The following interview is a conversation we had with Douwe Kiela, CEO of Contextual AI, on our podcast Category Visionaries. You can view the full episode here: $20M Raised to Build the Future of Enterprise Language Models
Douwe Kiela
Yeah, thanks for having me on the show.
Brett
As were joking there in the pre-interview, I was going to butcher your name, and I’m quite confident that I did just that. So can you go ahead and tell us how to pronounce your name and tell us a little bit more about your name and where you come from?
Douwe Kiela
I think you did a great job. Actually, I’ve heard much worse. So the French call me duvet, and I hear all kinds of different versions. So what I usually tell people is to just call me Dao, as in Dao Jones. Or if you want to try the proper dutch version, it’s Daoquila. But I don’t blame anyone if they can’t pronounce it properly. So, yeah, I’m CEO at Conceptual AI. I’m also an adjunct professor at Stanford. With this company that we started at the beginning of the year, we’re really trying to build the next generation of language models specifically for the enterprise. So we see this great need in the enterprise to use this brand new cutting edge technology, but it feels like things aren’t quite ready yet. And so our mission is to close that gap.
Brett
What do you think about the last two weeks of AI and everything happening at OpenAI? What was it like for you?
Douwe Kiela
Yeah, so it’s been interesting to follow from the sidelines. I think a lot of the information that ends up in the media around this is probably not close to the truth. So I’m sure there’s a lot more kind of subtle stuff going on there, but I’m happy that it wasn’t too disruptive, at least from the sounds of it at this point in time. It wasn’t too disruptive to the folks over there. I mean, they have some really great people, and bad for them. If this kind of led to bad outcomes because of some internal bickering.
Brett
Yeah, totally agree. Now, when it comes to your inspiration, are there any specific founders that have had a major impact on you along the way?
Douwe Kiela
Yeah, so I think there are the classic examples, right. And I don’t think I should even mention them. But the one example that I think is really inspiring, but maybe it’s also a little bit obvious at this point, is Jensen Wang from Nvidia, where they just made some really tough calls in the beginning of the company, and they paid off extremely well. But these were really like dutsy moves at the time around betting on Cuda and betting on sort of hardware accelerators and things like that. And what I really think is special, there is this respect for thinking from first principles and thinking about the scientific approach almost to how you can really build technology that is ahead of what everybody else is thinking. And so that’s really a special superpower.
Douwe Kiela
I think if you look at the most famous ceos right now, random people in the streets, anywhere in the world, they will know, like Musk or Zuckerberg or Bezos. And I think Jensen probably needs a lot more credit, deserves a lot more credit than he gets.
Brett
I was just listening to a really long form podcast called acquired on the story of Nvidia, and they were talking about different quotes that he said along the way. And I’ll probably get it wrong a little bit, but I’ll try to capture it. It was something along the lines of his desire to live and survive was always greater than the competitor’s desire to kill him, and that’s why he’s still standing today. I thought that was just an epic line and epic quote.
Douwe Kiela
Yeah. So they’ve built something extremely durable out of that. So, yeah, they have this massive moat. They’re just very hard to compete with, I think. Yeah. So wild.
Brett
It’s such a cool story and a fascinating story. Now, what about books and the way we like to frame this? We got this from an author named Brian Holliday. He calls them books. So a quickbook is a book that rocks it to your core, really influences how you think about the world, how you approach life. Do any quickbooks come to mind?
Douwe Kiela
Yeah, so one of my favorite business books is called the Culture Map. So I was first exposed to this when I was at fair Facebook AI research, working on a very international team, folks from Poland, India, China, me from Holland, people from the US, really all over the place. And we had a hard time, I think, working together as a team because of the cultural differences that just existed and that were sort of inevitable, but they just made it hard to work around because we didn’t have the right tools to talk about the cultural differences that we experienced on a day to day basis. So then I was exposed to this culture map book, and it really changed the way I look at the world, basically.
Douwe Kiela
So it’s a very helpful book in helping you understand how cultural differences can lead to different behaviors, and it really helps you understand how other people are doing things differently. So me personally, I’m Dutch. As you can hear in dutch culture, we’re very explicit, very direct. If you behave that way to somebody from an east asian culture, that can be experienced as extremely rude. So dutch people are known for being a bit blunt and rude. So this book also helped me understand myself a lot better and helped me with some tools for maybe trying to be slightly less Dutch.
Brett
How would you summarize the culture at Contextual?
Douwe Kiela
Yeah, so I think we have a culture that’s really about learning and working together to do something amazing. So we have a very flat hierarchy, and everybody is trying to make this a success. One of the things I really like about being in a startup is that everybody’s kind of naturally aligned to make this a success. And that’s very different from if you’re in a big organization like I was in Facebook before, where everybody’s kind of trying to promote their own career, maybe sometimes at the expense of others. So, yeah, with us, it’s really about just doing things together as a team and really trying to solve the hard problems that other companies are maybe less equipped to solve.
