Skip to content
The Twenty Minute VCThe Twenty Minute VC

Cohere Founder, Nick Frosst: How To Compete with OpenAI & Anthropic, and Sam Altman’s AI Disservice

Nick Frosst is a Canadian AI researcher and entrepreneur, best known as co-founder of Cohere, the enterprise-focused LLM. Cohere has raised over $900 million, most recently a $500 million round, bringing its valuation to $6.8 billion. Under his leadership, Cohere hit $100M in ARR. Prior to founding Cohere, Nick was a researcher at Google Brain and a protégé of Geoffrey Hinton. ---------------------------------------------- In Today’s Episode We Discuss: 00:00 Intro 00:50 Biggest lessons from Geoff Hinton at Google Brain? 02:14 Did Google completely sleep at the wheel and miss ChatGPT? 05:41 Is data or compute the real bottleneck in AI’s future? 07:16 Does GPT5 Prove That Scaling Laws are BS? 15:47 Are AI benchmarks just total BS? 22:01 Would Cohere spend $5M on a single AI researcher? 32:59 Open vs Closed AI Models 36:01 Future of Prompting 38:43 Lessons from a $600M Fundraise 42:11 How do Cohere compete with OpenAI and Anthropic’s billions? 46:17 Do Enterprise Companies Trade at Lower Multiples? 56:15 Should countries fund their own models? Is model sovereignty the future? 01:05:12 Why has Sam Altman actually done a disservice to AI? ----------------------------------------------- Subscribe on Spotify: https://open.spotify.com/show/3j2KMcZTtgTNBKwtZBMHvl?si=85bc9196860e4466 Subscribe on Apple Podcasts: https://podcasts.apple.com/us/podcast/the-twenty-minute-vc-20vc-venture-capital-startup/id958230465 Follow Harry Stebbings on X: https://twitter.com/HarryStebbings Follow Nick Frosst on X: https://twitter.com/nickfrosst Follow 20VC on Instagram: https://www.instagram.com/20vchq Follow 20VC on TikTok: https://www.tiktok.com/@20vc_tok Visit our Website: https://www.20vc.com Subscribe to our Newsletter: https://www.thetwentyminutevc.com/contact ----------------------------------------------- #20vc #harrystebbings #nickfrosst #founder #cohere #chatgpt5 #anthropic #europe #aimodels

Nick FrosstguestHarry Stebbingshost
Sep 1, 20251h 14mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:00 – 0:50

    Intro

    1. NF

      I don't think Sam Altman has done a service to the world by talking about how close AGI is. I think he has made several predictions now that are wrong, and that were obviously wrong at the time he made them. I think AI will probably lead to the end of the world. You know, he's made allusions to things. He did a world tour where he spoke to every major leader, the world over, to tell them, "Hey, this technology is gonna pose as an existential threat." And I think that was academically disingenuous, and I think did a disservice to the technology he loves.

    2. HS

      Ready to go? Nick, I'm so excited for this, dude. When I had Aidan on the show-

    3. NF

      Mm-hmm.

    4. HS

      ... he was like, "You've gotta have Nick on. He's the real star of the show." And he introduced us way back then.

    5. NF

      Mm-hmm. Hmm.

    6. HS

      So, I'm so excited that we could make this happen.

    7. NF

      Yeah, man. I'm happy to be here.

    8. HS

      Now, before we dive into

  2. 0:50 – 2:14

    Biggest lessons from Geoff Hinton at Google Brain?

    1. HS

      Cohere, I have to ask, you were Geoff Hinton's first hire at Google Brain.

    2. NF

      Mm-hmm.

    3. HS

      And so then you're put in a room with Geoff Hinton. You get to work with him every day. What was the biggest lesson from working with Geoff, a legend of the industry?

