That hey, if your entire technology is coming from a single country and that country decides that every now and again they're going to shut off access to you, that's not a foundation you can build on.
Hello and welcome to another episode of the Betakit podcast. I am Douglas Saltus, editor-in-chief of Betakit. This week we are talking AI sovereignty. Now, in case you missed the big news, the US government recently ordered Anthropic to ban all foreign access to its most advanced AI models. And in response, Anthropic removed access of those models for everyone. Now, the decision has sent shock waves throughout the industry, and it's prompted this really important question of ultimately who should be in control of frontier tech. It's a question that is top of mind for Canada's AI champion Cohere, which has said that it has seen a huge number of inbound requests after what happened to Anthropic. Why? Well, according to this week's guest, Cohere co-founder Nick Frosst, Cohere is simply built different, especially when compared to those other global AI giants. They're not trying to be huge. They don't sell to consumers, and they're not trying to build a digital god. They also don't control their customers' use of their models and in fact have recently open sourced several of those models so that anyone can give them a go. You might call it a more Canadian approach to LLM. As we noted in our most ambitious feature on Cohere, both the Canadian flag and Cohere's approach have been pretty good for business lately. And on this week's episode, Frosst and I get into all of that, including what it means to be Canada's AI champion, and how Star Trek informs the AI future that the company is trying to build. This episode was recorded during Toronto Tech Week before the Anthropic news broke, but you'll note that Frosst still had some interesting things to say about the company as well as other peers in the AI space. So stay tuned for that, alongside what it means for everyone when your technology comes from a single country, a country that can shut off your access in a moment's notice. So, did the Anthropic model ban just prove Cohere right? Let's dig in.
Amid global uncertainty, the path forward is clear. Canada's moment to build is now. Presented by Uber Canada, DMZ and National Bank of Canada. Betakit Most Ambitious is back telling stories of nearly 100 Canadian innovators strengthening our nation's autonomy, security, and prosperity. To see these innovators and read this beautiful 80-page print issue, head to mostambitious.betakit.com.
Nick, man, thank you for doing this.
Yeah, I want to start off by I'm gonna put some of the burden on you. I want you to make the argument as to why Canada needs or should have an AI champion. And then I think you can maybe make the argument about why you think it should be Cohere, but set up with the premise of like why do we need globally competitive leaders in this space?
Yeah, great question. So let me a few things. So there's first of all the technology that we make. When you talk about AI, really what you're talking about is large language models, which are transformer-based neural network architecture. Really when people talk about AI, they mean statistical models that take in text and predict more text. That's what we're talking about. That's what powers ChatGPT, that's what powers the consumer like when you do a Google search now and a thing comes up, it's a transformer that's powering that. There's very few companies in the world that can make those. There's like 12, 10, I don't know, somewhere between there. There's four countries that can make them. There's four countries that have companies that can make them. And the reason it's so difficult to make them is it's very capital intensive. You have to spend a bunch of money on compute to train the machine learning algorithm that can do that. You have to spend a bunch of money on data and it's really hard. It's a really challenging engineering science problem. So you need very specific people and they have to try really hard in order to get it to work. So there's not many companies that can do it.
Why does Canada need one? Or is it like Canada has the opportunity to have one because we have some of those things that you mentioned?
Yeah, I mean I would say it's more like we have the opportunity. Again, there's four countries that can do this. China, there's America, there's a company in France that can do it, there's us in Canada. That's really the extent of it. So we kind of have an opportunity to do it. I think that's an opportunity we need to seize. And the reason for that is largely if you look at the history of technology over the past 20, 30 years, it has been one that consistently promises to uplift people and in some ways has, and in many other ways has not. People have a pretty openly hostile relationship with their phones and a pretty openly hostile relationship with the internet. And a lot of people feel like that technology promised freedom, autonomy, agency, promised access to more information, better connectivity, and it delivers on some of that, but it also does the opposite of a lot of that. And largely that's because of technological hegemony.
Okay. So it's the business is controlling the tech more so than the tech.