Brett
That’s a perfect segue to dive right into the big problems that you’re solving. So at a high level, when it comes to the problem that Contextual AI solves, how do you think about describing that problem?
Douwe Kiela
Yeah, so we’re solving the problems that language models suffer from right now. So I think everybody’s extremely excited about language models. Everybody can see that they’re going to change the world, right? Chat GBT when that kind of enter mainstream attention, I think that really changed a lot in how people are looking at artificial intelligence, adoption, kind of across the economy. But at the same time, there’s a lot of frustration, I think, especially in enterprises where you can build very nice demos, but to get these models to actually be production grade, so enterprise grade for a production use case, that requires a lot more work, and we’re not quite there yet. So there are a couple of big problems. Things like hallucination. These models make up stuff, often with very high confidence, attribution. We don’t really know why they’re saying what they’re saying.
Douwe Kiela
They can’t really point back to their sources. There’s issues with stillness and compliance, data privacy. You don’t really want to send your data off to somebody else’s language model, and then they can do with your data whatever they want. And then finally, there’s a lot of issues with cost quality, trade offs, where the best models are often pretty good, but they’re also so expensive that you can’t really use them for any serious use cases. So you can solve these problems one by one. That’s what our competitors are doing. Or you can try to really go back to the drawing board and try to design a better next generation of language models that overcomes these issues all in one go. And that’s what we’re building. It’s built on top of this idea called retrieval augmented generation.
Douwe Kiela
That’s really one of the main ideas around language model deployments everywhere right now. And that’s good for me, because I came up with that with my team at fair in 2019 2020. We wrote the first paper on retrieval augmented generation. And in that paper, we actually showed that what you want to do is train the entire system. So this is what almost nobody does right now, because it’s very hard to do and because most people don’t have access to the actual language model. Right. So this is an OpenAI language model. You can really access the weights yourself. So what we’re doing is building Rag 2.0 Contextual language models where everything is completely trained end to end for working on enterprise data, and in a way where it hallucinates less, has better attribution, respects, data privacy, and all of those things.
Brett
Can you take us back to, let’s say, maybe 2016? 2017, you’re at Facebook as a research scientist working on AI research, did you think that AI would be where it is today? Did you think it would have been further behind? Do you think it would have been further ahead? What do you think about the state of AI compared to your expectations as you were working on this technology in the early days, before it really entered the mainstream?
Douwe Kiela
Yes, that’s a very interesting question. I think I was quite surprised by society kind of suddenly waking up to the potential of the technology, because if you’re actually on the inside of one of these kinds of disruptions, it feels very gradual. So chat GBT wasn’t all that disruptive if you actually already were paying attention to the field. So it was more that suddenly everybody else woke up to the stuff that we had all been working on, and it’s been really interesting to see, actually how that panned out. I think this was always the promise of AI, and I don’t think there was, like a pivotal moment. And in a way, Chat GPT actually gets, and even the transformer architecture get way too much credit for this revolution.
Douwe Kiela
It’s been a very gradual process with hundreds or thousands of people all contributing little building blocks to this movement, basically. And, yeah, I think in 2016, you could already feel that this was going to happen. There were substantial advances happening basically everywhere in the field, a lot of it driven just by the availability of more data and better hardware. So better gpus, going back to Nvidia, and, yeah, that drove a lot of this technology forward.
Brett
Switching back to the problem that you’re solving. What was it about this problem that made you say, I can’t live with this problem existing. I’m going to go and build a company and dedicate the next 510 20 years of your life to solving this problem?
Douwe Kiela
Yeah. So, honestly, I think that this is the most important problem that AI can solve. So AI is going to change a lot of things in our lives, but the thing is going to change the most substantially is the way we work. It is literally going to change the way the world works. And I think if you look at the existing large language model companies that are a lot bigger than us, they’re really focused on kind of consumer market. They want to get to AGI, artificial general intelligence. They have a very specific worldview that I think wants to kind of try to solve everything. And where I think the real solution lies is in much more specialized solutions.
Douwe Kiela
So artificial, specialized intelligence, where you take these models and then you make them very good, that’s the one thing that an enterprise really wants to solve, or that somebody working in a company, if they have to do this repetitive thing over and over again, what if they can outsource that to their AI coworker and they are almost like their own CEO of their little team of AI kind of assistants? Then this will make them much more productive. So the most disruptive place for this technology, I think, will be in the workplace. And so that’s exactly where we want to be.
