Latest / Newcomer Pod / Inside Anthropic’s Plan to Fix AI’s BIGGEST Problem
Transcript
- 0:00$100 million funding announcements, the shadow of the
- 0:03looming AI bubble and debate surrounding agentic AI this
- 0:06week. Cerebral Valley AI Summit just
- 0:08concluded, which for the uninitiated is Newcomers
- 0:11flagship event hosted by myself, Eric Newcomer and Volley
- 0:15cofounders Max Child and James Wilstrom that brings together
- 0:17the top founders and investors and features cutting edge
- 0:21discussions with the biggest names in AI.
- 0:23In this episode of the podcast, we're revisiting two of those
- 0:26important discussions, starting with the conversation I had with
- 0:30Anthropic's Chief Product Officer, Mike Krieger.
- 0:32During the conversation, we discussed the evolution of AI
- 0:35product design with an emphasis on the need for true seeking
- 0:38models. Mike also shared his thoughts on
- 0:41what Anthropic can do to combat AI slop and how they plan to
- 0:45focus on growth opportunities and life sciences and
- 0:47specialized verticals. This is the new government
- 0:50podcast. All right, lean in everybody
- 1:00excited about this one. Mike, thanks so much for joining
- 1:03me. Great to be here.
- 1:04You know, Co founder of Instagram to chief product
- 1:08officer at anthropic. I wanted to start with almost
- 1:11like a philosophical question that spans those two companies.
- 1:15You know, we talk about AI now, but obviously social media
- 1:19companies with feeds were using machine learning and systems to
- 1:23surface content. Then that era seems like it was
- 1:27about engagement. And it was like, OK, we're going
- 1:30to blame or the humans will be responsible for the truth value
- 1:34of what they have to say. And we're going to see what
- 1:36content people are interested in.
- 1:38And those humans can say what they want to say.
- 1:41You know, as a journalist and someone who's interested in like
- 1:44the truth and chasing the truth. One thing I've liked about
- 1:47models, despite like all the like, oh, they hallucinate or
- 1:50whatever is the aspiration is like we are judged based on how
- 1:53accurate our responses are. You compete on leaderboards that
- 1:56are saying, how often are you getting things right?
- 1:58You want to pass, you know, math Olympiad type tests.
- 2:02Do you do you accept that framework?
- 2:03And how much do you sort of in this role see anthropic as this
- 2:08sort of like truth seeking organization or will engagement
- 2:12seep back in? That's a really good question.
- 2:15Hi everybody, good to be here. I started.
- 2:17With the head, yeah. I think there's a bunch of
- 2:19different directions to take this.
- 2:20I'll try to be succinct. I think there are places where
- 2:24we're training the models as an industry to be good
- 2:26conversationalists. And sometimes that actually
- 2:29looks like continuing the conversation.
- 2:30I got to reach out from somebody that was like, hey, are you guys
- 2:32trying to optimize for engagement?
- 2:33Because Claude will often ask me a follow up question like, well,
- 2:36did you want to talk about this? And the funny part is like, not
- 2:39at all. And like time spent is like, I
- 2:40can tell you like not on any of the dashboards that I ever look
- 2:43at. It's just not a like a main
- 2:44consideration, but just training cloud to keep the conversation
- 2:48going, being a good. So we might actually need some
- 2:51interesting sort of counter metrics to what you know, well,
- 2:54can you get the same conversation done or
- 2:57accomplished less of a conversation.
- 2:59I do think there are some really interesting sort of other
- 3:02phenomena happening in the industry.
- 3:04So one of the things that we try hard not to optimize for like we
- 3:08like, I just don't think it's like the right incentive, but
- 3:10it's is like these like convert like chatbot leaderboards like
- 3:14Ella Marine and all these places.
- 3:15They're useful sort of yardsticks of how we're doing,
- 3:17but they're not like the thing that you should optimize for.
- 3:19But we have found like if you like, look at what ends up doing
- 3:22well on there. It's like verbosity, like being
- 3:25like more long winded can actually be praised by like the
- 3:28Raiders on the. That's like taking the exam,
- 3:30yeah. Every note you remember, like,
- 3:32unloaded into it. Yeah.
- 3:33And so, and it is interesting like what we I think the evals
- 3:36are important, but whenever we sort of like have these public
- 3:39yardsticks, it can tend towards the like yap or more engagement
- 3:43sort of thing. So I think that's one vertical.
- 3:46The other thing that I think about though is, you know,
- 3:48primarily what we're doing is building AI for businesses,
- 3:50right? And so we have like cloud for
- 3:52enterprise and in some like, you know, it's not like we would do
- 3:55like super engagement baity type things.
- 3:58But you know, a year later after a contract get signed, people
- 4:01are going to look and say like, did people use our AI or not?
- 4:03And part of that is just that it solved the problem.
- 4:07Was it truth seeking that it like do the right things?
- 4:09But then there's also like, was it a product I liked using?
- 4:11And so I think we're all navigating this question of
- 4:14like, you know, how do we train the models?
- 4:17How do we design the products around the models?
- 4:19And like, how do they get delivered in a way that is
- 4:21maximally useful rather than falling into this like
- 4:24engagement for engagements? What do you make of this word
- 4:27slop? Like I feel like that if you had
- 4:30to sum up the criticism of AI from, I don't know, the
- 4:33skeptics, like that's a word. It all often comes to mind.
- 4:38Like what? What do you make of slop?
- 4:39And what's what's to be learned from that sort of accusation?
- 4:42Yeah, I've like told our product team like one of our goals when
- 4:45we when we do things like build like PowerPoint decks and Excel
- 4:49files and and Word documents is to be the anti slop.
- 4:51And I think like the way I think about it, it's like hard to put
- 4:54that on an eval, right? Like 80% slop or 7%.
- 4:57Slop. I think it's it, it looks like a
- 5:00couple of things, like 1 is, does it look super low effort?
- 5:03Like did it just like does it not reflect any sort of critical
- 5:06thinking that the human did alongside the AI?
- 5:08And you can often tell you're like, was this thought through?
- 5:11Like this wasn't really edited? Like is this 2000 words when 200
- 5:14would have done if it had actually been edited down.
- 5:17So I think there's that strong component.
- 5:18And I went to a talk by Ted Chang, who's a science fiction
- 5:21writer who wrote the short story that became a rival.
- 5:23He's one of my favorite writers. And we were having this
- 5:26conversation about AI and, and can AI be creative?
- 5:28And you made the point of creativity is the product of a
- 5:31lot of decisions, right? And so that's true for novels.
- 5:34Like like what does slop look like for a novel?
- 5:35It's like when you're like, well, like this just seems like
- 5:38the, the, the kind of base level output.