Yeah. And if you look at the history of mobile phones and the internet, our experience of the internet is like one of four companies or one about I don't know how many. There's many people whose experience of the internet is things owned by Meta. Just white American billionaires controlling access to most of our digital experiences. And I think in many ways that's responsible for why the technology, although it promised empowerment and autonomy and agency, didn't deliver on that and instead now people really don't like their phones and feel like it's making them less able to do the things they want. So I think technological hegemony has been bad and I think the view of technology coming out of Silicon Valley is often no longer delivering on the promises it's made.
Yeah. And so we have a technology now, transformer-based large language models, that is really changing the way people use computers and really changing what can be done with a computer. And it's a very good thing to have that technology be built outside of the traditional technological hegemony.
Okay. So what I'm hearing from that and I'm going to guide this a little bit but I think we're on the same page because again we put those pages in our book. It's a really if Canada wants a different digital experience than the one provided or it even wants the option to have a different one, to build it.
Yeah, we have to build it.
Okay. So we're talking about sovereignty here folks and we obviously have released our 80-page issue entirely about that as it relates to tech. Earlier today you were on stage at Homecoming talking about sovereignty a little bit and you kind of were talking about it in terms of agency and autonomy. Is that just circling around this is what we're going for?
Yeah. Let me just say, in Canada everybody's talking about sovereignty. The whole world is talking about sovereignty right now. And what does that mean? It's a word that prior to a few years ago many people had not, it was not part of their daily vernacular. I think it would have accurately been described as an archaic term.
Yeah. And I think it's I like that it's back in the public consciousness, but I want to spend some time defining it.
So when I think about sovereignty, what I think about is autonomy and agency. And I think about that not just for a country, but at many spatial scales. So you can have autonomy and agency as an individual in your own life. Are you largely in control of what's happening to you? As a company, you can have sovereignty as a group of people working on something. Is your enterprise able to function independently of the market? And then as a country, are you in control of what's happening within your borders? Are you in control of what's going on for the people, what are the things you're investing in, what are you building, or are you beholden to other partners? And I like to think about it as autonomy and agency because then you realize that's not mutually exclusive. When people talk about sovereignty, sometimes they think it means protectionism or not working with people, but that's not true at all. You can develop your own agency and autonomy without infringing on the agency and autonomy of other people.
So there's got to be fire breaks, right? There has to be the capacity that whatever that digital relationship between nations, you set the control like that you can turn off that pipeline. It seems like right now one of our motivations and we had Vas Bedar from the Canadian Shield Institute write an op like how sovereign is sovereign enough. We're not going to put up a digital wall but to your point of the consumer digital experience, Cory Doctorow is great. Canadians do not have the option to change any of their devices because there are laws that make that illegal. So there's got to be some point where whatever interconnection that we have with other nations and other digital systems, we can either turn that on or off or as a government body and as a legislature decide which things we want to implement. And actually, shout out to Cory Doctorow for giving very pragmatic, practical, specific policy recommendations that I think will result in more digital autonomy and sovereignty for people as well as for the nation. The solution is a hacked app store apparently on your iPhone.
Well, but so it goes back to your point of tech and I think we're closeish in age, but I'm assuming we had a similar experience of coming up with tech and being like this computer is something that we can totally fiddle with. You can put in your own RAM, you can code whatever you want. You are direct to the terminal. And it is default built as an open system.
Yeah. And I think there are a generation of people growing up now. All their experience with technology, all of their digital experiences being closed down, terms of service controlled. I definitely grew up as a technological optimist. And I grew up hanging out in the hack lab of Kensington in a little hacker collective that was running there. And I was really into mesh networks and really thought that through the ubiquity of technology people would get more independence. And I'm very disillusioned with that now. That very much didn't happen. There's ways in which it did. I hang out online all the time and I'm always learning about things and I'm amazed at the amount of information that is accessible to me and so I love that. I still love technology but I hate Twitter. Right. I still look at it all the time. And everybody has this pretty adversarial relationship with the tech that they use and it mostly makes them feel disempowered. So, I'm now at this stage of my life trying to figure out why did that happen? What was the result? And one of them was that it ended up very centralized. It ended up very controlled. So, when we're building tech, we're thinking about how can we build useful tech and then give it to people so that they control it, so that they can do things with it that they want to do and they can be empowered with it. And we're making a business. We sell access to our models. We sell private deployments of language models. We connect them into enterprise data so they're actually useful for those.