    4. NF

      Yeah. I, I learned... Yeah. I, I loved working with Geoff. Um, I learned everything I know about research, um, from those, those... I think we were there for four years, three years? Um, I think I was very surprised at how creatively and playfully he approaches research. Um, when we would discuss, like, algorithms or, or, or, like, optimizers or, or, uh, loss functions, we would discuss them often in, like, through physical analogy. So, we'd spend a lot of time talking about, like, imagine there's like a ball here and like an elastic band to this thing, and a pulley here, and like this is what the... you know, it's on this kind of a surface. And like a lot of it was descriptions in the natural physical world. And that was very, yeah, like playful. And a lot of it was approached with like, "Oh, what would happen if..." You know, with, with curiosity. Um, and I didn't ex-... Uh, when working with him, I didn't expect that. Um, I expected hi- it to be much more like, you know, just, "Here's the equation. Let's, let's figure out what the derivative is and let's, let's go from there." Whereas instead, a lot of it's based on like intuition.

  3. 2:14 – 5:41

    Did Google completely sleep at the wheel and miss ChatGPT?

    1. NF

    2. HS

      When you look at Google Brain and you look at DeepMind, a lot think that really kind of Google were asleep at the wheel, given them not being at the forefront in what was the consumerization of it with ChatGPT.

    3. NF

      Mm-hmm.

    4. HS

      Do you think that's fair?

    5. NF

      I don't know. I mean, uh, it's certainly interesting. Look, like the transformer was invented at Google, right? Like, there was... Uh, in 2017, Aidan, uh, um, amongst with, um, many other brilliant people in Google Brain published the transformer as an architecture. Um, it wasn't, it wasn't then commercialized very quickly within Google. It wasn't scaled up very quickly within Google. Um, a lot of that work had to be done elsewhere and years later. So, that's interesting. And like w- why that is, like what, what, what systems are in place to make that be the case? I, I don't know. I, I, I will say there's still a ton of brilliant people in DeepMind, I think, now. It's, it's just cons- sub- subsumed the rest of it. Doing great work, um, and they continue to make good products. Uh, it is interesting that all the people who worked on the transformer left to continue to work on the transformer.

    6. HS

      If we kind of go down your tangent, for people who don't know, and just to set the scene before we dive in-

    7. NF

      Mm-hmm.

    8. HS

      ... properly, what is Cohere and how does it differentiate from more generalized models that are maybe more well known, like your OpenEyes-

    9. NF

      Mm-hmm.

    10. HS

      ... and your Anthropics?

    11. NF

      Yeah. So, we're, we're a foundational model company, um, like those other two. So, we, we build foundational models, we build language models. Um, there's maybe... I don't know. Maybe there's like 10 companies in the world that are building lang- large language models.

    12. HS

      In the West.

    13. NF

      In the... Yeah. Maybe the... In... Maybe there's like 15 in... I don't know. We'll have to figure out how many. There's a few that have popped up recently.

    14. HS

      Okay.

    15. NF

      Um, but there's some num- some number. Less than 20 in the whole world. Most of them in America. Uh, a handful of them in China. Uh, us in Canada, and one in France.

    16. HS

      Yeah.

    17. NF

      Uh, so those are really the, the companies out there. We're unique in our singular focus on bringing this technology to enterprise. So, that means we train a model that is good at enterprise tool use. So like, we train a model that you can, you know, give it a bunch of tools and APIs within your business, give it access to your business's data, and then you can ask it to help you with something in your work and it does a good job of it. So, that's what we train it for.

    18. HS

      How does the focus on enterprise over consumer change the way in which you train and build a model?

    19. NF

      Hmm. So, the models themselves, like transformer architecture, which is the, the original model that... Yeah. That was introduced in 2017, uh, hasn't changed very much, right? Like every... All, the whole industry is still using transformers. We've changed the way we train them, but the model architecture itself, you know, we're approaching 10 years of, of the same model architecture. Um, when we train our model, we're not training it to be like an amazing conversationalist with you. We're not training it to like keep you interested and keep you engaged and occupied. We don't have like engagement metrics or things like that. We're just training it to augment you in the workplace. We're just training it to help you do your job. Um, and that means the type of data we train it on is very different. So recently, we started doing a bunch on like synthetic data. So, we, you know, generate a whole bunch of data, um, to create like fake companies and fake emails between people at these fake companies, and fake APIs within those fake companies. And then we train the model in that synthetic environment to help out within that fake business.

  4. 5:41 – 7:16

    Is data or compute the real bottleneck in AI’s future?