Brett
This show is brought to you by Front Lines Media, a podcast production studio that helps b, two B founders launch, manage, and grow their own podcast. Now, if you’re a Founder, you may be thinking, I don’t have time to host a podcast. I’ve got a company to build. Well, that’s exactly what we’ve built our service to do. You show up and host, and we handle literally everything else. To set up a call to discuss launching your own podcast, visit frontlines.io podcast. Now back today’s episode. When it comes to the solution and the business model, what does that look like? Or maybe a blunt way of asking that how do you make money? Or how will you make money?
Douwe Kiela
Yeah, so it depends on the exact deployment model. Right. So one of the things that we can do with our technology is deploy the AI models inside the VPC of our customers. So then you get a very different business model where you’re basically just licensing software and they pay for the infrastructure cost. But we also offer a SaaS solution where we essentially just host the infrastructure ourselves. And so then it’s much more of a traditional SaaS business model.
Brett
Got it. Makes a lot of sense. When it comes to your market category. How are you thinking about your market category?
Douwe Kiela
It’s very hard to think about that, actually, when you are trying to tackle something as ambitious as what we’re trying to tackle. So the market is essentially the world economy, right? Yeah, I don’t know. It sounds a bit fluffy, maybe, but I really think that’s true. But at the same time, obviously, we have to focus on specific verticals and specific use cases. So the verticals we’re looking at right now are things like finance or just technology companies in general, companies that are a little bit more tech forward. Because like I said, there’s this kind of demo disease almost going on where a lot of companies are building cool demos that kind of show the potential of the technology. But then they have a hard time bridging the gap to a production deployment.
Douwe Kiela
And so, yeah, we need to have people move beyond those demos, and that’s where we come in.
Brett
When you’re starting from a place that’s as broad as the global world economy, how do you choose which verticals and choose which use cases you’re going to really focus on? And the reason I ask, because that’s something that I’ve heard a lot of other founders struggle with as well, is choosing your market. And at some point, you do have to choose your market or choose a couple of markets to focus on. So how are you thinking about making those kinds of major decisions?
Douwe Kiela
Yeah, so we’re in a very fortunate position where we’re basically not doing any outreach and folks are coming to us with their problems. So they tried the demo. They couldn’t really productionize it. So then they’re coming to us, Fortune 500 companies saying, like hey, can you help us actually get this to where it needs to be? And so that allows you to really go in the direction of the market pool. And so I think that you can really tell initial conversations with these kinds of companies how tech forward they really are. So in some cases, they haven’t really thought about what they want to use AI for or what a production use case looks like. They don’t understand success criteria.
Douwe Kiela
They are basically asking a company like us, can you please figure out how we should be using AI, because we have no idea what we’re doing. So that’s a very different set of companies than the most tech forward companies who already know. Exactly. These are the top ten use cases that we’re most interested in. We’re not going to put all of our eggs in one basket. So this one we’re going to try with OpenAI, this one we’re going to try with Contextual, this one with somebody else and really have a strategy in place for what they’re trying to achieve. Those are kind of the ideal customers, I think, for a company like us.
Brett
I think any Founder listening in would love to answer that question, that Fortune 500 companies are knocking out their door and coming to them. What are you doing to get on their radar, and how do they know that you exist? And what are you doing to rise above the noise? Because there is a lot of noise in AI.
Douwe Kiela
Yeah. So again, I think we’ve just been very fortunate with kind of the history that we have. We have a team with, I think, a pretty amazing pedigree in the field of AI. So there are very few companies in the world that can say that they came up with something like rag with their previous teams. So we have some street cred kind of coming from that, basically. Also with the other folks on our team who have really done amazing things in their previous careers, so that helps. And then for the rest, I think just doing good work and talking to the right people and having the right investors and network is super important. So we’ve been very lucky with the set of investors that we have and the connections that you get from that also helps a lot.
Brett
What would you say has been the most pivotal decision you’ve made to date for the company?
Douwe Kiela
Yeah. So initially, when we did our fundraising, we thought that we would have to really train big language models ourselves from scratch. So this was before the Facebook or the meta, I should say llama models and llama two. We thought we had to train everything ourselves from scratch. And what’s been kind of pivotal for the company is that when it comes to the bigger scales of more billions of parameters in the language models, we can leverage open source models, which are a very good starting point and kind of contextualize those rather than having to train our entire system at that scale from scratch. So that’s been really good for us. And yeah, I think it’s really incredible that companies like meta are putting out these very high quality open source models that other people can build amazing new technology with.
Brett
If we look ahead for the new year 2024, what’s keeping you up at night? Or what’s going to be keeping you up at night, do you think so?