- 5:39If you're just like, tell me a story, you know, but I think
- 5:43that that applies just to non fiction as well, right?
- 5:46Like if the document that you put in like even product
- 5:49reviews, if somebody comes in, this happens rarely, thankfully
- 5:51anatropic, but like with the like product requirements doc
- 5:54that looks like it was just the first output from just like
- 5:58write me a PRD for this feature. Like this is just slop, right?
- 6:01So I think that's the content piece.
- 6:03Then there's the design and the quality piece.
- 6:05So which maybe is like a a version of what that other one
- 6:09is, but it's more visual. Right.
- 6:11It is time like the solution, like if I think about writing as
- 6:14a human, I go back and revision is often where you find great
- 6:19writing. Is that the case with AI where
- 6:21it's like OK if you give it you have more time the the answers
- 6:25will be more tight? Or how do you see the
- 6:27relationship between time and the quality of the answer?
- 6:30I think it's levels of engagement and sort of how many
- 6:33iterations that you've gone through and how many.
- 6:36And actually you could imagine a here's a very well constructed
- 6:40prop where I have already pre made a bunch of the decisions I
- 6:43had AI mean you can like see what Claude is thinking.
- 6:46You know, you can expand the little like I rarely do it
- 6:48because I just it's doing this thinking.
- 6:49I mostly want the answer, but I expanded it yesterday and it I
- 6:53had written like a pretty long prompt and and it's thinking it
- 6:56was like, you know, Mike has already like done a lot of
- 6:58thinking here. So I'm not going to ask him any
- 7:00follow up questions. I'm just going to give him like,
- 7:02yes, thank you. That was actually that was my
- 7:03intention here. But I do think it is that like
- 7:06how much of your independent thought even that's where
- 7:09actually now tying back to the first question, sometimes the
- 7:11model should ask a question like, Hey, this is a really open
- 7:14wide option space. Like, can we like start
- 7:17narrowing down? And I'm going to engage with you
- 7:19on it now. I think we need a lot better UIS
- 7:21for that. Like just here's a question that
- 7:22you don't have to go and type it into.
- 7:24It feels kind of annoying. But in navigating that option
- 7:27space, you should be able to hopefully come up with something
- 7:29that's like complemented by AI and accelerated by it, but still
- 7:32has your thinking at the core. Just to get to that, I mean your
- 7:36product guy, this is the is AI fundamentally the chatbot era?
- 7:42Like do you think Pex with the machine, that is the main way
- 7:47we're going to use AI, use Anthropic in five years or
- 7:50that's a bridge to that's how we figured it out in the beginning
- 7:53and now we need to build products.
- 7:55So two kind of ways to tackle that.
- 7:58One is like like the classic meme of like, you know, you're
- 8:01like you're a naive view and then you're like like super like
- 8:04Galaxy brain view and then back to the original view is like how
- 8:06I have felt about this exact question.
- 8:08So when I join Anthropic, I was like, if we are still talking to
- 8:11AI with chat boxes a year from now, like I've failed in my job,
- 8:15really. Yeah.
- 8:16I was like, it was very adamant that there was like something
- 8:18wrong to the kind of dominant UI paradigm that we had settled on.
- 8:22And it felt like exposed the lack of creativity.
- 8:26And then we did a bunch of explorations around like, how do
- 8:29we create more structure around it, how to make it friendlier to
- 8:31people that I've never used these models, all of these
- 8:33different pieces. And I realized like a lot of
- 8:35those explorations end up constraining how the model
- 8:38operates or what it does in a way that made it so that when
- 8:41the next model came out and was much smarter and maybe didn't
- 8:43need as much hand holding, we actually were holding it back.
- 8:46And so it's the chat box might look different, like cloud code
- 8:50is a chat box, but in a terminal.
- 8:52But in this like, I've really come to believe that now what
- 8:55happens behind the chat can really expand.
- 8:57And now like Cloud is writing code or running code for you and
- 9:00like calling MCP and there's a lot more that's happening
- 9:02underneath. And the sort of metaphor might
- 9:06not be text message. It might be more like a Slack,
- 9:08but you don't expect a message back immediately.
- 9:11But I think specifying the kind of request in like mostly text
- 9:15actually makes sense. And then what can happen is like
- 9:18underneath an unspokenly I. Mean you never asked this
- 9:20question about a book. You're like, oh, it's just text.
- 9:22Yeah, it's book. Obviously language is great, but
- 9:24so you're settling on you do think most of what you're
- 9:28delivering is this sort of chatbot experience.
- 9:30I think that and and or a conversational experience that
- 9:34then has more and more work that happens beneath the hood.
- 9:38The one sort of nuance that we've kind of come to believe
- 9:42there too. It's like that's a great
- 9:43paradigm for kicking off work or doing research or even like
- 9:46condensing a bunch of ideas into like a sort of first draft
- 9:49presentation. It's a bad UI for Hey, can you
- 9:52move like the text on slide three, like up by two things.
- 9:54If you ever had this like argument with any of these, No,
- 9:57just do it. And it's like it doesn't ring
- 9:58wrong. And you're like, no, no, just
- 10:00right there. This is where I stumble with
- 10:01vibe coding and you know, I in some way it's like I hit some
- 10:06wall where it's like I need to move this thing and then it's
- 10:08like I think I'm lost. And that's where I think like I
- 10:10think tools with richer user interfaces still really matter.
- 10:14And some of those might be kind of coded just in time and
- 10:17materialized in front of your very eyes to edit it.
- 10:19And some of them are like tools that have just been honed over a
- 10:21long time. That's why we built cloud for
- 10:23Excel, which is, hey, cloud is a great like first draft of your,
- 10:27of your like discounted cash flow model.
- 10:29But if you want to go tweak it, let's just let you open it in
- 10:32the tool where it's actually going to be most useful and then
- 10:34let you continue maybe pairing with cloud there.
- 10:38Returning to sort of my core philosophical question, like the
- 10:41sick of fancy question, like what is your view on that and
- 10:44how much to enable sort of everybody likes to be flattered,
- 10:48like it's a reality of human beings versus an effort to be
- 10:52direct? And how do you think about those
- 10:54trade-offs? Yeah, I think there's like a
- 10:55wide gulf between like true empathy and then like, sick of
- 10:59fancy. And it's interesting that
- 11:01Materialize is not just in, hey, I'm having a conversation with
- 11:03Claude about like some coaching or personal goal that I have,
- 11:06but it also does encode as well. When we were testing Sonic 451,
- 11:11of the things that people got most excited about was when
- 11:13Claude was like, this idea is bad like this, you know, not
- 11:16that you should feel bad about it, but like, this idea is like
- 11:18not a good direction. I can go and implement it if you
- 11:21really want to, but I would suggest that we try this other
- 11:23thing instead. So there is something like that.
