Our guest today is Aidan Gomez, the CEO and co-founder of Cohere, and he's been at the center of generative AI since the beginning. He was an intern at Google in 2017, when he helped write the foundational paper that conceptualized the transformer. That's the tech that makes generative AI possible. Now, he's focused on making AI products for companies instead of consumers and making them profitable. Here's our conversation. Aiden, thank you for joining us. Um, let's start with your work at Google and on the transformer team. And I just want to kind of talk about how that idea started. What was the original goal on that team and what you guys worked on? Yeah, the original goal was improving Google Translate. So it was a very well-known problem. Translate had been around for years, and our objective was to improve the quality of the translations that the service provides. So it's been pretty extraordinary to see the breadth of impact of a technology that was really

fundamentally developed to just improve translation. Did you guys imagine this is where it would end up or progress to? For me personally, I had no idea. But I remember a moment when we had just finished the project. We were submitting the academic paper to the conference that we were submitting to. It was like 2 a.m. and I was laying on the couch next to Ashish, and I was like, we did it, we're done. And he turned to me and said, this is going to be a huge deal. And I was an intern at that point. And so this was kind of like my first research project that I was working on and contributing to. And so I was like, what do you mean? Isn't research always like this? And funnily enough, that sort of pace of progress has only gotten faster since then. So in the past seven years, I think now at this point, there's no one in

the field who was around back then working who could
have foreseen where we are in terms of technological capability. The models are doing stuff that I personally thought
maybe I'd see at the end of my career, maybe in like 40 years. We'll see something that looks like this. And it just showed up in under a decade in just a few
years. Yeah, yeah. So Google was kind of sitting on this and like you
said, so many people, including the people you worked with, didn't know it was going to turn into what it
turned into. Should they have should they have seen it? Because the narrative right now is Google kind of got
caught flat footed by the launch of ChatGPT. Do you think that's true, or do you think Google had a
reason for not launching publicly? What's your take on that? I think what it required to actually see the future was
a perhaps even irrational conviction in this idea of scale.

If we make these models larger, if we train them on more data, if we give them more capacity, more computing power, they're going to continue to get smarter, and it's not going to slow down. Back then, we were training models that are a thousand times smaller than what we train today. And so to have the conviction to say we're going to spend a thousand times as much to continue scaling this thing up. And, you know, hopefully it continues to get smarter. That's a big and risky bet to make. I'm glad that we've made that bet. It's proven out over time, and I think it'll continue to prove out. Do you consider yourself a founding father of this technology, and if so, when people say things like that, maybe not founding Father directly, but hey, you started this thing. How does that feel to hear that kind of commentary? I don't I would never assign that title to myself.

Never, never. How does it feel when I call you that? I it's a little intimidating, I think. I think it's it's too much credit. I think really the progress that we've seen, it's the work of hundreds, thousands of people who are pouring themselves into this technology, pushing it forward. It's literally hundreds of small ideas that have been adopted and pulled in and have contributed towards the position and point that we're at today. Humans love credit assignment. We love to be able to say, you know, that person did it. Al Gore invented the internet, right? Right, right. But there's no I really can't believe that for this technology or AI in general. Google throughout its history has been famous for letting people work on side projects. Was it that kind of culture that allowed you and your colleagues to come up with the paper 100%? Google brain was extremely self-organized.

It was just a collection. A lab of people who knew each other talked about
stuff, and the folks who had shared interests would kind of congeal around a project that would last maybe
three, 4 or 5, six months, and it would succeed or fail and you'd move on to the
next thing. Um, but it was totally a self-directed research
organization. And I think that was that was a huge part of why
Google got so far ahead of everyone else. It was such an attractive place to go. If you're a researcher that is like your dream job,
that may have changed within the culture of the engineers and researchers that are building
this technology. At this stage, folks are less interested in research
for research sake, and what they want is impact. They want to be able to actually create a technology
that gets used by people, and that does stuff out in the real world, because for a long time, AI was a pure
R&D project

where there wasn't a ton of value that you could bring
to the world. It's finally crossed some sort of threshold where now
the utility is actually felt and used. And that's why you see consumers adopting this thing
like crazy. It's why you see enterprises adopting it like crazy. Even pre-launch of ChatGPT. There were some, you know, high level departures and
controversies within various AI groups at Google. What's your perception of what happened? Was there a cultural shift at Google that precipitated
all these people leaving, including yourself? I think that Google is doing really fantastic work in
AI. Still, the models they build are incredible. They have extraordinary talent. Many of my friends and mentors are still there. You know, I think that they catch a lot of flack in
the press. Um, but in reality, I think it was an extremely
non-obvious bet to pour capital into scaling, like I mentioned
earlier.