    1. NF

    2. HS

      Do you think data is a bottleneck given the ability for synthetic data to produce infinite supply?

    3. NF

      Yeah. Data is still a bottleneck. And you, you need, you need real world data in order to start a process of synthetic data. Synthetic data has helped a lot, um, and it's made models better than they would be if they didn't have access to it. Um, but getting access to high quality data, uh, is still something people think about. We still, you know, we still make a whole bunch of data in-house with annotators, um, who are making real data and not synthetic data.

    4. HS

      When you think about kind of the three pillars of compute, algorithms, and data-

    5. NF

      Mm-hmm.

    6. HS

      ... which one do you think is most constrained or the biggest bottleneck?

    7. NF

      ... it's interesting. I mean, the algorithms haven't changed very much. Like that's an interes-... Like, they've changed a little bit. You know, when we started this industry, you know, originally we were just training base models, which were not called base models at the time. They were just called large language models, um, but they weren't trained from human feedback. So all they would do is, you know, take in the first part of a sentence and write the second part of a sentence. Um, but if you tried to have a conversation with them and like... it wouldn't work, 'cause that wasn't the data they were trained on. Um, since then, you know, now we, we train models in a few different steps. There's like a base modeling step, then there's a reinforcement learning step from human feedback with SFT data. After that, you know, there, there might... there's a variety of other reinforcement learning techniques you, you can do. Um, but the algorithms, I, I think, are not the bottleneck in terms of making those models more useful. I do think a, a lot of it is still getting good quality data and then making good quality synthetic data from your good quality real data.

  5. 7:16 – 15:47

    Does GPT5 Prove That Scaling Laws are BS?

    1. NF

    2. HS

      When we think about kind of the bottlenecks, that leads to potentially a plateauing that people are worried about, and everyone seems to now be on the train of, "Hey, more compute, scaling laws are more real than ever, and we will continue this exponential progress with more compute." Do you agree that we are seeding the benefits of scaling laws for the continuous next 12 to 24 months? Or do you think that actually more compute will not just lead to more progress?

    3. NF

      How much better do you think GPT-5 was than GPT-4?

    4. HS

      I actually think it was worse.

    5. NF

      So I think that was kind... I, I think that tells you something about the nature of just throwing more compute at the problem.

    6. HS

      Does it or does that show an o-... So why do I think it was worse? I think it was worse because actually the way that they now do model selection is slower and more cumbersome, and actually it's a pain. It gets it wrong sometimes, I mean, I just want a quick answer and it suddenly goes into deep research-

    7. NF

      Mm-hmm.

    8. HS

      ... and I'm like, "Oh, for fuck's sake, I just want a quick answer."

    9. NF

      (laughs)

    10. HS

      Right? Yeah. All right, PhD, calm down.

    11. NF

      (laughs)

    12. HS

      (laughs) Uh, I do know what I mean? And so, uh, I think it's a worse product in that respect, and I think we, uh, waited for y-... a year or a year and a half-

    13. NF

      Yeah.

    14. HS

      ... for model auto selection.

    15. NF

      I, I think, like if I go back to your, your, your original question of like, you know, do I, do I think just throwing more compute? Like some people are thinking there's a plateau.

    16. HS

      Yeah.

    17. NF

      Do I think sh-... there's more to compute. Uh, like under your... Like, I think we need to agree on where I th-... where we think the technology is going to establish whether or not there's a plateau. Right? Like, I think language models are incredible. I think they're super useful. I use them in my work life as often as possible. One of the reasons why we're focused on the enterprise is because that's really where I think lar- large language models are useful. Like if I look at my personal life, there's not a, there's not a ton that I want to automate. You know, like, I actually don't want to respond to text messages from my mom faster. I want to do it more often, but like io, I want to be writing those. I want to be like engaged, you know? Um, whereas in my work life, there's a ton of stuff I don't want to do. Like I... Like we need to get to a stage where I can, you know, open up North and I can say, "Hey, uh, file my expenses," and then it can figure out, okay, cool, I got to, you know, look through all your emails. I got to look through photos of receipts you've taken. I got to cross-reference that with the things you're allowed to expense via internal documentation. Then I got to figure out what the API is for how to expense things within your company, and then I got to do all of those and get approval before I do them. Like that's a super... that's a many step process, um, but that's where the technology is going. So that work, the work of making a m- model do that is not plateauing. That's more modeling work, that's more, uh, product work, that's like building better connectors, that's building, uh, safer data integration so that you can trust giving a model access to the types of stuff I just said. That stuff's still ongoing, and that's what we're, we're working on. I think when people are talking about building towards AGI, um, like, I don't think this technology gets us there.