Douwe Kiela
I’m still a little bit scared of regulation, although it seems like the market is not getting as aggressively regulated as some of us might have feared, and for good reason. AI, I think, is an incredibly powerful technology, so we definitely need to be very careful with how we use it. But if we overregulate the market too early, then we’re just going to have a few incumbents kind of take the entire market because they can afford to hire hundreds of lawyers and a company like Contextual just can’t afford to do that. So that’s been one of my worries.
Douwe Kiela
And I think the other worry, or maybe that’s not really a worry for us, but for the AI industry in general, is this hype train is going to stop at some point and so the tide is going to run out and a bunch of people are going to get caught swimming naked. And so we’re already kind of thinking about what comes after that. We basically want to be one of the winners coming out of the future, kind of drying up of the current hype cycle. But I think that’s going to not be great for a couple of companies and it’s probably going to be next year that we’ll start seeing some cracks appear.
Brett
What do you think is the appropriate level of regulation? Or at a high level, how do you regulate this? Or is the answer no regulation until the technology has a chance to mature a bit more?
Douwe Kiela
Yeah. So I think where the debate about regulation is misguided is that we also didn’t regulate the Linux kernel or like a browser. And I think a language model is very similar to those pieces of technology where it can be used for good things and for bad things. Right. So if you have a browser and you have an operating system, you can use it to do all kinds of bad things. But that doesn’t mean that you should regulate browsers or operating systems. It means that you should regulate applications of those systems. So rather than the fundamental technology itself, we should regulate the application of the technology.
Douwe Kiela
And yeah, I think debate around regulating just the model itself and not the application of the model is really driven by a kind of cynical group of people who are, I think, trying to lock down the market in their favor.
Brett
Have you watched that Bill Gurley talk on regulation at all? No, I’ll send you a link after the call. I think it’s called like, I’ll get the number wrong. It’s like 2890 miles, and that’s the distance between Washington DC and Silicon Valley. And he just goes through all these examples and basically makes the case that the big organizations want regulation, and then they use that to essentially dominate the market and force out innovators and force out the startups that are coming in to try to win that market and take away from that market. I’ll share with you a really fascinating and fun video.
Douwe Kiela
That sounds amazing. Yeah. So this is exactly how big tech companies can stifle innovation, or how politicians can stifle innovation at the behest of lobbyists.
Brett
Yeah, makes a lot of sense. Now, as I mentioned in the intro, you’ve raised 20 million to date. What have you learned about fundraising throughout this journey?
Douwe Kiela
I personally really enjoyed the fundraising process. I think. So it can be quite nice for your ego in a way, but I think the advice I would give other founders is really to stay very close to yourself. So the more you can just be yourself and speak to your strengths and come from a position of strength and kind of confidence about who you are. And so I think the right way to do that is to stay close to yourself. That maximizes your chances of fundraising successfully and on your own terms. So if you start lying or pretending that things are better than they are, or if you haven’t really thought something through, but you’re pretending to bullshit your way out of it, people will catch up on that very quickly. So just be yourself, and that should be good enough.
Douwe Kiela
And if it’s not good enough, then you probably shouldn’t be fundraising in the first place.
Brett
Final question for you. Let’s zoom out three to five years into the future. What’s the big picture vision that you’re building?
Douwe Kiela
What we really want to do is we want to change the way the world works, literally. And so we want to be the go to enterprise large language model platform for all enterprises that really care about having high quality language model deployments all across the different workflows that folks in companies are doing. So we are working on this next generation of language models, retrieval, augmented, really, from first principles. And once that generation of models is ready, then we’ll have a good shot at helping people become their own ceos of their own little groups and teams of coworkers, and being much more effective at doing their job and changing the way the world works. Amazing.
Brett
I love the vision and I really love this conversation. We are up on time, so we’ll have to wrap here. Before we do, if there’s any founders that are listening in and they just want to follow along with your journey, where should they go?
Douwe Kiela
Follow me on Twitter. It’s probably easiest or LinkedIn, but yeah. So, yeah, I’m all over the place. You can find me on social media, and I have a very unique name, as we established at the beginning.
Brett
Yeah, we’re not searching for John Smith. It’ll be easy to find you. Exactly. Amazing, man. Well, thank you so much for taking the time to chat. It’s been a lot of fun.
Douwe Kiela
Thanks for having me.
Brett
Keep in touch. This episode of Category Visionaries is brought to you by Front Lines Media, Silicon Valley’s leading podcast production studio. If you’re a B2B Founder looking for help launching and growing your own podcast, visit frontlines.io podcast. And for the latest episode, search for Category Visionaries on your podcast platform of choice. Thanks for listening, and we’ll catch you on the next episode.
Twitter
facebook
linkedin