And so I actually don't fault them that much. I'm glad someone made that bet. I'm glad someone made that bet, and that we have realized this future. But I think Google is still an incredible company, incredible technology creator, and I really admire the work that they do. So what made you want to leave? I think for me it was about realizing the vision that Nick, Ivan, and I had. So we wanted to bring this technology to the world. We felt that we couldn't do that within Google, and that Cohere was the best way to bring it. What we wanted to build is an enterprise company. We want to build a platform for enterprises to adopt this technology, and Google is a consumer company, and so we couldn't do that within that environment. And so we left and created Cohere really for that purpose. When you saw ChatGPT launch and just the immediate reaction, did you think there was kind of a missed moment or did it surprise you?

I mean, when you chat to folks at OpenAI, they were also surprised by the popularity of ChatGPT. So I think it was surprising across the entire market. It it wasn't the technology itself wasn't surprising. Like we had chatbots internally. We were sort of iterating on chat as an interface onto these language models. What was surprising was how much people loved it. And they made it useful for them. They made it fun for them. And so that was a huge surprise, was that the technology was ready for the world. And it was a hugely positive event for, I think, everyone. We all got access to an intelligent assistant that is awake 24 hours a day that you can ask anything to. And for enterprises, which is Cohere's focus, the opportunity for driving automation or driving brand new product experiences, right? Computers can actually speak to us now. So the intellectual modality that we we all use to communicate and store our information, it's

accessible to machines. And so, that means we have a new interface onto every
single product, every single service that we've created inside of those machines. So let's talk specifically about Cohere because you
mentioned, you know, your interest in the enterprise. First of all, let's just start there. What is Cohere? What do you guys do? What do you sell? How do you make money? Yeah. So we're a model builder. We develop models. Our focus isn't on the consumer side. What we build is a platform for enterprises to adopt
this technology. And our priorities are privacy. So the fact that you can deploy completely privately
add in your data and it's on your hardware, we can't see anything, completely secure. The second is cloud agnosticism. So the fact that we're not locked into one cloud,
you're not buying proprietary tech that's only available on cloud A, instead it's available
everywhere, even on prem.

And then lastly, on the technology side, we focus on
the use cases that are high priority for enterprise. So it's not like, you know, help me with my homework. Let's chat about the news. It's very much I need to get work done. Help me with that. The people I talk to who are basically in charge of
buying these products for their companies don't necessarily see it that way. At least with a product like Microsoft Copilot, it's
too broad. It doesn't. It's unclear if that productivity is
actually attainable. And boy, is it expensive. What are you selling that differentiates from that
vision? Is it more niche like literally what are people using
cohere tools to do day to day? Yeah, I'd say there's two different categories. On the one side there's like external facing
applications which might be adding new features into your product that drive efficiency, you know, build
new experiences for the user.

The second side is internal facing within the organization. So driving productivity, building assistance for different functions. And you know, employee types. We're partnered with Oracle, and we power tens of new features that came out in Oracle's Fusion apps. So these are like, you know, the software that powers HR departments, customer support departments, et cetera, et cetera. And there's just tons of features that are coming out to make that work easier, to drive automation. You know, we're having research come out of MIT and Harvard showing the productivity gains. You can just quantitatively measure it. You sit a knowledge worker down next to one of these models. You train them on how to use it, how to make it useful for them. They have to learn how to use the technology, but once they do, you see productivity lifts that are like 40%. And it's not just volume lift, it's also quality lift.

As judged by their manager. There's barriers to access, the first of which is
privacy. Like there's a real risk here that your data could
leak if you know, you're just your employees are sending out sensitive documents to a consumer service. It's it's going to get trained on. Right. And then that exposes that that data to all
future users of. You're right now talking about someone who's at work
using ChatGPT--their free or personal version and uploading company documents. If you don't give it them secure access, they're going
to get insecure access from just, you know, using a consumer service. So certainly the data privacy piece is one of the
blockers. And then like you mentioned it's expensive. The cost. These models are massive. And so cohere has really focused on compressing our
models, not just building the largest model for the largest model's sake because it's it's not ready for
market, it's not actually scalable into production.