    18. HS

      When you say gets us there-

    19. NF

      Yeah.

    20. HS

      ... what is there?

    21. NF

      W- uh, well, yeah. Great question. We've had many years of people discussing AGI, um, and not many definitions thereof. (laughs)

    22. HS

      Next to none.

    23. NF

      Yeah. Yeah.

    24. HS

      I mean, my, my definition is when Sam Altman and Microsoft decide.

    25. NF

      Yeah. They've changed their definition a few times on that. When I say AGI, what I mean is a computer that you treat like a person. I mean, when you use a computer and you expect it to behave like a person and treat it that way, um, I'll call that AGI.

    26. HS

      Do you not think we're already there then?

    27. NF

      Uh, people do not treat language models like they treat people.

    28. HS

      Do you think OpenAI and Sam Altman then now realize that more compute does not lead to this exponential progress when they look at GPT-5?

    29. NF

      I don't know. I think... Like, uh, you know, they're a great company, they built a really cool consumer product. Um, I, I don't know what they're thinking.

    30. HS

      Well, I guess my question is, why does the world still think scaling laws are so prevalent when you don't?

  6. 15:47 – 22:01

    Are AI benchmarks just total BS?

    1. NF

      on that stuff.

    2. HS

      Do you think the benchmarks are bullshit? 'Cause we place a lot of emphasis on them in, on Twittersphere, on the Reddit sphere. Are they bullshit or- or are they a accurate reflection of model progress?

    3. NF

      Let's go back in time a little bit. Um, when we first started in this industry, the benchmark that was used the most was called LM1B. Do you remember that benchmark?

    4. HS

      No.

    5. NF

      Okay, cool. Uh, that was a benchmark that was like taking in the first part of a text, like of a, of a newspaper, and then writing the second part of a, of a, of the newspaper article. Um, after that there was a benchmark called Hella Swag. Do you remember that one?

    6. HS

      Yeah, I do remember that one.

    7. NF

      All right, cool. That, so that-

    8. HS

      Yeah.

    9. NF

      ... okay, so that's like 2022, so that's like-

    10. HS

      Tha- that's my introduction. Yeah, yeah, yeah.

    11. NF

      That's when you started. All right, cool. Uh, no one's talking about that anymore, right?

    12. HS

      Right.

    13. NF

      Yeah. Um, now a lot of people talk about like AIM as like a math reasoning AMI, or actually don't, that math reasoning benchmark.

    14. HS

      Okay.

    15. NF

      None of our customers ask the model to do math reasoning. That doesn't come up in the workplace that often. That comes up in a few workplaces where mathematicians work, but there aren't a ton of people out there making a living doing math reasoning. Um, stuff like the ARC AGI Challenge is a benchmark that people talk about, but that's like a pixel manipulation challenge. It's like, you know, taking in like a grid of pixels and based on rules predicting the next one. That's not a thing any of our customers have ever asked the model to do, nor do I think they will. So do I think they're all bullshit? Um, I think taken hol- like I, I think, I think it's interesting. I don't know. There's, there's good scientific work in some of them. I think it's very interesting to evaluate emergent capabilities for models.

    16. HS

      But they're not an accurate reflection of the utility value of models.

    17. NF

      They're a reflection of how much the model had been trained on those benchmarks.

    18. HS

      So you can gamify them, essentially.

    19. NF

      Oh, you can definitely gamify them. Yeah.

    20. HS

      Do the big players gamify them?

