their large language model platform.
Joel: They just didn't learn how to demo it apparently.
Ryan Steelberg: Well, well, well, they have another problem. They gotta make money, right? And so, so they, you know, so they, when they're sitting on it, this reminds me when the world transitioned from primarily desktop, right? Usage on the web to mobile. Google and these ad titans are like, oh my God, all of our ad screen, real space just disappeared. And how are we gonna make up the difference of billions of dollars if I can't serve, you know, display banner ads everywhere. When you go from, and we're talking about obviously conversational AI and Chat GPT here is when you go from a search result page, right? Where it's very structured and Google has perfected ironically, using AI to observe which ads to you when you're getting a search results, trying to yield, get as much ad dollars from a conversational search result is challenging.
Ryan Steelberg: Candidly, it's like doing, you guys doing a native integration on one of your podcasts, right? In terms of like a sponsorship. So I think that more than anything is Google has had the capabilities, mind you, arguably with better technology than Chat GPT and open AI with better ironically guardrails. But they can't release something that's potentially gonna undercut their business model. That's why their stock dropped a hundred billion as you guys saw. And it wasn't, it wasn't that they answered a question wrong, it was, oh my lord, [laughter], where the hell are they gonna make money? [laughter]
Chad: Yes, exactly. Okay, so back to the question. Back in those days, that was kind of, that was the novelty, you know, oh, look, Jeopardy champions are getting beaten, Gary Kasparov, blah, blah, blah. What's the huge difference between the AI of then other than, I mean, obviously processors versus today, and it seems like the promise, I guess you can say, "The promise of AI," is starting to actually come to fruition.
Ryan Steelberg: It's the beautiful mind in the past that always had many of the answers, but they couldn't communicate them effectively, like humans like to communicate. So, you know, the calculator, right? I mean, again, we've had some power in our pockets now for years, right? And a lot of crazy similarities of what when Chat GPT and large language models were rolled out, schools are freaking out. Like, oh my God, it's banned. Can't use Chat GPT for school use. [laughter] As we all know, calculators, right? You know, depending on what class you're in, you're finally in advance. I guess they think you're smart enough that you can use, right? In an advanced geometric or an advanced calculator in the future. That's where we're, so I think where we're seeing here is you, just a slow progression.
Ryan Steelberg: And if we, we think AI technology is slowly trying to replicate and augment the human mind, right? And the number of synapse points, et cetera, et cetera. You're just, all these things absolutely are AI. They're just now getting smarter. They're getting more powerful. And it's kind of like that, again...
Joel: More dangerous.
Ryan Steelberg: Or dangerous. It's, but I, when I'm saying like that, it's the brilliant kid that can't speak, right? You have to, you have to ask him a specific question, and if you ask him in the right format he'll give you a brilliant answer. Now what you're seeing, because they have so much training data, and they now have the successful technology at scale to put conversation around it. Now, when I ask a question and saying, "Hey, can you make me a compilation song or write me a song that blends the style of Eminem and Katie Perry together, and I want it to be very explicit, right?
Ryan Steelberg: What I just did that prompt, that prompt that you could have gotten a search result from a multiple different sources, but now they can present it to you. Literally, you have to wait. You only have to wait a couple seconds and something comes out at you that is so good that David Guetta is putting it on his stage these days. [laughter] Yeah. So I think that's the difference, is the, the brain power has been there as we're moving, I'd say higher functioning cognitive processing is there. But now the presentation, the ability to communicate very human-like in terms of conversational format in speed, so right for immediate gratification is what everybody is seeing as a quantum leap now in AI.
Chad: But that's the thing though. It's the immediate gratification. We've had so many awesome technologies that have been behind the wall and nobody's had access to them. So what really Chat GPT has done is it has exposed the power of technology that many of these companies already have, and they've had for years, but they've finally, they finally let out the, the genie out of the bottle. So the big question is, how many other companies other than Bard, let's say for instance, they start to publicly display just how much power they have behind the wall.
Ryan Steelberg: You know, this is really gonna show the big boys digital divide, right? How expensive it is to run and train these large language models. You're not gonna see a ton of large language models come out. You're gonna see hundreds and millions of, I'll call domain specific language models, right? So, I'd say it's the world we live in in chatbots today, when you go and have a horrible experience trying to find out what your bank account balance is, right? Through a chat. That's a very specific, domain specific language model. Building and training a large language model is incredibly expensive. And again, for it to be really effective, there's only gonna be a handful of large enterprises, ironically, those who dominate in cloud and AI historically like the big fangs, right?