This is not the fun Arduino chip level. This is enterprise level optionality.
Well, it's still pretty fun but it's not Arduino chip level. But we make this tech. We give it to people. We sell it to them. But then they control it. Then it is privately deployed in their environment. We can't shut off access. Then it's deployed. And that's what we're doing in order to make sure that the tech is in power.
Okay. Off the top, we were talking about whether or not Canada needs or should have or should honestly take advantage of the opportunity for a Canadian champion. You had mentioned some of the other countries that can do this. There are a couple of companies that are doing similar things or broadly play in your cohort. Between the global potential champions and if we're going to choose our fighter here, what's Cohere's fighting style? How would you differentiate yourself from OpenAI, Google, Mr. philanthropic? Is it just that element of choice? Is it the enterprise focus? Like how would you at this stage our business looks pretty different than those? Like we don't have a consumer offering.
You can't pay us $20 a month to have access to our LLM. Instead, we're making foundational models. We're open sourcing them so that people can use them and build with them as they want. And then we sell those to enterprises and deploy them privately.
So you're not doing any Gen AI video deals with Disney right now?
No, we don't. We can't. We don't generate images. We don't generate videos. We make a tech that can be a fundamental interface for a computer and you can use it to do the things that are boring. And we think the best place for that is at work where there's a huge amount where people are often bottlenecked by administration, bottlenecked by data processing, bottlenecked by reading through documents and finding relevant documents. And we think we can deploy them privately and securely so that it can access real data without leaking that data, without people training on it. And that's the way that we can make the tech the most empowering. So that looks very different than $20 a month for a ChatGPT subscription or a consumer offering being built out by Meta or Google.
How much of that is actually the tech? Because we can talk about efficiency of models and we can talk about what you guys have built to run on certain chipsets and I want to talk to you about open source a lot. But I feel like less of that's about a VHS versus Beta decision here and more of a although even in that example I'm using what I'm trying to get to is those are technological decisions driven by business decisions or like a founding principle but they're totally different. VHS Beta is like a completely different format. Whereas we make large language models, OpenAI makes large language models. And the language models they make are pretty similar. The way that we differ is that we're not trying to build the biggest thing.
Right now you have this war going on between the consumer offerings where each of them is trying to make a way bigger one so that it's marginally better on some data set so that they can have a press cycle where they say we're the best one for like six weeks.
For flip six weeks. And that has a handful of people flip over to pay them $20 a month and they keep undercutting it so you get the most one. And we're not interested in that. Instead, we're making one that is efficient and good enough so that it can be deployed and be value adding into a company right when we deploy it.
That's a pretty different approach. The tech is the same. It's still a foundational transformer-based large language model. Still takes in text, writes more text. That's what it does. So, how would you define Cohere's ultimate goal relative to and you don't have to you can name any names you want but like say relative to the field in those other examples.
I think it's worth talking really about so much of the tension, the distrust, the malaise, the disillusionment with tech and we experience as Betakit with the media comes when you put all of the spectrum in one box and bludgeon it.
And I'm hearing from you that you are trying to stand out in a bunch of different ways. We're going to talk through beyond just the Canadian stuff, interesting partnerships like the open sourcing, but can you clearly articulate an outcome or a vision or a goal that's separate from some of these other companies that just want to be big?
Yeah. I envision a world and I think we're relatively close to this in which you open up your computer and primarily you use language to tell it what to do. And if you're going to do a task on the computer, it is language that you are using to get it to do those things for you. And subsequently, you're much faster. You're much freer. You're more able to do the interesting and the strategic and the creative and the processing of it, the administration, the research and reading and all of that. The connecting this spreadsheet to that document and sending that document to so-and-so so they can read it and send it back to you and fill out this form and add your signature here. Send out 200 emails notifying our honorees for Most Ambitious where they needed to be on Monday for our town hall. And the entire time I was being like you shouldn't have to do that. There's got to be a better way. That's not a good use of your time. So I envisioned a world where you open up your computer and primarily use language and that's it. Now I notice I didn't say you open up a computer and there's a little person in there who tell you. I didn't say we're not building Clippy.
Well, I mean Clippy was a funny throwback.
But we're not trying to build digital people. We're not trying to build AGI, whatever that means. We're trying to make this technology useful as it is.