    21. NF

      I don't know. Yeah, I don't know. I, uh, none of it's too, like none of it's too relevant to us, right? Like I don't think those leaderboards are that helpful. I think in a consumer space, it's cool. I think if you're making a consumer app and it's like exciting and fun and people like to look at it and they want to try out the most recent thing, like that's fun. That's cool. Um, I think if you're not in a consumer space, uh, people don't care about the hype as much. They care about like, "Hey, did I get to production? Hey, like did, you know, did I buy LLMs, deploy them, and then get ROI on that?" You know?

    22. HS

      Gi- given the pace of deployment, we are seeing model evolution so fast and so rapidly-

    23. NF

      Mm-hmm.

    24. HS

      ... that you're essentially seeing this kind of decay rate on models being greater than ever because it's like next one, next one, next one. And actually-They're still being trained, though, on H100s or NVIDIA chips from 18 months ago. Is there a misalignment in terms of the progression of models versus the progression of chips?

    25. NF

      You can cycle through new versions of models quicker. I mean, it's still, uh, it's very slow. Still, like I, when I was training neural nets in 2000 and, I don't know, 11, I mean, it would take, like, you know, hours to days. I remember being like, "This is crazy. I can't believe this takes so long to train this model." Now we spend months, months training models. So like, you know, that's, that's a timescale, I, I, I didn't anticipate when I was working on this, um, a long time... when I was working on neural nets a long time ago. Um, but that's still very different than the timescale of working on chips, right? Like, that's still slow. I think when you talk about, like, we're, we're seeing all these models iterate so quickly, like yes, on the one hand we're seeing models iterate really quickly and people are releasing new models. On the other hand, there's still the transformer that was invented in 2017, and there's still sequence models, and they still take in words and predict the next le- word. And we've changed how they're trained a bit. Uh, we've added on steps, like now there's a, you know, base modeling step, then a SFT, like supervised fine-tuning from human feedback, where like if somebody writes a sentence, someone writes the sec- the response they want, and we train on that. And then there's a reinforcement learning aspect where the model is generating and you're telling it that's good, that's bad or something. So there's like new ways of training it. But fundamentally, the tech is still the same. Um, we keep making them better, keep iterating on them. But it's not as though we've like c- anybody has, you know, trained a, a model that's fundamentally different than a transformer. So, it's, it's an interesting dichotomy. Like on the one hand, there's constantly new stuff. On the other hand, eh, we've been working on the same stuff for a while.

    26. HS

      We have been working on the same stuff for a while. The thing that has seemingly changed is the value of the people working on this stuff. You know, we're now seeing billion-dollar people in terms of Zuck's willingness to pay for, like, chief scientists. You recently hired, um, and I just wanna get the name right, which is Joel Pineau? From...

    27. NF

      Yeah, Joel. Yeah.

    28. HS

      Joel from Facebook.

    29. NF

      Yeah, yeah.

    30. HS

      Or Meta. My question to you is, how do you think about the war for talent that we're seeing today?

  7. 22:01 – 32:59

    Would Cohere spend $5M on a single AI researcher?

    1. NF

      but not all.

    2. HS

      Do you sit down, though, with Aidan and the team and go, "Shit, we need to step up what we pay people because we are in a war for talent, and budget is a big part of it"?

    3. NF

      Mm-hmm. We, we definitely think, I mean, we, our, like our company is, um, you know, what, what we spend the most time thinking about and talking about. Um, and the company is the people who are there, you know. That's fun- functionally like we, we, we are only the people who we have the privilege of working with. Um, and so we do think about, hey, is this the right place for... Like, are we making sure that this is the right place for people to work? Um, are we making sure that this is getting the best in their lives, uh, in a financial perspective and in a personal...

    4. HS

      But would you spend five million on an AI researcher?

    5. NF

      I mean, if they were bringing in the right value, yeah. I mean, like there are people who are... There are certainly lots of people who, through our equity, like own portions of that, you know, own, own what you're talking about. Um, and I feel great about that.

    6. HS

      Do you worry that the industry's kind of becoming commoditized or transactionalized with the hype around it?

    7. NF

      I don't like... Yeah, I, I do think the hype around it is misleading sometimes.

    8. HS

      Mm.