So we focus on building a smaller category of model
that's much more scalable, completely private, and cost less. So Cohere right now, at least as of a year ago, was
valued at little over $2 billion. Sure, you're probably raising money, but these
investors want to return. And it's this technology is not cheap to run, not
cheap to operate, not cheap to build. What do you tell them about when they're going to see
a return and why they're going to see a return and how they're going to see a return? I think 2023 was really the year that everyone woke up
to the technology. And so it was like the year of the proof of concept,
little tests, you know, like five, ten employees are just building and hacking and sort of gaining
familiarity with the technology that's changing. Now. The technology is starting to go to production. These POCs are coming to fruition.

And they're saying, I like what I built. I want to actually push this into prod. And those friction points that I just mentioned are starting to come into play. So when you have five people using a POC, it's not expensive. Scalability, cost, it doesn't really matter because you're not actually deploying it at scale. And so when they say, Okay, the POC succeeded, let's push this to scale. How much is that going to cost? There's sticker shock. And so that has been a point of friction that we're trying to resolve by building stuff that's faster and cheaper. The way that we funded Cohere has not been by taking massive, multibillion dollar checks from cloud providers, where we're just going to pay that money right back to them for the compute to build the models. So what we raise from, who we raise from, are investors that want ROI.

And so there's a very, very high bar for us to meet in
order to get their investment. And so it's it's very much a, you know, competitive,
legitimate non-hype company building process, which is which is
what we want. That's how you build a good company. That's how you build a good business. You talked about taking money from cloud companies and
so forth. And you know to drill down on that. There's the OpenAI deal with Microsoft where
everything OpenAI does has to run on Microsoft Azure. We've seen Anthropic make similar deals both with
Google and with Amazon, that they have to run some number of their business or some amount of their
business on those respective clouds. And by the way, it's not like these companies are
necessarily writing couple billion-dollar checks. A lot of that investment is by cloud credits. Is that the wrong model because you guys aren't going,
you guys aren't doing that.

You guys are going a sort of more traditional route, like, yeah, write us a check and we'll give you a return down the road. Yeah, I don't want to comment on my my competitors fundraising strategies, but I will say what Cohere is doing is trying to build a real business, a real independent new company for the world. There's no shortcuts. You need to convince investors who are deploying capital for a return in some number of years to give you that capital. And so that's what we're doing. We think that's the way that healthy businesses are built and always have been. And it shouldn't be an unusual strategy like it is the the right way to build companies from scratch. And so that's what we're doing. Just give us an idea of how many employees you have, where your offices are located, just the size of the company. Yeah, yeah. So we were born out of the pandemic. We're super remote, and we have offices in Toronto and New York, San Francisco.

London and folks everywhere. Yeah, we're currently a few hundred people, 400 people
approximately, but expect to grow quite a bit, particularly on the go-to
market side. So that's really where we've been underpinning like a. Salesforce. Salesforce. Exactly out there hammering the phones. And yeah, so everything to date has been done with like
five people. Right? Okay. And so we're super happy. Microsoft has a few more than five people selling
this stuff. Right. I've heard yeah I've been told. Yeah. But we're fortunate that we've partnered closely
with folks like Oracle and McKinsey and Accenture, and they have much bigger sales forces than we do. And so we can scale through them. But we need to build our own engine. And so that's what that's a huge area of investment
for us for the next year. What do you have to say to the AI doomsayers, the
people who say AI is going to go rogue and destroy us

all? I think Elon Musk yesterday or a couple of days
ago said something to the effect of there's a 20% chance that's going to happen, like tomorrow. What's your take on that? I think I'm empathetic to the fears, you know, the sci-fi narrative of computers or AI taking over and
destroying the world. It's been going on for decades, and so it's really
deeply embedded within our culture. It gets lots of clicks, headlines, it gets attention. It's shocking. So I, I understand why people are scared of it, and I
even understand why people are saying it in order to get attention. But it's not a technical truth of this technology. And it assumes continuous exponential scaling. And that doesn't happen. It doesn't happen. There are friction points. There are complexities. We're bottlenecked in intelligence of these models by
the humans that create them, because it's our data, our