And that's a very different goal than some of the companies that say we're going to be profitable when we make a digital god and then we ask it how to make money.
That's like not what we're trying to do. We're instead making this technology useful for people.
Okay. But like people in business, is that because the secret history of Canadian tech is that we all grew up loving B2B SaaS and we love selling enterprise and we just that's where it's most useful. Like if I look at my life, there's not a ton in my personal life I want to optimize or automate.
You know, I'm not trying to respond to my friends faster. I'm trying to do it more because I like them and I should respond to people's texts more often, but I'm not trying to automate it. Whereas in my work life, there's tons of stuff that I don't want to do. I would way rather have a computer do it for me and allow me to think strategically, try other stuff.
I agree with you. I would say though, what I see online is maybe a bunch of people who are like, "Hell yes, I would love a digital butler to automate every experience I have with people." And I'm wondering if you just like
I just don't think that's a good there's certainly elements of it that are really great in personal life. And there are like tons of use cases for LLMs in a person's life, but when I think of where we can add the most value, it's in augmenting and automating the things that people don't want to do rather than augmenting and automating the things they do want to do. There's another component to that too which is the data component. And if you want to automate and augment my personal life, you need to know a lot about me. And for the last 20 years, it's been a really bad tradeoff to give my data to every large tech company for a small amount of value.
Is it just easier in an enterprise environment where they're like, "We've got all the data. Please help us organize this."
Well, the incentives are a lot better, right? So, I think one of the things that's strange about all is people talking about putting ads in them already. And that just like as a person who, I've been working in neural networks for like 15 years now, I find the technology beautiful. When I hear people talking about putting ads into a thing that's trying to help you think, it makes me uncomfortable.
It's gross. Makes me uncomfortable.
Already at a level where fundamentally the technology cannot guarantee you the same answer every time. So there's a trust issue and then you're gonna throw a banner ad on that. I don't like it. Whereas when we give to an enterprise, we deploy our model in their environment. We don't see what they put into it. We can't train on their data. It stays private to them. So we're not trying to extract data from them to train on or get better. We put it in their environment. We're not trying to train models that keep optimize for engagement. So our model won't ask you a bunch of follow-up questions.
Other things that you can do to.
It won't be like, "Hey, let me help you." We're not trying to do that. We just deploy it in their environment, then we connect it to their stuff and then they can use it for any of the things that people in their company don't want to do themselves.
Is that fundamentally a bad thing? I go back and forth against that where particularly with a nascent technology where the UX is language either text or voice when you can't see the buttons you can push sometimes prompts to prompt is helpful.
Yeah, I think there's a version of it that's useful. But I also find it when I'm talking to the consumer applications and they're asking me like trying to keep the conversation going. I'm like I don't want to talk to you. I want you to do the thing. I want you to do the thing I've asked you to do.
And that's it.
Is that is that why the chatbot you made for your secretive Toronto Tech Week party was like so rude and did not want to engage in conversation over text for more than two minutes? Just like no engage in farming.
We certainly were not engaged in farming. We were having a party for tech week and we had a little thing you had to text in order to see if you get in. We just thought that was funny. That was just a silly thing to do.
All right. So we're I wanted to set this up by talking through philosophically some of the differences so that we can get into how that's working out for you guys and like how that is proving out and how it's related to our sovereignty and what this country needs. But we are going to do that after the break. So stay tuned.
The Betakit podcast is brought to you by, drum roll please, the Betakit newsletter. Every Sunday, Betakit's flagship newsletter provides in-depth analysis and insights on Canadian tech and innovation news served piping hot and straight to your inbox. Think of it as a podcast that you read. With the Betakit newsletter, you'll get the top Canadian tech headlines of the week, the week's juiciest deals, Chris commentary on the moments that matter, events you'll actually want to attend, and exclusive news for Betakit's partners. Your inbox deserves better, and so do you. Stay informed on all things Canadian tech by signing up for the Betakit newsletter today by heading to betakit.com/newsletter.