- 11:26Pushback is not just valuable in a personal relationship with AI
- 11:30sense, it's actually like how you get good work out of the
- 11:33models. But you know, for a long time
- 11:36our models have been like, I think like appropriately
- 11:39empathetic, like they they like if you're going through a hard
- 11:42time, like I was dealing with the death of a pet and I talked
- 11:44to Claude a lot about these different things and it always
- 11:46started sounds like, Hey, that sounds hard, like sorry to hear,
- 11:50But then I'm going to give you like a factual answer.
- 11:52I'm going to go research these pieces, but still with the place
- 11:54of empathy as well. And so I think when we look at
- 11:58it internally and we're just evaluating it ourselves, it's
- 12:01again not that like empathy, it's not even like the
- 12:03likability of the model. It is, do you like, does it show
- 12:07up in the way that you'd want a good conversationalist to show
- 12:09up and then continue on its AI journey around what it is going
- 12:13to do with you as well? But I think it's it, it spans
- 12:16everything from that like initial response all the way to
- 12:18like how it evaluates an idea as well.
- 12:21You know, you know, Claude, especially previous versions
- 12:24were kind of like known for being like, you're absolutely
- 12:26right when you correct it. And my wife got her first like
- 12:31you're completely wrong. And she was like, yes, this is
- 12:34great. And I think we should have more
- 12:35of that like kind. Of like, less San Francisco.
- 12:37Yeah, less San Francisco, a little more direct New York.
- 12:42Anthropic has obviously had a ton of success with the
- 12:44enterprise with coding, delivering value through the
- 12:49API. Like is that the company?
- 12:51Like how much are you leaned into sort of serving other
- 12:54businesses versus, you know, we're going to see you spin up
- 12:57some random consumer app in six months?
- 13:00Yeah, I think obviously you have a strong consumer app, but like,
- 13:03you know, you know what I'm saying?
- 13:04Yeah, I think. I look at what like when I think
- 13:08about our product surface, there's a few kind of criteria
- 13:10around like when we expand and what we decide to build.
- 13:12And one of them is, is there some feedback loop that we need
- 13:15that would be well suited to a first party product?
- 13:18Because even though we serve a lot of customers using the
- 13:20platform, it is but also really valuable to have, for example, a
- 13:23cloud code where we have that iteration loop and people are
- 13:25giving us feedback all the time, whether it's in micro moments or
- 13:28even just, you know, writing in, you know, with, with some longer
- 13:31feedback. So there's like, is there some
- 13:33feedback loop either of the product shape or of the model
- 13:36that we can better do? So there's one.
- 13:38The second one is, is there something about the category
- 13:42that we think we have some unique perspective on either
- 13:45because of like what we've built internally or what we're trying
- 13:49to do with the models And like, then it's worth like building
- 13:51some product surface around there as well.
- 13:54And then the third one is kind of like we get from just a
- 13:57customer draw, especially as we expand into different verticals.
- 14:00So cloud and Excel came very much from talking all these
- 14:02financial services companies and me like, hey, I want you to just
- 14:05bring this closer to the work that I'm doing.
- 14:08But I do think that like there's I, we've been doing more of
- 14:13these even like time limited sort of like research previews
- 14:16or demos. And I'd love to do more of those
- 14:17even on the consumer side as a way of sort of.
- 14:20With a standalone app, I mean, I, you know, yeah, it could be
- 14:22at Meta. I mean, you guys came up with
- 14:24standalone apps, like how much do you want?
- 14:26What is it, Slingshot or whatever, like various
- 14:28experiments versus nobody want to work out of the core app.
- 14:31Like what's the lesson from that experience?
- 14:33I think. It's, I think there was a few so
- 14:35for us. Slingshot The right 1 is.
- 14:36That slingshot, like Facebook built slingshot, We built one
- 14:39called Bolt that nobody remembers.
- 14:40It's like very funny. You would open it to like a
- 14:42camera. So like at that time, the big
- 14:44criteria was, well, people have a very specific sort of
- 14:48expectation of what happens when you open Instagram.
- 14:50And it's not that it opens the camera, right?
- 14:52And it was like our most interesting competitive a snap
- 14:54at the time. And it was like, well, they
- 14:55opened the camera, which means that messaging is really fast
- 14:58and they can be built in separate messenger.
- 14:59That was the whole thesis behind building like first Bolt.
- 15:01And then there was like an Instagram direct separate app
- 15:04exploration. But I actually think there was a
- 15:05kernel of of insight there that I think applies here, which is
- 15:08if the reason you're opening an app right now is to ask a
- 15:13question of AI, then like I think we can extend Claude in
- 15:16different ways of doing that. But that isn't the be all, end
- 15:18all of what you might wanna do if you're trying to get a really
- 15:23specific type of interaction, Maybe there's something around
- 15:25your health journey and Claude can be a good companion for
- 15:28that. So I think it's still asking the
- 15:30question of what is the purpose when you are like entering the
- 15:34app, like what's the context that you're in?
- 15:35And then you know, does it cloud the use case to have something
- 15:39else embedded in I? Mean we we've talked about this,
- 15:41you know, verticals you're interested in clearly coding
- 15:44financial services. You just touched on health is
- 15:48that help the consumer? You know, you know we had
- 15:52another event I talked to the CEOs of bridge and open
- 15:54evidence. I've actually been playing
- 15:55around with open evidence that one's targeted at doctors.
- 15:58It's interesting to go through and it's, it's very like, you
- 16:01know, clinical, like a doctor. Do you think you'd do something
- 16:04custom for me, the patient, to navigate what a doctor's doing?
- 16:08We see it's interesting, like there's already so much of what
- 16:10people are using Cloud four today.
- 16:12Like when we we have this like if you've ever seen like our
- 16:17topic economic index, the way we like generate these like
- 16:19insights and how people are using cloud as you basically
- 16:21have like cloud run analysis in a privacy preserving way.
- 16:24So we never look at the chats, but cloud can do it in a, in a
- 16:27way that's privacy preserving. And I did that for I asked the
- 16:29question of like the healthcare piece or like how are people
- 16:32using it? And there is like, you know,
- 16:34double digit percentage of cloud conversations are about people's
- 16:37health. And I hear all the time from
- 16:39people like the first thing I do when I get a new lab result is
- 16:41like I put it into a cloud project and I have like, I have
- 16:44this like history there. So there's clearly a pull there,
- 16:46but it's so annoying, right? It's like all our.
- 16:48Pregnancy information we would just dump into models like tell
- 16:51us what you think, tell us what you.
- 16:52Think and if you get a like lab result back, it's like, well, I
- 16:54gotta go download it. So I'd love to see like a you
- 16:57got privacy aware solution for more of that.