    9. NF

      Like the technology... Like I, I'm in such a strange place of being caught between the technology is the most beautiful technology I've ever seen. It's the most transformative technology, uh, I've ever worked with. I'm... Like, it, it is already fundamentally changing the way I do work, and I'm very sure it will fundamentally the change, change the way we all do work soon. Like, I think that. On the other hand, there's a lot of hype around it. There's a lot of misleading rhetoric. Uh, there's a lot of misinformation. And I don't think the hype is necessarily helpful for getting to the truth.

    10. HS

      Can I just dig in there?

    11. NF

      Yeah.

    12. HS

      What do you think is the hype and misleading rhetoric that is most damaging or confusing?

    13. NF

      Yeah, I think the hype around AGI is the most damaging and confusing.

    14. HS

      This assumption that we will all have no work to do, we'll all be in UBI and...

    15. NF

      Yeah. Yeah, and even bef- I mean, this is kind, this isn't really in the discourse as much this year as it was last year and the year before that. Um, and that's because it's pretty clearly not true. But the idea that, oh, this technology is, like poses an imminent existential threat to humanity writ large, um, was not a, wa- was incorrect, um, and not helpful for talking about the real ways in which this technology...... could be damaging, the real ways in which this technology, uh, could, could shake up the system and, and cause, you know, rapid changes. Um, and it, it's, was not helpful for getting people to understand what the technology is, right? So I, I don't, like, I don't hear that as much anymore these days. Um, I think that's because people have realized that that's not the case. But the remnants of that discourse are still in the, in the world, are still out there, yeah.

    16. HS

      I think the remnants are, and I think they're most prevalent in the way the internal employees in large organizations respond to AI being introduced. People do not welcome the introduction of AI in large companies. They-

    17. NF

      Yeah.

    18. HS

      Maybe a European sign, but a lot are very nervous and scared and do not embrace it wholeheartedly.

    19. NF

      Interesting. Yeah, I think we, we, I haven't found that as much, um, when we've worked with our customers in our enterprises. Um, I, I find a lot of people are, are interested and excited about using an LLM. Mostly that's because, I think, they realize that the LLM is augmentative, um, for the most part. And it, it allows them to not do the things they don't want to do.

    20. NA

      Yeah.

    21. HS

      I mean this in a nice way. Do you actually buy that?

    22. NF

      Do I actually buy that? (laughs)

    23. HS

      Yeah, like ni- nice, yeah, yeah.

    24. NF

      I love it. Yeah.

    25. HS

      I mean, I had Benioff on the show from Salesforce, like-

    26. NF

      Yeah.

    27. HS

      ... two days ago, and he says, like, "Oh, the same human plus agent, yeah, great."

    28. NF

      Yeah, yeah.

    29. HS

      Are you serious?

    30. NF

      (laughs)

  8. 32:59 – 36:01

    Open vs Closed AI Models

    1. NF

      around employment.

    2. HS

      Can I ask you, when we think about, like, problems to solve, I think a lot of people also get worried about the open versus closed argument.

    3. NF

      Mm.

    4. HS

      Um, how do you feel about where the future of efficient AI lands in the balance between open versus closed models?

    5. NF

      Mm. So at Cohere, we, we make our foundational models, and then we release the weights for non-commercial usage. So, we're somewhere in be- in the between the, like, open and closed, right? We're a for-profit company. Like, we, we exist to make money. Um, and so we don't, we release our weights for scientific and research and, like, you know, you can download it on your computer and run it. Um, I think that's a good sweet spot for us as a business. That allows us to, like, you know, build credibility within the community. If people want to check out our weights, like, they can go check them out, right? Like, there's lots of companies that started out as open, um, who no longer release the weights of their models, or who never did, right? So, we have our models out there. You can go look at them, you can use them, you can validate, "Hey, do they work on my problem? Yes or no." Um, but if you're using them for commercial purposes, you gotta talk to us, and then-

    6. HS

      (laughs)

    7. NF

      ... and then we figure out a commercial relationship so that we can, you know, exist as a, as a business. That works for us. Um, I'm surprised to s- I'm surprised there aren't more businesses taking that tact, uh, more foundational models taking that approach.

    8. HS

      Do you think Meta will move to a closed model approach from an open?