knowledge that teaches them. And so there's not going to be some sort of exponential, continuous increase of model intelligence. I'm extremely skeptical of that. I think that it is mostly used as a marketing strategy, mostly used in order to get in front of more eyes. The fear mongering, the fear mongering. And so I think what I really want to see from the world is there are real risks with this technology. It's not Terminator taking over the world. It's like if we deploy this stuff into scenarios where mistakes have extremely high consequence. For example, in medicine, if we let these models prescribe drugs right now, they are not equipped to be able to do that effectively. And the cost of that could be a human life. And so we shouldn't deploy these models against those use cases. That's where we need scrutiny and a kind of tough conversation about where we want to deploy. Instead, we go for the very like eye-popping sci-fi narrative stuff.

There's so much money sloshing around. Is it are we still in the early days, or are we on the cusp of a bubble somewhere in between? Where do you see it? I think we're super early, like we've just dipped our toes. I think we're at a inflection point now that we're past the shock, the technology just landed. What does this mean for my business? What do I do? And now there's much more understanding. People are familiar with the technology. They know the use cases they want to deploy against. And so now it's really an engineering project to integrate. So I think that there's going to be massive acceleration over the next two years in terms of adoption. And for us as enterprise employees, we're going to start using this technology every day as part of our work. And as consumers, we're going to see this technology and be using it as part of our, you know, the products

that we use, and the services that we use. And so, it will become a present part of our life interfacing with these models. You said you were an intern at Google when you wrote the transformer paper. Were you still a student at the time? I was, yeah. What would you say to students today who either are, you know, studying computer engineering or what have you who are saying, okay, I want to be the next AI employee. I can only give the advice that I've lived myself, which was I got obsessed with AI. I was reading papers constantly. I would in my backpack, carry around a stack of research papers that had just come out. I'd be in the gym, and in between sets, I would pick up the paper and make notes. And initially, it was really hard. It would take me like two weeks to get through a single paper, and I didn't know what any of these words meant. I would have to.

I'd be googling, trying to understand it. And so I think commit. It's going to be hard at first. Now you can pick up a paper, skim through it in like
30s and understand it. Um, but you have to get to that point and so push
through that difficulty, stay committed, stay excited, and really pursue your
passion. If AI isn't it, do something else, right? Like I don't think everyone should be working on AI. Um, but if AI does excite you, if this technology is
what inspires you, pour yourself into it. And then finally you talk to so many businesses. I'm sure, uh, just by the nature of what Cohere does
and selling B2B. Speaking of non-tech companies, obviously all these
companies use tech. What is changing about their business models, or how
are they thinking about changing their business models because of artificial intelligence, including what you
sell or what other competitors are doing?

Yeah, I think they're looking at the technology as a competitive advantage. That's what they're looking for. What's an example of that? I mean, one example is we worked with an insurance firm for natural resource projects like pipelines, mines, you know, these sorts of projects. And it turns out when there's a mining company that's going to open a mine, they put out an RFP for insurance providers. It's essentially a race, the first insurance provider that puts a reasonable bid in front of them wins that contract. And so what we did with them was we augmented their actuaries, the folks who are doing the research on the project, assessing the risk, coming up with a quote, we augmented them with a model which had access to all the data sources that the humans making these decisions use in order to come up with that quote. And it sped the actuaries up dramatically, letting the business win more.

contracts. And so I never thought that an insurance company for natural resource projects would be adopting large language models, but they are because they're focused on winning more business. And that's their competitive advantage: speed. And so that's where they invested, and they saw fantastic outcomes as a product. So there's a productivity gain right there. Exactly, exactly. Aidan, thank you so much for joining us. Yeah. Thank you so much for having me.

How Cohere CEO Aidan Gomez Says AI Will Directly Profit Companies

Welcome to the official YouTube channel for CNBC. From 'Wall Street' to 'Main Street' to award winning original documentaries and Reality TV series, CNBC has you covered. Experience special sneak peeks of your favorite shows, exclusive video and more.

For real-time news updates from around the world, visit: http://www.cnbc.com

For CNBC's primetime TV lineup, visit http://tv.cnbc.com

For CNBC's International YouTube channel, visit: http://www.youtube.com/cnbcinternational

How Cohere CEO Aidan Gomez Says AI Will Directly Profit Companies

Scripts