So we're back and we're going to talk about this wonderful profile that the OG Daryl Eington wrote in our new Most Ambitious issue. Kind of talking about again we're kind of saying that you are standing up like the loadbearing structures of the Canadian economy which is a leap and exciting. But the other side of this feature and why we have you in the section which we call the sovereign stack which is focused on all of these Canadian companies that are building some sort of capability that we need that we can use that hopefully will not be used against us because we have it. Is how well the sales pitch seems to be going for you guys internationally when it comes to the approach that we talked off the top, but also the Canadian component. And I think Daryl quoted you in the issue just saying that you know the Canadian flag is a really good brand these days and you know the RAH is really really fun and like sovereignty on a cover is really really bold and interesting but like let's get into why people are responding to the candidness of Cohere.
Yeah. We were founded here. We have a huge presence here, but we also are a global international company. We have offices all over the world. We have employees all over the world. We have customers all over the world. And selling large language models to enterprise out of Canada has been good. There are lots of people over the world that are interested in diversifying where they get their technology from and us both being us Canada from Canada and also not the US and not China.
And also the nature of our deployment in which we do these private deployments that there's like on the stack on prem.
On prem. Yeah. On prem private deployments. We do some hosted as well. We do have a range. We'll bring our models to wherever we need to bring them in order for customers to access them. But both the where we come from, the flag that we have, and then the nature of our offering really lets people know that we're building we're selling them a piece of technology that will empower them as opposed to lock them in to something that could be switched off.
Right. Like the world has seen that hey if your entire technology is coming from a single country and that country decides that every now and again they're going to shut off access to you that's not a foundation you can build on.
So the point of leverage question that our country is dealing with is being also dealt with elsewhere and when enterprises are running the evaluation they're like these Coheres might be on to something that's useful elsewhere. Was that a pitch or were you just guys because you've done partnerships and like it's listed in the issue. It's been listed in all of our coverage. You guys were really really early working in like Korea. You've partnered with Swedish companies. You were really early to caring about the large language models working in a language other than English very well.
Was that strategy or just again from that principle? No, I think actually so we invested really early on in multilingual data and making the models work in languages that weren't English. And I guess there's a few reasons for that. One is it's technologically interesting and it's the type of stuff that you get to if you don't think LLMs are going to get to AGI immediately. If you're aware that what you're building is not a digital god but is instead a statistical machine of language and you're interested in having that statistical machine of language be useful in other languages, you know that you need data in those other languages in order to perform well. And language also changes the inputs and the outputs of those like the probability machine gives very different probabilities in French versus English versus Chinese. If you take a model and it's only ever trained on English and you removed from the data any other languages, it was really only English and then you tried it on those other languages, it would not work.
Is there a language in the world where like Reddit isn't trash and it's just like fun and everyone's polite?
No, just no. In all the languages that Reddit has spoken. There are lots of languages that have huge numbers of speakers but very little data online. The internet is way overbiased on English in terms of number of speakers to how much data there is online and that's largely due to English itself being a language that people learn in order to engage online and also due to the global imbalance of economic resources.
Like actual lingua franca.
Okay. Talk to me then about I guess you are I'll start it again whether or not this was intentional or a byproduct of that you know AGI evaluation which we'll get to. You're obviously leaning into it and we have covered and listed off a number of partnerships. I think the biggest one's like we got to talk about the Germans here. Buying Alf Alpha. I'm wondering if that is a function of the tech, the team, access to European markets, alignment of like nations who are dealing with the same evaluation that we talked about before. Like can you walk me through?
Yeah, I think it's all of those. We're really excited about it. They're a great team. They built great tech. We're really interested in having deeper roots within Europe.
Okay. And being able to collaborate on building tech better, making sure that it works across all European languages and then selling into Europe. So, it's all of those reasons.
Is there would you have made the same decision if one of those reasons wasn't there? Is there a leading decision that implies more?
I think at this scale you only make decisions like the one we have made in order to work together and collaborate if all of those things are correct.
Okay. A lot of that work we're now not just talking about like hey it's really cool that you're working with like South Korean companies or you're working with all the companies that want to sell us submarines so when we get subs and boats they're going to have Cohere on them. I'm sure you'll come up with like a cool branding for that. This is being done at a level that seems less like global commerce and a little bit more diplomacy. Like the federal government has stood on stage multiple times. I think we were there at All In in Montreal last year being like Cohere is our Canadian champion. We will champion them. We will help them win. And they always seem to be around these announcements that lead to stronger relationships not only with international companies but maybe like international markets broadly. That is not the standard startup experience or like outcome or process. Is that weird? Is that like something you guys had to get used to being like I don't want to have the tall poppy conversation, but just the fact that you guys are in some ways doing diplomatic work on behalf of Canada's new economy through tech is not something you can say about like again use a hand. How many Canadian companies on that hand could you list that are in a position like that?