    9. NF

      They've certainly hinted at that, right? There certainly looks like... But I don't know what they're, I don't know what they're doing over there. Yeah, I don't, I don't think a lot of people know what they're doing over there, and I don't spend a lot of time thinking about it.

    10. HS

      Do you not think it's helpful for founders to be very aware of competitive landscapes in case they're asked about them by customers, in case customers are going, "Hey, w- why aren't you more open? Why aren't you more closed? Are we, is our data secure-"

    11. NF

      Yeah.

    12. HS

      "... if you're..."

    13. NF

      This, as in, as with most things, like, a middle ground is the right place to be, right? You could spend your whole time as a founder only looking at competitors and being like, "Oh, why are they doing that? Why are they doing this? What's going on with that?" You know, you know, um, and that will, I think, not be helpful for you. You could also spend your whole time, you know, with your head in the sand, only thinking about what's going on in your company, and I think that would not be helpful either. You have to find some middle ground. The discourse around AI is inescapable. You know, you, you, you would be hard-pressed to ignore it. It is every other headline, you know, it is, it's all over the place. I don't think there are many people who work in the industry who suffer from not enough information about what's going on in AI, right? I think there's a lot of people who suffer from way too much of it and obsessing over the minute details of like, you know, how, w- so-and-so got 0.2% better on this thing, or like, you know, is, you know, constant min- like, constant small changes in businesses out there. Um, and I think that can mislead you from staying grounded and, like, what are you actually doing? Who are you actually helping? How is this making, you know, things

  9. 36:01 – 38:43

    Future of Prompting

    1. NF

      better for your customers?

    2. HS

      Do you think we will still have prompting as the core user input guidance mechanism in five years' time?

    3. NF

      Prompting as in, like, you write something to a model and it writes back?

    4. HS

      Y- yeah.

    5. NF

      Yeah. Yeah, what else would it be?

    6. HS

      The way that it changes, the way that you do it changes. You wouldn't say, "Hey, make it a funny tone," um, or, "Hey, add in a light personalized style that also is sincere."

    7. NF

      I think the idea of prompting as a skill will become less relevant. And if you look at, like, that's the trajectory. Like, when I started doing this, if you wanted to get a model to summarize something, you wrote the first paragraph, and then you wrote, "In summary," colon, new line, and then you'd generate it.And like that was, like that was the skill of prompting, was figuring out how to trick a model into getting it to do what you want it to do. And that's because they weren't trained on feedback from people, they were only trained on text from the web. And so all they were were sequence models from text on the web, and nobody on the rep, on the web wrote, like, "Please summarize this for me," and then a summary. They wrote a paragraph and then wrote, "In summary:" and so if you wanted to get the model to do that, that's what you would do. Language models are like, ff- w- we're training them more to fit how people expect them to work, and that means that getting good at prompting is less important. So I think the idea of saying, like, "Oh, yeah, you gotta learn how to prompt" is gonna go away. I think the idea of saying, "You need to learn how language models work and you need to know what they can and can't do," in the same way you had to learn how a computer works and what it can and can't do, and you had to learn how a telephone works and what it can and can't do, like I think that's going to exist. But, um, and, and that means, like, prompting is gonna exist. The idea that, you know, you write to a, you write something to a model or you say something to a model and then you get the response back, and if it's not what you like, maybe you iterate a little bit, like that's gonna exist. That's fundamentally how the technology works.

    8. HS

      Mm-hmm.

    9. NF

      But the idea of it being a discipline that you have to train to do, um, we've already seen that trajectory get, uh... Yeah, it's al- it's already gotten easier. I, I look for people who know how a language model works, and who know... One of the, one of the core things that's, you know, one of the things that's been, uh, a necessary component about working at Cohere, uh, is you can't think the technology is magic. You can't think, uh, th- this is like we're doing spells. You have to know that a lang- how a language model works, how it's trained, and what that means for it. What emergent capabilities happen, uh, which ones don't. You can't think, "Oh, yeah, I just ask the digital god to do my work and then it does." Like, that's not what the technology is. And thinking that will not help, will not help you build it, and it will not help you

  10. 38:43 – 42:11

    Lessons from a $600M Fundraise

    1. NF

      use it.