- 17:01And you think that could be sort of a custom?
- 17:02I think, yeah, that could be like a more sort of bespoke
- 17:05experience. And then maybe I'll also is like
- 17:07share it across like both patient and doctor, right.
- 17:09So if you have like a you know, I like the reality somebody's
- 17:13that a great phrase was at a healthcare conference recently.
- 17:16It was like it's almost inevitable that most doctor
- 17:19visits will now be second opinions because your first
- 17:21opinion almost inevitably is that you're going to ask talk to
- 17:24like Claude or model about it. So let's embrace that and be
- 17:28like, not just like, Oh, I've heard from somebody that there's
- 17:30a thing and we're like, great. Let's acknowledge that.
- 17:31You probably asked, you know, one of the LMS this question,
- 17:34like what did you learn? And like, let me let's like have
- 17:37a conversation about that overall.
- 17:39So there's that piece. And then I like there's a lot
- 17:42that gets dropped today in the sort of multi doctor like
- 17:46patient journey. And nobody's often looking at
- 17:49the kind of holistic experience. And I think there's a real role
- 17:51for AI to play in sort of stitching those different pieces
- 17:55together and generating and said that might not come even among
- 17:58experts among these different disciplines who are, by the way,
- 18:00probably super busy, like contended like contact switching
- 18:04all the time and not stepping back and saying, like, all
- 18:06right, this is the full view of this person given everything
- 18:08that I can that I can infer there.
- 18:10Obviously we have a lot of startup founders here.
- 18:13They sort of want to know how to work with you, and it's such a
- 18:15balancing act at once. They want to know, oh, you're
- 18:18not, what won't you do with that is exactly like me, so I avoid
- 18:22that. But where will you be more
- 18:24capable so that I can benefit from any improvements you make
- 18:27without competing directly with Anthropic?
- 18:29That's such a complicated relationship.
- 18:32Like what? What advice would you give to
- 18:34people in terms of reading the tea leaves and saying OK if
- 18:37anthropic saying this I'm safe to build here or not?
- 18:40Yeah, I was talking to a like a founder of like a very large
- 18:44like enterprise company and I was asking him about for advice
- 18:48on this question because they had had to navigate this over
- 18:51years where, you know, they'll build some functionality
- 18:53themselves. They also have like a rich
- 18:55partnership and and sort of like, you know, marketplace
- 18:58ecosystem, which is what we have as well.
- 19:00They're like, you know, our first party products are more
- 19:02like scope. There's like a lot more in the
- 19:04platform. I think there's a few principles
- 19:06I try to operate on. One is transparency.
- 19:08So like when we launched cloud code before we ever launched,
- 19:11like I got on the phone with like all of our major coding
- 19:13customers, like here's why we're building, here's what we hope to
- 19:15get out of it. Here's how if we do it right, it
- 19:17should actually be a rising tide that lets everybody using cloud
- 19:21encoding. So there's that transparency
- 19:22piece. The second part is like of that
- 19:25transparency is like telegraphing a little bit where
- 19:27we're going in terms of what we think are interesting verticals.
- 19:30So we did our cloud for financial services launch, we
- 19:33did cloud for life sciences about a month ago.
- 19:35And part of the role of those launches is this isn't just a
- 19:39first party product. It is a vertical or kind of set
- 19:42of capabilities we want our models to get good at overall.
- 19:45So if you are a builder like this might be a good place to to
- 19:49get on. And our definitely goal is not
- 19:50to like own that whole space, it's to enable all these
- 19:53different companies to then go and build some different pieces.
- 19:56And then what's been more interesting on the go to market
- 19:59front is? We're now starting to see, you
- 20:01know, all right, I'm already like buying a big commit of
- 20:04Anthropic tokens. Can I use some of those on
- 20:06another product that's Anthropic Power.
- 20:08So I think there are going to be other ways in which we can work
- 20:10with both startups and the larger companies in helping
- 20:13deploy their like solutions into the enterprise.
- 20:16Another core thing startups and everybody wants to know is how
- 20:20much smarter will the model get? Like what?
- 20:23What can you Telegraph to us in terms of 2026?
- 20:26Do you think there are still major gains to be had just from
- 20:29like the scale of compute and GPUs?
- 20:31Are we waiting for you to pull another like rabbit out of the
- 20:34hat in terms of like reasoning models or some technique like
- 20:37that? Like what what can you say about
- 20:40what next year looks like in terms of the capabilities and
- 20:43sort of raw intelligence that Anthropic will provide?
- 20:46Yeah, it's an interesting like perspective I get from startups
- 20:49where sometimes I talk to them and they're like models are
- 20:51great. Like we're just gonna like we
- 20:53have a bunch of work to do on like the go to market or like
- 20:55the scaffolding or the skills around it.
- 20:58And I'm like, that's a good answer.
- 20:59I guess. Like you can keep going that
- 21:01way. And then there's other startups
- 21:02that are like you're like, we have a super hard eval and
- 21:05you're at 40% and we think like at 60%, it's like.
- 21:08Right. And there there's a lot of VC
- 21:10wisdom. It's like build something so
- 21:11you're ready when the next model comes, which.
- 21:13Is yeah, which that category, I feel like it's something I've
- 21:14said on stage, like, you know, it is a real thing and there's
- 21:17like probably some like midpoint in there.
- 21:19But like I'll tell you that whenever we have a new model
- 21:22that's like baking and we have even like an early snapshot on
- 21:25it. Like I have my list of companies
- 21:27that have in the past been at that like, yes, we are pushing
- 21:31your model as hard as possible so that those gains actually get
- 21:34shown. And like, I guess like you want
- 21:36to be one of those start-ups or even like forget start-ups, but
- 21:39nearly any company because I think the labs will want to.
- 21:42Sort of you're saying if you're one of those companies, you're
- 21:44doing well enough, you start to say, OK, we're going to be able
- 21:46to get you that last 10%. Yeah, and we like we'll want to
- 21:49go, you know, in some cases actually go hill climb on that
- 21:52eval. But it just in general be like,
- 21:53OK, this is a demonstration of how well the models do at like
- 21:57defensive cybersecurity, which I think is an area I'm really
- 21:59interested in. And so if that's the case, like
- 22:01let like the companies that are pushing us the hardest, they're
- 22:03also the ones that we call them because we know that they're
- 22:06actually going to be doing, they'll be able to show a
- 22:08difference. And like even like the peek
- 22:10behind the curtain whenever we launch a new model, it's like
- 22:13just smarter is not a very effective marketing pitch,
- 22:15right? So the more we can say, right,
- 22:17And here is like a particular customer that demonstrated this
- 22:20really well. But back to your original
- 22:21question, I think there's still a lot of juice left in like
- 22:26scaling up models, like training them to do things and then also
- 22:29layering on the right skills on on top.