Yeah, I mean let me say a few things. One, I find it deeply humbling. We didn't anticipate that when starting the company. We expected that this technology would be really influential and would define how people used computers. I didn't necessarily expect that it would be something that governments around the world would be so interested in.
So, with in a good faith way.
Yeah. Sensing good faith with that.
Yeah. Absolutely. When I hear the Canadian government say, "Hey, what you're doing is important and you've got to make sure you do it correctly." I'm very moved by that. It's very motivating. It's very humbling. It's very daunting. It's like, yeah, we have to make sure what we're doing is now relevant. And so we have to have an enormous amount of respect for that and make sure we're doing it correctly. So I find it very humbling and very exciting. But I also, you know, this is my first startup. This will be my only startup. I don't think of myself as a serial entrepreneur. I don't even really think of myself as an entrepreneur. I think of myself as a co-founder of Cohere.
So is this normal? I have no idea. And I'm not an expert in politics or in geopolitical negotiations. And so I don't really think about that. I don't really know like what is there a co-founder on the team whose job it is to think about that? Do you guys huddle up at McDonald's and you're like, "Guys, we're sending we all think about what we can do to support the country because we all feel deeply supported by Canada." And we had that like we all felt that, well before starting Cohere.
We all felt that because we all went to school here. We went to high school and university here. We got our first jobs here. We live here. This is where we're from. And we all feel very Canadian. And so we think about how we can support and if what we're doing is the right thing to support the country in whatever way we can.
But like I like on a secret mission to the Arctic and like posting selfies like that's like there's got to be a moment where you're like this is a novel experience.
Ivan definitely I think that was a novel experience for him. He's definitely very motivated by doing whatever he can to help the public sector. And so if they invite him to a place, he's more than willing to go to learn. That's very much been something he's been extremely motivated by. And it's been great to see him do it. He's so called to action by the public sector saying, "Hey, we need your help here." And that's great.
Okay, let's talk about how let's evaluate how our government is now doing in this perilous environment. And we, you know, we just had Minister Solomon on stage for our most ambitious town hall trying to get detail of the work that he's doing and it's been a little tough because I actually feel he has like the hardest portfolio of any minister ever because it's completely ill-defined. It's constantly changing and it's desperately important and we will only look back in retrospect to see if it worked. Maybe at time of recording as well also note it's Prime Minister Carney has announced that I guess the AI strategy is going to be announced next week. Mister did not have it ready to speak to us on stage which sucks. But he did talk high level about it. We reported when they released the kind of like survey results of the task force and all the participation but then also the broad Canadian submissions across the country like some 20,000 things like that. Canadians seem to be really really equally concerned about the need for us to increase productivity and then the harms for AI and in tech media circles we're all like okay well how is the AI strategy going to solve this. Do you think there is like a policy approach that Canada can have that solves these problems or creates the scaffolding to have the outcome of this technology not be like the way we described the top of this episode?
Yeah, I believe there's policy decisions that the government can make so that this technology ends up mostly empowering people and they find it useful. I think there are policy decisions.
Are you comfortable sharing those publicly because I also know because no my second sentence is I don't know what those policy decisions are and I am not an expert in policy nor am I an expert in the government. So I don't really it's not my place to make recommendations of what policy we should have.
I am an expert in this technology and I am an expert in how the technology works and what it can do and what it can't do and I'm an expert in what goes into making this technology.
So it is my place and it is the place of our company to inform and educate whenever we can about how to make this tech, what it does and what it can't do. And you guys submitted a thing and you have lobbyists and stuff like that. I don't want to get too far. This isn't a gotcha thing. I don't want to get too far.
No, I get it. But if the question is like, well, what specific policy can we make? I don't have recommendations for that just because I'm not an expert in it and I would trust other people like Cory Doctorow making very specific policy recommendations to come up with policy recommendations to do it.