Ryan Steelberg: The Facebooks, the Alphabets, the Microsofts, it's the big boys. NVIDIA has the chips, but NVIDIA's, you know, but NVIDIA's not running the cloud-based architecture as well? So I don't think you're gonna see millions of companies, you're gonna see the blending of this, right? Meaning because they're being, because of what Chat GPT and Open AI did is open it up to your point, it's almost gonna force the other big boys and players to open up their kimonos.
Chad: Yes.
Ryan Steelberg: And which thank God, because now as companies and others, let them go fight it out, [laughter] and now we get to actually use these things for whatever crazy purposes we want to use them, and it's not gonna be locked up and just kept behind these walled gardens.
Joel: Does this get commoditized? I've heard that recently that there's a risk that this whole thing just becomes ubiquitous and free.
Ryan Steelberg: Which would be great, [laughter], because I, it's the blend of the large language models with the domain specific models. It's the orchestration of those two things, which really makes it powerful. So for example, if I'm having an HR discussion with a fully programmatic AI experience with, and I'm trying to evaluate somebody who wants to join our company, you can do it right now, we have it, Wade and Wendy, right? Panda Logic Company we own does that very effectively. But in the context of the conversation, if you wanna get to know the person and ask 'em, Hey, tell me about, you know, what's the recent sporting events? What kind of sports are you interested in your local market, right? The bot just sits there and doesn't know what to do.
Ryan Steelberg: Now you can orchestrate a large language model, right? In coordination with these hyper trained domain specific models, and you can have a real conversation. You can really learn a lot more about Chad and Joel about what makes them tick, what they like and they don't like in context of the real world. So I think it's gonna be the blending of the commodity. And again I don't wanna minimize the power, the brilliance of what these guys are putting in the market, but I do think that everybody's gonna be able to tap into the large language models. It's gonna be how you use it and how you differentiate it to make it work for your business.
Chad: So what you're saying is we have companies that are out there right now that are domain specific, and they are tight on what they know, the application process, the interview process, et cetera, et cetera, et cetera. Right? They know their shit and they've got tons of data, decades of data, right? But then could you play that almost in series with a large data model so that it would like, it would automatically know, oh, wait a minute, this isn't a part of the specific domain, so I have to go out to large domain, is that what you're saying? So that the actual chatbots can have different training models that they would actually be working off of?
Ryan Steelberg: Exactly. And they're calling it the... You know, see some terms about chained language models in orchestrated language models, and that, and they're doing that now in context, in the same conversation. That's what's so dynamic now, right? And that, of course, you can choose to go one path Chat GPT, right? Public large language model, or you can go domain specific like HR Panel. Now you're gonna blend it in the same conversation, which by the way, is like having a conversation with a human, right? So if you're talking to your surgeon, right, your surgeon can go very, very deep domain specific, let's say about spinal surgery. But that same surgeon can, you can have a conversation depending on, well how charismatic he is, and you could break the conversation but, you know, and have a conversation about sports, right? So again, I think you're gonna, it's, that's where I think the differentiating value is gonna be for a lot of companies and people who are applying these technologies, is how do you blend these different AI solutions in a congruent, seamless experience that is gonna be the magic.
Joel: Pivoting to sort of our space a little bit. Ryan, we've heard about scams on job seekers for a long time. From my perspective, this technology and deep fakes and sounding like the CEO, right? As a robot, where does this thing go in terms of scams on job seekers? How dark is it gonna get?
Ryan Steelberg: I mean, it's gonna be messy, right? I think the, it's gonna, there's gonna be a lot of time and inefficiencies trying to, both sides of the equation to figure out, okay, geez, I feel like I'm being recruited I'm gonna spend time. Am I supposed to actually show up for a live video interview, or am I gonna show up and find out that there's nobody there? Or inversely you're seeing a lot of now bots applying for jobs, right? They're just flooding the systems on the other side and they're, and they're actually able to answer questions. So I think it's gonna take time for all of us. I mean, we're in this business for us to try to make it more efficient. You know, the... And so I think in immediate short term is you're gonna have to see, it's gonna help create, it's gonna help expedite a lot of the process.
Ryan Steelberg: But I think you're gonna see a bottleneck where actually human to human interaction, at some point, it's gonna have to become back into context here just to validate the truth behind the situation. 'Cause again, some CEOs are open, they're not trying to hide it. They want use their fake voice or their synthetic digital twin to... Because they feel that's part of the experience if you're applying for Tesla and they wanna use their voice versus other groups who are using it as a spoof. So again I don't think there's a perfect solution to verify the integrity of the human in the short term, and that's gonna work itself out over time.
Joel: So you mentioned in one of your interviews, like, if someone asks you for money, like that should be a red flag, right? That that's something that everyone should look out for. In our business it's, if they ask you for a social security number, if they ask you for, you know, intimate information or bank account information, what should job seekers be looking for to keep them safe?