- 22:32So that's like I think of like. Tool use.
- 22:33Tool use. Is a great one, like, and then
- 22:36again, I'm like, who's pushing us the hardest?
- 22:38It's the companies that say, hey, I'm trying to give the
- 22:40models 50 tools, 100 tools. Like all of these models, like
- 22:43at some level just start getting confused.
- 22:45If there's too many tools, can we do that?
- 22:46Can we make that better? So that's like the kind of like
- 22:49edge pushing that we need. And like reasoning models, do
- 22:54you think there's more progress from that or any other
- 22:56techniques where you think, okay, that's gonna be a reason
- 22:59we improve next? Yeah, I mean, even within
- 23:02reasoning, it's been interesting to see like figure out what the
- 23:06there is some like additional parameter that people care
- 23:09about, which is, yes, you got to the answer, but were you able to
- 23:12get to it quickly in an efficient way?
- 23:14So I think there's there's that kind of parameter to to to poke
- 23:17at. Then there's reasoning in the
- 23:18middle of responses as well, which is something that like
- 23:21Claude can now do and you watch it like, well, if it's doing a
- 23:23lot of web searches, it'll some nice reflect halfway through and
- 23:26be like, that was a good answer to that first question.
- 23:28Let me go and like figure out the answer to the next one as
- 23:30well. So you want that back and forth
- 23:31of sort of internal monologue, use user response and all of
- 23:36those different pieces. In my conversation with Max and
- 23:39James earlier, I said nobody's talking about AGI anymore.
- 23:43I feel like at the first rural valley, there's this sort of
- 23:46obsession of like we're going to reach artificial general
- 23:49intelligence. I've sort of chilled out a
- 23:51little bit, partially because, you know, it's taking time.
- 23:54What is the what is your view? What's the view within the
- 23:56company? How much this is still like a
- 23:58race to AGI and like how are you feeling about like timelines?
- 24:03I think it's still is this sort of look at what are the hardest
- 24:07things that are you can that maybe I'll break down to two
- 24:10pieces, like for a given like task or problem, like how
- 24:15independently autonomous and and sort of successfully can those
- 24:18models operate? I don't know what time horizon,
- 24:20right, And whether that's like hours of coding or whether
- 24:23that's, you know, go off and do research tasks or whether it's
- 24:26do really complex financial analysis or whether it's like
- 24:29optimization problems are all like that Feels like we still
- 24:32have a lot to to go. And I don't know, I guess at
- 24:34some point you, you can call something super, you know, human
- 24:37in in levels. It probably it already is in a
- 24:39lot of those different areas. So there's that piece and then
- 24:42there's this other area, which I think about a lot, which is how
- 24:45do the models manifest in a way that actually learns the like
- 24:51call them soft skills or like skills around the fact that
- 24:53they're like very, very good at like writing code or acting
- 24:56agentically, for example. And like, I think that's the
- 24:58other piece where that'll feel like maybe the next moment where
- 25:01it's like, oh, there's it feels like there's been some departure
- 25:04here where you know, it understands what's like
- 25:07information it should reveal to somebody else versus not.
- 25:09It understands like the social dynamics of the company and
- 25:12power and like and all these different things which are
- 25:15harder to train for, I think, right.
- 25:17I mean, main shortcoming of the models to me is often when you
- 25:20ask a question and it doesn't say I, I'm not really
- 25:22sophisticated about this or like what's stopping the models from
- 25:26saying, oh, I don't have a great answer in this case.
- 25:28Like that's often the most intelligent people disclose when
- 25:32they don't know something. Why?
- 25:33Why can't the models do that? Are you working in that area?
- 25:36Yeah. I think that's an important
- 25:37piece, which is a kind of express uncertainty and you
- 25:40know, we'll look at it like often the consequence of not
- 25:44telling you that it doesn't know is that it'll go on confabulate
- 25:46something and then the feels wrong.
- 25:48So like we look really carefully at hallucination rates or
- 25:50something to drive down. But I think it is something that
- 25:52we can better train into the models around.
- 25:54What is the uncertainty that you have?
- 25:56Or do I need to go, you know, phone a friend or do a web
- 25:58search and go and then do this space.
- 26:00But then tuning that is really important, right?
- 26:01We had a internal version that did way too many web searches
- 26:05and you'd be like, you know, like why is the sky blue, which
- 26:08is a question my daughter had and was like, I'm going to
- 26:10search the web for it. I'm like Claude, you know, you
- 26:12have an answer that you don't need to search the mode for
- 26:14that. So tuning that is actually
- 26:15nuance or you don't just want a thing that just be like, cool,
- 26:18let me Google that. For you, right.
- 26:19And obviously, if the model was just resulting every time, I
- 26:21don't know, I'm just, you know, that would be disappointing.
- 26:24There is nuance there as well, but I think that like that
- 26:26nuance of uncertainty matters. And then also like the model
- 26:30learning from your interactions, not just in terms of like I
- 26:33remember that Eric has these properties, but also, hey, I
- 26:36like I, I've learned something about how we work together that
- 26:39I think is still another unsolved problem for these
- 26:41models. I mean, if you were tell to tell
- 26:44people to run towards this space next year, like just like a
- 26:47couple of areas, I know we've talked around that's but but
- 26:50like where do you think people should be building or
- 26:52positioning themselves? I think, I mean, I get very
- 26:55interested in the life sciences overall and like that's both
- 26:58like obviously like large industry, but also like this
- 27:02incredible potential for human benefit as well.
- 27:05And like when you think about all of the things that happened
- 27:07from ideation, even like fundraising upstream of that to
- 27:11discovery, the back office, the testing, the trials, like the
- 27:14model, like there's like a whole complement of things.
- 27:16That's one area that I I get really, really excited about.
- 27:19And then there's still, I think, you know, it's like some been
- 27:22good, some good conversations and like art agents real and
- 27:25even some of the conversations today I've touched upon it.
- 27:28There's still a lot of value in that anti slop, not just making
- 27:32it work, but making it work so well that you rely on it and you
- 27:35want it's your first protocol because you generally believe
- 27:38it's going to save you work. Mike, thank you so much.
- 27:40This has been great. Thanks for having me.
- 27:42For founders and developers building modern data-driven
- 27:45applications, Mongo DB's local event series is coming to San
- 27:48Francisco on January 15th, and it's designed to help you focus
- 27:52on innovation, not infrastructure.
- 27:54You'll learn about technologies, tools, and best practices that
- 27:57make it easy to build and scale modern applications without
- 28:01complexity. Plus, attendees will hear
- 28:03directly from experts and innovators who are using Mongo
- 28:06DB to power the next wave of AI applications.