Can you appreciate I think you might how let's say people in tech or more broadly people who are impacted by tech would hear someone like you who has ears of power but then also like you know 15 years in this advanced stuff being like oh I don't have the answer and how scary that is when like who.
Yeah but think about it here's the flip side of that. How many times have you heard tech people say like, "Well, I'm an expert in tech and therefore I'm an expert in all things and here's what I think." I very specifically don't want to do that. I want to be like, I am not an expert in policy. What I am an expert in is in what this technology can do and how it's built. And so I will tell anybody and we do all the time and I have been into anybody who will listen. I'll explain to you what goes into building this technology, what it's great at and what it's bad at. And then I will leave it to the people who study policy and who study the functioning of the government and the functioning of policy on technology to figure out what the right way to do it is.
Okay. Is there beyond Cory Doctorow who again has a wonderful piece in this issue about jailbreaking Canada in a really cool way.
Oh, I got to read that. I haven't read that.
You got to look at the photo spread that we did for this was so good. Beyond that, is there anyone that you're reading where you're like, "These seem like smart takes. Let's expand the sphere of people who should be involved in the conversation then."
I don't have a specific name to give you, but in general, I think the right people to learn from are the people who are building large language models. The reason for that is that there's a lot of discourse that goes on about AI. And I don't think discourse about AI is super helpful. I think discourse about large language models is helpful.
And we should talk about the technology as it is as opposed to a generic digital god that doesn't exist.
Yes. And so if you're writing policy about a digital god that doesn't exist, I mean maybe that's helpful if that ever happens.
But there's an immediate discussion that has to happen about policy for large language models and how those are implemented and what those do. So I would encourage people in thinking about this is to talk to the people who are studying and building large language models.
That's really helpful because I want to talk about the AGI thing. And I'm also wondering don't answer this yet if you are drawing a distinction between the people who are building the large language models and the people running the companies who build the large language models because you were on the 20 VC podcast talking about Sam Altman, OpenAI CEO, and like just the rhetoric around approaching AGI being like disingenuous and basically a disservice. In the past couple of weeks, people like Mark Andreessen and others have come like, "Oh, we're basically at AGI. We never defined it, but we're like totally past it because 90% of the time you can get a banner." They're all doing the mission accomplished thing. The banner's up. And they're going on the podcast saying that it's already happened, so stop worrying about it.
That's a lie. It obviously has not happened.
Okay. I just it just obviously has not. And by that I mean in a heuristic evaluation, I mean technical evaluation because I really want to like I'm truly interested in parsing this in a way that the people impacted by this both in their daily lives but then also all the young entrepreneurs who are like holy crap this is the most exciting time to ever be involved in technology in my lifetime. I want to be a part of it. And they're seeing people who have all the money and power in the world saying this has to happen. And I'm putting that on a Mark Andreessen right there where it's like this will happen, it must happen.
He is a techno optimist. Anyone who criticizes that is the enemy.
Yeah. I mean, I wouldn't make a point of listening to Mark Andreessen very often. And I am still optimistic about the way technology can go, but I'm also very realistic about it and I am jaded in some other ways as we discussed earlier. But to say that AGI is here and we have it just really flies in the face of people's experience of the technology.
I love large language models. I think they're very useful. I also treat them very differently than I treat a person and I expect them to behave very differently than a person behaves and that's because they do. And there are things that they're very good at that people are not. Like I can go to our language model and I can say to our agentic framework, I can be like hey I have a meeting in an hour. It's about a whole bunch of topics and I need you to write a 20-page report over everything we're going to discuss. Pull from Slack and from Gmail and cross reference it with the online, make a table that tells all of our reports based on this whatever. And I can have all that expected to do it in a really fixed timeline and I can expect it to be cited and accurate. Can't go to a person and do that. But if I wanted to sit down and strategize about what I think the company should do or how is this particular thing going to be received, I'm going to go to my co-founders for that. I'm going to go to the people who are experts in that. So if somebody says hey we have AGI, I would encourage them to go back and look at the media and the art that was written earlier that envisioned what AGI looked like and then see if that's what the tech looks like now.
Yeah. Neal Stephenson in all things has you covered for all of this stuff, right? Like choose any speculative fiction.