Ryan Steelberg: Yeah, I mean, any PII information to that level is a big no-no, you know, personal identifiable information, your address. I mean, any, you know, forget just spoofing, I mean just self-preservation and protecting yourself is imperative. The challenge for us is, people who are older are more skeptical, right? It's the generation below us that's terrifying. I mean, they've been willing to compromise their personal information to get a little bit higher scale in this, in their snap social circles. And that's what's really scary is again, the younger generation, again definitively have actively, and in many cases are aware that they're giving up their personal information to experience a social dynamic. So I think where there... And that's one good news, the good news about what you just said is we can identify a lot of those things.
Ryan Steelberg: It's, it's, it's great. It's like creating an index of certain things that should not be disclosed. There's redaction based software that's been around for a long time. That also is getting very good and sophisticated. So our ability to identify what is a phone number, what is a social, potential social security number. We can be overly conservative on a lot of these things. Are there gonna be gaps and holes and loopholes? Of course. But again our big thing is, somebody's really interested, very rarely are they gonna ask you for any personal identifiable information outside of stuff that, you know, just common sense of, you know, I mean, are they gonna ask you what your GPA was at UCLA? Who knows? But again, I think rule of thumb is don't share any personal information they really want. If they're, if you're gonna get to that next level, somebody will call you back and they'll re-engage you.
Chad: So tell us about some of the biggest challenges Veritone is solving for customers today with AI, with these voice models with chatbots. What are you doing to help companies and why? Why is it a problem?
Ryan Steelberg: Well, it start, well, we talked earlier about like how AI's progressed. So let's talk about, I'll put it in context of one big media customer, so CBS News. So CBS News is a big customer of ours, and what we do, it starts with us ingesting all of their content. So they have tons and tons of programming, they have their legacy archive, the Apollo moon landing footage, right? 'Cause that goes back decades. So number one is we help ingest and organize all their content, and we use AI models. And so instead of humans having to look through and listen to all that content, we have AI models that sift through it and identify when Walter Cronkite's speaking, right? When there's a spaceship on, you know, and the lunar module capsules on screen, the AI does that.
Ryan Steelberg: So it's organizing it. What you're seeing now is that when they say, okay, geez, can we remaster something, like this footage is cool but we can't really hear Walter Cronkite's voice. Can we remaster his voice? But we wanna keep the authenticity of the grainy video. But again we can't hear half the things he's saying. Now, you know, again, with approval and permissions you can do that. So you can start to now go from not just indexing. And I'd say like, that's kind of like the version A one of cognitive services that now it's extension. It's, we can actually create new derivatives of the content, synthetic voice is one of them, obviously we can clone Walter Cronkites's voice and we can, and we can turn that into a potentially with approval, original new programming, a whole new series, right?
Ryan Steelberg: Narrated by Walter Cronkite if that's the goal. So I think what you're seeing now is companies, by the way, it's a daunting task. You know, Moore's Law, the joke is Moore's Law, which was obviously the ever increasing speed of technology innovation is obsolete. It's growing so much faster than the Moore's Law is predicted. Well, companies are sitting around like, wait a minute. I thought guys were trying to figure out like the Metaverse and NFTs, like what happened? Like that was, that was last month, right? And now we have Chat GPT. So for us, with our customers, we're, we are now almost like a wilderness guide. Like, with all these new technologies, it's impossible for a lot of our customers to keep up on all these different things and that's really where we are.
Ryan Steelberg: There's so many different AI models coming onto it. And our core offering is called AI Wear, right? Which is our sort of baseline operating system. So we look at a large language model, just like any other model, right? If you, instead of running for face detection, and if you want to incorporate a chat AI solution that includes a domain-specific model and a large language model it's a common infrastructure stack. And that's the big thing is for our customers, whether it's an iHeart or a CBS News, if they're looking to do something new with all these new technologies come in the market, we're the well one stop shop. And that's, and so we're like the trusted AI partner, tech partner for a lot of these companies. And some of these ideas they're coming out with are crazy. You know, they're really original new ideas and interactive cartoons that are gonna take it to a whole new level. I mean, what we're doing with Cameo and a lot of the, bring interactivity to animated characters, I mean, you're gonna start to see some really next generation stuff here in months, not years.
Chad: Wow.
Joel: Quick example of what you're doing with Cameo, and some people may not know what cameo is.
Ryan Steelberg: Cameo.com, they're, they've been around for a while. They're, they have like 55,000 celebrities and brands on there that if you want to have Chad and Joel like, call you and sing you happy birthday, you can go to Cameo and you can pay to have usually somebody famous, right. Do a custom call out to you. And not that you guys are not superstars, but [laughter], but to get Tom Hanks to do it, you know, they don't necessarily want to voice everything. So we partner with Cameo and we started initially with children characters like the big brands, Blippy and, yeah.