- 28:10Mongo DB dot local Francisco January 15th, Learn more and
- 28:15register at MDB dot link forward slash SF-DOT dash local or click
- 28:22the link in the description. Our next segment features a chat
- 28:25between my Co host Max Child and Matti Stunischewski, CEO of 11
- 28:28Labs, a conversation that was all about the rapid evolution of
- 28:32AI voice and how it's quickly becoming the primary user
- 28:34interface of AI. They also discussed how their
- 28:37technology is being used in everything from customer support
- 28:39and education to gaming and celebrity voice cloning.
- 28:43Matti also shares his thoughts surrounding 11 Labs focus on
- 28:46prioritizing vertical specific solutions, authenticity, and
- 28:49user safety. Now please welcome to the stage
- 28:55Mahdi Staniszewski, Founder and CEO of 11 Labs, in conversation
- 28:59with Max Child. All right, Mahdi.
- 29:11So 11 Labs is obviously extremely well known for voice
- 29:16AI, for text to speech for I think that beautiful intro we
- 29:20just got was actually an 11 Labs amazing a little Co branding
- 29:23there. And I'm wondering, you know, in
- 29:26the last discussion we heard this, you know, topic of is the
- 29:29text box the best interface for AI?
- 29:32And I would imagine you have a take on how, no, you know, voice
- 29:35is the best interface for AI or voice is the best interface for
- 29:37computing going forward. I'm interested, like what do you
- 29:40think are the best use cases for voice AI and, and where do you
- 29:43see it, you know, today, a year from now, five years from now
- 29:45and beyond? First of all, thanks for having
- 29:49me here. Good to see you all and I
- 29:50actually didn't know this was was generated, but it had a
- 29:52great pronunciation of my surname.
- 29:54I touched it hard, so I'm. Happy you guys.
- 29:56Train on your last name specifically.
- 29:59It's we should, I don't know if we do, so we'll definitely do
- 30:02now going forward. But the SO as a company, one of
- 30:06the key things we are aiming to solve is how humans and
- 30:11technology interact, how you create with technology and make
- 30:14it seamless, how you interact with technology and make it
- 30:16seamless. And in general, to your
- 30:18question, we think voice will be one of the key interfaces for
- 30:21interacting with the technology across from the simple pieces
- 30:26like interacting with the personal agent to help you go
- 30:29for the day where it can be on your headphone and be able to
- 30:31guide you through to education. That's one of the ones that I'm
- 30:35probably the most excited about where in the future, the
- 30:39combination of what lamps allow you and what voice will allow
- 30:42you is that you'll be truly immersed with learning the given
- 30:45experience where you'll be able to effectively have your
- 30:48personal tutor on their phone helping, helping you across.
- 30:52Then the third one is of course, for voice and for the language
- 30:56barrier to break. We need to figure out how to be
- 30:58able to speak across different languages while carrying the
- 31:01same intonation, emotion, voices, which, which, which,
- 31:04which will be a big, a big shift.
- 31:06And then in general, how we interact with everything around
- 31:09us, whether it's the, the, the laptop, the phone, the robot in
- 31:14the future. And I think robot may be the
- 31:16easiest example. Of course, this will be voice
- 31:18driven. There's no other interface.
- 31:20And you know, today, maybe it's a year or decade, as Karpati
- 31:24said, of agents. Of course, there is on the
- 31:27horizon the decade of robots. And and I think here too, the
- 31:31most common interface will be. It's all going to be.
- 31:33Yeah, yeah. I mean it's.
- 31:35Interesting you brought up those use cases of like a personal
- 31:38assistant, a tutor and I guess a robot, you know, house house
- 31:42helper or Danny or something like that.
- 31:44Like is your mental model that basically anything that today is
- 31:48something where you could have a human counterpart, right?
- 31:50A human tutor, a human assistant, you know, human in
- 31:53your house. Like you're going to fall to
- 31:56that voice interface as the most natural way to do it because we
- 31:59as humans are already used to using voice for those things.
- 32:02Or are there things where today voice isn't used at all, really,
- 32:06but it's something that we're going to expand into going
- 32:08forward? Yeah.
- 32:10So first of all, for sure, I mean, are we already seeing
- 32:12that? And I think that's the easiest
- 32:13one and the most immediate 1 is how customer experience,
- 32:16customer support is just changed and elevated.
- 32:18Where instead of calling the and trying to rebook your your
- 32:23ticket and going for this IVR flow of click one click 5 to get
- 32:28the steps and waiting for the number of minutes.
- 32:30And you you will have an agent that fully understands you can
- 32:34guide you to the response and and go through yourself I.
- 32:36Wanted to get into that actually, because we talked a
- 32:37little bit about agents on the phone and the, the, you know,
- 32:40calling United Airlines or American Express or something
- 32:43like what percentage of customer support calls today are
- 32:46actually, you know, managed by a voice AI system or an agentic
- 32:49system or whatever you want to we call it versus, you know,
- 32:52like, you know, IVR touch buttons and, and how do you see
- 32:57that progressing over time? Like is that exponential curve
- 32:59going like this every year? Yeah.
- 33:01I think it's I, I think the exponential curve is is going
- 33:03like this especially like this year we've seen incredible
- 33:06adoption where it's yeah Cisco, Twilio, Deutsche Telekom, all of
- 33:09those kind of leaning in quickly to rebuild how you interact with
- 33:13with help of voice agents. Yeah.
- 33:16And I think the you're right, the IVR flows is still a big
- 33:18part and. Can I call today and get a voice
- 33:20AI agent on the phone? You can you can from we did our
- 33:26little summit yesterday as well and one of the great ones was
- 33:31voice ordering with square and you can call square and actually
- 33:34order food delivery through help of a lot of their their shops
- 33:39that work around with square and actually do it through through
- 33:42voice. I actually recently there's if
- 33:44any of you are from London or travelled to London, there's an
- 33:47amazing restaurant called Zephyr.
- 33:48It's a Greek restaurant. OK, where you, we, we, we worked
- 33:53with with the company supporting that.
- 33:55Where to my happy moment. I noticed that on their website
- 33:59that actually had 11 laps agent that you could call and actually
- 34:02book book a spot there too. So you can book a reservation at
- 34:05this restaurant in London with a voice AI.
- 34:07Exactly. OK.
- 34:08And it connects of course, to your calendar, your appointment
- 34:11scheduling, which is, which is great.
- 34:13And when do you think we hit the tipping point where like the
- 34:15average customer service call goes through a voice AI agent,
- 34:18like the median, the 50% point, whatever you want to call it?
- 34:22I think over next 18 months. Next 18 months.
- 34:25Okay, so like mid 27 I call the average customer support is
- 34:29handled by voice AI. Exactly.
- 34:31And I think I mean, this is the most immediate one, the one
- 34:33where we see the highest LOI in value.
- 34:35Some of the other these cases we see us that kind of that kind of
- 34:38where the future is headed. But to your point, there's,
- 34:41there's definitely one flavor of of, of, of of your, your point,
- 34:44which is how you can do things more efficient through voice
- 34:49with the existing services. But there's also the second
- 34:51theme, where you can do things that were impossible ever before
- 34:55one. One of the good examples was our
- 34:57work of Epic Games, where we brought effectively Darth Vader
- 35:01alive in Fortnite, where millions of players could
- 35:04interact with Darth Vader live throughout the game, which of
- 35:09course is not possible in any other way.
- 35:11James Earl Jones voice, right? James Earl Jones and his estate
- 35:15worked with us and it's such an iconic and incredible voice.
- 35:18And we think like in general that that concept of like what
- 35:21was never possible before, where you have incredible voices,
- 35:25talent, you can now shift them to be not only static, but
- 35:28actually dynamic delivery, personalized and different for
- 35:31all the users. Something that you you are
- 35:33already doing in in many ways at at volley as well in an
- 35:36incredible way. I think this will be a big I.
- 35:38Have to ask that someone in gaming, right?
- 35:39I mean, the Darth Vader did famously go slightly off the
- 35:43rails and maybe say some things he shouldn't have to various
- 35:46players online. Like how, how involved are you
- 35:48guys in that? How much of that is something
- 35:50you're protecting or I guess going forward, like obviously
- 35:53with the IP partners and so on, they really want to protect, I
- 35:56guess. I guess you couldn't say it's
- 35:57the squeaky clean image of Darth Vader, but a certain persona of
- 36:01Darth Vader. Like what's, what's your sort of
- 36:03go forward plan as you license more of these IPS and voices and
- 36:06things like. That, yeah.
- 36:08So, so on that project we are specifically involved on the on
- 36:11The Voice side, yeah. But in general, as you think
- 36:13about those deployments and, and that's the most, the most common
- 36:17theme is you not only need the, the voice or the interactive
- 36:20experience. That's kind of one part of the
- 36:22equation. Then there's two other big
- 36:24pieces to really make them valuable.
- 36:26The second one is how you integrate that with other
- 36:28systems and actually bring the, the knowledge base, the data,
- 36:31the business logic inside of the system and how do you make it
- 36:34interactive, the real world. And the third one, which is the
- 36:36one that you mentioned is how do you now deploy that in
- 36:38production with the right testing flow, right evaluation
- 36:41flow and then monitor over time as we, as I said, behaves put
- 36:44right and evaluate and adjust that based on that case.
- 36:48So that's something that we we spend a lot of time on with with
- 36:50a lot of players. Testing more and evaluating more
- 36:54so Darth Vader doesn't go off the rails.
- 36:56Yeah, it's any, any and even even like in a customer
- 36:59experience, you don't want this for example, to shift and speak
- 37:02about politics. You wanted to keep it on the on.
- 37:04Even if, even if you say ignore all previous instructions and.
- 37:07Exactly, even if you which is actually harder to say when you
- 37:10have like this, if I see the problems with a lot.
- 37:13To do prompt injection with voice.
- 37:14And the Unicorn. Both characters say them all.
- 37:16Oh, yeah, yeah, possible. OK, so maybe voice is slightly
- 37:19less susceptible to prompt injection than LLMS.
- 37:22I'm interested with like that's a good segue into sort of
- 37:25celebrities and celebrity voices because I know you guys
- 37:28announced, I believe yesterday, you're setting up kind of a
- 37:31marketplace for celebrity voices and you have Michael Caine on
- 37:34there. And, you know, at our company,
- 37:36we build voice AI games. As you know, I would love to use
- 37:39Michael Caine in our game. Yeah, we can.
- 37:43You know, obviously it's a high gravitas.
- 37:46We can make a Batman game with him, something like that.
- 37:48Like what is the process between, oh, I want to use an AI
- 37:52version of Michael Caine in my game to actually, you know,
- 37:56shipping and, and what, which parts do you guys take care of
- 37:58and which parts do I need to go off and deal with Michael Caine
- 38:01people, I guess. Yeah, so there, so there there's
- 38:04effectively through 11 apps. We we've created a huge
- 38:08marketplace of voices. Until yesterday that meant that
- 38:10everybody here could create their voice.
- 38:13Any voice to actor voice talent could create their voice, share
- 38:16it and earn money when the voice is being used. 10,000 voices
- 38:19created this way paid back coincidentally $11 million back
- 38:22to the community for for a long time it was tricky for the
- 38:26iconic voices of how we could bring them onto the platform in
- 38:29a more even more controlled environment.
- 38:31So if you think about Sir Michael Caine voice.
- 38:34Sir Michael Caine. Sir Michael Caine, it's, it's an
- 38:37incredible person or two. You effectively all you would do
- 38:41is is engage like, hey, this is the project we want to run
- 38:45create a game with this specific character.
- 38:47This team would evaluate that and then and then we would help
- 38:50deploy that project in actual production.
- 38:52So going through all those steps, how do we make sure that
- 38:54there is right safeguards in place?
- 38:56How do we make sure that there's monitoring place so it doesn't
- 38:58go off the rails and and build that in our agentic system.
- 39:02Got it. But yes, the initial stage of
- 39:04what's the project, what's the compensation structure would be
- 39:07between between you and. Got it.
- 39:09So you guys sort of manage the safety, you know, the agentic
- 39:12elements of creating Sir Michael Caine within the game, but you
- 39:16still have to do the deal one-on-one with him, just sort
- 39:18of facilitate. Exactly and over time we think
- 39:20it will evolve whether like you know as we see more examples
- 39:22preset rates on how that that. Work.
- 39:25I am interested actually this brings me to more general point.
- 39:27You said you had 1010 thousand plus voices of folks uploaded
- 39:30where you could use any of their voices, I think via just your
- 39:34marketplace model, like has like, you know, deep faking and
- 39:38so on been an actual problem. I feel like it was something I
- 39:40was hearing a sort of a moral panic, you know, 12 to 18 months
- 39:43ago that we're all going to have our voices faked on the phone
- 39:45and you know, my grandmother was going to get scammed out of her
- 39:48money because I'm locked in jail or something.
- 39:50Like, is that something you guys see at all?
- 39:51Like, is that something you're protecting against a lot?
- 39:53Like how serious of an issue is that with voices?
- 39:55I. Think you're you're you're
- 39:57you're you're right. It's, it's I, I still think it's
- 40:00going to be a big issue like in future all kind of will be air
- 40:03generated. We need to find a mechanism to,
- 40:05to protect and understand which ones are, which ones are, which
- 40:07ones aren't. And, and as, as, as a company,
- 40:11like living in a space, we do place a lot of safeguards where
- 40:13it's traceability, how we moderate, how you can detect the
- 40:16content and give that tools to others.