Joel: Blippy.
Ryan Steelberg: Blippy and, geez, JJ and Moon Bug, that whole phenomena, Coco Melon, that whole phenomena that now we launched this right before Christmas. That kid, that parents can come in and have a personalized Graham from these characters call their, you know, speak to their kids, sing 'em happy birthday, wish them Merry Christmas, happy Hanukkah, whatnot. And that's kind of took off. And now kinda what we're teasing about here a little bit on the show is making those fully interactive experiences. So it's not just one way, it's these children are gonna be able to interact back with these characters in real time.
Chad: So Veritone owns Wade and Wendy, a chat bot for our industry. So will Wade and Wendy have their own voices sometime soon?
Ryan Steelberg: Yes, they will. [laughter] Wade and Wendy, we've kind of gone away from the branding of Wade and Wendy. It's just it's more of I think the... It's made available in the commercial sense now through Pando Select. But yes, I think you're gonna have personas, right, that companies are gonna use, and we're in the middle of making brand new personas for companies today. That's gonna be their ambassador, right? And it's not just a clone of the CEO's voice. They're coming up with a unique new digital twin, digital version.
Joel: I think they're getting away from Wade and Wendy, because when I go to Wendy's for that double stack, I want actual Wendy's voice in that drive-through. [laughter] That's, no, you're gonna sell that to Wendy's, aren't you? I know what's going on.
Ryan Steelberg: Or Ben Affleck, right? He'll serve your Dunking Donuts, I guess. [laughter] That was pretty awesome in the Super Bowl. [laughter]
Chad: Yes, it was.
Joel: Chad and I will live forever. I love that. I love that.
Chad: Ah, yes, yes, yes, yes. Which, yeah, which means, you know since our voices are already cloned everybody out there in HR land, they're available.
Ryan Steelberg: You know what? I'll leave you this morbid to point. So right now, today, it's gonna change fast. The laws, because of what we're talking about, digital twins, copyright and IP laws are changing, but for you to protect your voice and your name and likeness, you need to die in California.
Joel: That escalated quickly.
[laughter]
Chad: What?
Ryan Steelberg: Yeah.
[laughter]
Joel: That is dark.
Ryan Steelberg: It's dark, right? And then not only do you, and so after you die in California, do you then have to register or have your estate or power of attorney register those attributes, right? So it becomes protected by statutory law and not subject to argument in common law. And in California, California is obviously because of Hollywood, and this the, yes, let's say it is way ahead of the curve on that. But just the joke is that, again, if you're getting near the end capture it, make sure you create your digital twin. And you gotta make sure you get your death certificate in California so you can get your [laughter], posthumous, like protection of your IP.
Joel: I'm sure Progressive state, like Indiana's right behind California. No problem. No problem. [laughter]
Chad: Ryan Steelberg, everybody. Ryan, thanks so much. Thanks so much for coming back on the show. If somebody wants to find out more about Veritone or Panda Logic or the myriad of companies you guys actually own, where would you send them?
Ryan Steelberg: Just go to veritone.com. I mean, it's the best way, you know, it's our corporate site. It can navigate you everywhere and what you're gonna see over the next month is HR and HR Tech is a, is a huge focus of ours. You're gonna see us investing even in more in that market.
Chad: Say more. Say more.
Ryan Steelberg: Yeah. It's one of our most ubiquitous markets. We have, you mentioned, we do a lot of different things that it's gonna be a tight, crazy labor market here for the foreseeable future. And the last time I checked every single business is not gonna be completely automated with AI here shortly. And everybody's looking to hire employees at some point over the course of their... And so it's one of our most ubiquitous offerings and Panda's just the start. And we're really excited about investing more into that space.
Chad: Excellent.
Joel: And we are excited to talk about it, whether it's really us or some robots. And Chad, another one is in the books. We out.
Chad: We out.
Ryan Steelberg: Thank you guys.
Speaker 5: Wow. Look at you. You made it through an entire episode of the Chad and Cheese podcast. Or maybe you cheated and fast forwarded to the end. Either way, there's no doubt you wish you had that time back. Valuable time you could have used to buy a nutritious meal at Taco Bell. Enjoy a pour of your favorite whiskey. Or just watch big booty Latinas and bug fights on TikTok. No, you hung out with these two chuckle heads instead. Now, go take a shower and wash off all the guilt, but save some soap because you'll be back. Like an awful train wreck, you can't look away. And like Chad's favorite Western, you can't quit them either. We out.