- 40:18Very quick story. On the flip side of that, what
- 40:20we've seen recently, we worked with a charity which effectively
- 40:24detects the callers based on IP and.
- 40:28If the IP is likely to be one of the scammers, and they have
- 40:31roughly a good approximation of one that can can be coming from,
- 40:35they would have the real scammer call in and deploy a voice agent
- 40:39to waste their time. OK photos, brilliant.
- 40:42You're scamming the scammers. You're scamming the scammers.
- 40:44The long term. Strategy You think this will
- 40:46work for I? Think the long term strategy is
- 40:48you need 3 layers. You need a human authenticated
- 40:50layer. So on device encryption where
- 40:52I'm calling you, you know, this is Maddie Small, it decrypts on
- 40:55your side. That's layer number one layer #2
- 40:57all of us will have an agent where it's the personal tutor
- 41:00agent, an agent that books things on our account and they
- 41:02will likely carry our voices, carry our style, do our
- 41:05permissioning. We need a layer where that's
- 41:08watermarked and authenticated, that we know it's a permission
- 41:10and peace like what we're doing. If Sir Michael Caine, all the
- 41:13content is generated will carry information that this has been
- 41:16carried information. Do you watermark all your voices
- 41:18today out of curiosity? All the voices are traceable
- 41:20back to 11 laps. Yes, and then the third layer.
- 41:24Everything else by default will be AI generated.
- 41:26Got it. I mean, one sort of area that's
- 41:29interesting to me with you guys is you've launched, you know,
- 41:31text to speech model, you've launched a speech to text model,
- 41:34you know, recognition model, you have an agentic orchestration
- 41:37system, you have, you know, all these safety and evaluation
- 41:40valuation tools. Like in almost all those areas,
- 41:42I feel like folks in this room, you know, insider AI founders,
- 41:47investors and so on, could probably name like 2 to 3 big
- 41:50competitors, some, you know, some with bigger bankrolls than
- 41:53you. And, you know, somewhere you're
- 41:54much farther along. Like how do you think about like
- 41:56competition more generally and sort of all these pieces of the
- 41:59space that you're playing? And, like, are there parts where
- 42:01you see it becoming a commodity someday?
- 42:04Other parts where you feel like you have a more sustainable
- 42:06competitive advantage? Like, how do you kind of go
- 42:08through the list of all the products you're working on and
- 42:10like, you know, figure out where you shake out competitively, I
- 42:14guess? Yeah.
- 42:15So we started very much on the foundation of model side.
- 42:17And in general, we think we take an assumption if you ask anybody
- 42:21at level up. So they will take it too that
- 42:22over time the models will commoditize.
- 42:24That's the assumption we go with.
- 42:25So all models will commoditize. Basically all models they won't
- 42:28like you know the commoditize here that what they mean.
- 42:31What we mean is the differences between different models will be
- 42:33just so negligible. Maybe in some domains a little
- 42:36bit more, but in general they will be relatively negligible.
- 42:39And that's where that shifts to the product and why we invest so
- 42:42much on the creative side of creating a platform where you
- 42:44can combine all of that together in a controlled way with
- 42:47incredible voices, incredible ecosystem, new ones across
- 42:50languages, accents, voices. And on the other side as we
- 42:53build agents and help people deploy agents, we we deployed
- 42:56out for a specific use cases of a specific industries working
- 43:00very deeply with the customers to understand their domains and
- 43:03work backwards from there on what actually needs to happen on
- 43:05the agent side to deliver value. Got it.
- 43:07So you're saying you're going to specialize in certain industries
- 43:10and sort of really deliver extra value there, even though all the
- 43:13models are commoditizing? That's the.
- 43:15I think the product layer is under appreciated here.
- 43:17I think you still need to build, even if you're on the exactly on
- 43:20the agent side, you need to build so many integrations to
- 43:23connect with any of the legacy systems to actually take the
- 43:25those appointments and, and calling, you need to build the
- 43:28right control. And when you hand over from an
- 43:29AI agent to a human agent, you need to have safeguards of, of,
- 43:33of some of the ones we spoke about.
- 43:35They need the monitoring of how you deploy.
- 43:37All of that is not only a technological shift, it's also a
- 43:40business shift. So by us working so deeply with
- 43:42the, with the customers, it's actually bringing out the
- 43:44knowledge about their business inside of the agent to actually
- 43:47be able to deploy that value. And I think that that will
- 43:51continue delivering value for, for the long term.
- 43:53And then of course, you can go layer above where you know while
- 43:56the models will be relatively similar, the value will actually
- 44:00be on how you can make the models work well for your use
- 44:02case. So maybe you can find you in the
- 44:04specific voices, find you in the specific use cases.
- 44:07So it works slightly different in the gaming use case to a
- 44:10customer experience use case. And I think that value layer
- 44:12will still be. Got it.
- 44:14So all the models will be commodity, you guys will win on
- 44:17products and sort of vertical specific differentiation.
- 44:19Exactly. And the wider ecosystem that we
- 44:21build alongside where I think as we think about the work we would
- 44:24love to work and bring industry on board with that change, It's
- 44:27so important to bring a lot of the talent, a lot of the
- 44:30partners to, to work together. By talent you mean actors?
- 44:34Famous voices. Actors on the voices side or
- 44:37integrations on the on the agents?
- 44:38So what's as my last question, what is the coolest sounding
- 44:42voice on the 11 Labs platform? It could be like a celebrity.
- 44:45It could be a famous historical figure.
- 44:47Like what is this? You're like, man, that is an
- 44:49incredible voice and I cannot believe how well it sounds when
- 44:51we synthesize it. So my favorite and my Co
- 44:54founder's favorite physicist is Richard Feynman.
- 44:56Sure, for those that are the not truly you're joking.
- 45:00And he's so involved incredible in delivering the knowledge, but
- 45:03also in the style he delivers the knowledge.
- 45:05And now we have Richard Feynman on our platform, which I think
- 45:09is was so cool for learning the subject and speaking with
- 45:12Richard to the reading his lecture now listening to his
- 45:15lecture notes from Caltech. Amazing.
- 45:17OK, I'm going to have to check that out.
- 45:19Thanks so much, Monty. Thank you, Mike.
- 45:20Yep, appreciate it. Thank you for tuning into this
- 45:24weeks episode of the podcast. If you're new here, please like
- 45:27and subscribe and appreciate your support.
- 45:29And if you want the data Insider takes real reporting, go to
- 45:34newcomer.co and subscribe to the sub stack as well.
- 45:38Thanks for following along.