    2. HS

      If we, like, hone in a little bit more on you, you led all, w- a large part of the latest fundraisers we were chatting before. Um, when you think about the fundraising journey, how was that journey? And there, are there any big lessons from, it was 600 million, no?

    3. NF

      Uh, yeah, so I, I, but I, uh, fundraising at Cohere is a lot, lots of people are involved in it. I by no means led the, led the, uh, the efforts. But I, yeah, I was, I was involved in it. Um, and I, uh, yeah, enjoyed talking to people about the tech and what we're building. Um, yeah, I actually quite, I quite like talking to, to VCs, um, and to pension funds and to people. I think it's f- I love Cohere, I love what we build, and I also like talking. So I like- (laughs) I like talking about both of those-

    4. HS

      Were their any questions very similar?

    5. NF

      Oh, between them?

    6. HS

      Yeah.

    7. NF

      Yeah, actually. That's an interesting question. Yeah, I, I do think, look, the industry's a lot more mature. Like, two years ago, you know, when we were fundraising, or three years ago, a lot of the questions were like, "What is this? Like, what, what, how are you gonna, what? How does that work? What is this?" You know, and so we'd spend more time explaining stuff. Mm, people mostly know how it works and know what it does. And now we can say, "Here's what we're doing for our customers specifically." You know, like, "Here's how RBC is using it. Here's what we're doing with Fujitsu. Here's what LG is doing with it." You know, like, we can talk about those things specifically, um, and that's, that's more interesting.

    8. HS

      How much of the 600 million will be spent on compute?

    9. NF

      Oh, yeah, there's like three components that go into making language models, right? There's, there's talent, is like people, it's um, engineers and researchers. There's compute, um, and there's data. Uh, the importance of those has shifted, and they, yeah, the spend of those has shifted over time.

    10. HS

      Mm-hmm.

    11. NF

      Um, we train very efficiently, right? We train models, uh, we train efficient models, so like our, our model Command-A and the Command-A reasoning model, which we just released, Command-A vision model, which we just released. Um, those, they're all trained to fit on two GPUs. That's like a really important part of our business strategy. It turns out if you go talk to a lot of companies who wanted to deploy models into production, they were bottlenecked on deploying because they don't have enough GPUs.

    12. HS

      Huh.

    13. NF

      Two GPUs turns out to be like a sweet spot between performance and a- uh, performance and cost, and like ha- and actually how many GPUs they had access to. Um, so that means we train very efficiently as well, you know. We have dr- spent orders of magnitude less on creating foundational models than some of the other foundational model companies out there. Truly orders of magnitude less. Um, and I'm very proud of the efficiency of the team and like what they've done with the, you know, the resources that they have. Um, so we, we think about efficiency a lot for ourselves and for our customers, and those two things are, are related. Um, but how much of our funding goes to compute? It shifts over the years, um, but a lot. You know, it's, it's, compute is, is a, it's, it's expensive.

    14. HS

      Ho- how has it shifted over the years?

    15. NF

      I mean, when we first started Cohere, one of the very first things we did, uh, because we had no funding, or, n- we did have funding, but a very small amount of it, um, was we, we spent next to nothing on compute. And we showed that you could train a model by having, like, a bit of a GPU over here and a bit of GPU over here, bit of a GPU over here, and you could link them together, and we published a few papers on that, uh, on training models with, like, the scraps of GPUs in data centers (laughs) , right? That was what we started with. Um, and we showed that you could do that. You can do that. Uh, it's very slow, and it's much easier to just rent a big data center and train the model there. Yeah.

  11. 42:11 – 46:17

    How do Cohere compete with OpenAI and Anthropic’s billions?

    1. NF

    2. HS

      The question that everyone asks is, how do you compete against competitors who have billions and billions of dollars? Do you hate that question, and how do you respond to it?

    3. NF

      No, I don't hate that question. I think, yeah, I think that's a fine question. Um, you know, uh, we've, we've announced funding rounds, um, and you can see that they are smaller than some of the other f- funding rounds out there. Um, yeah, I mean, like, w- we're pretty singularly focused in a way that the other companies who build foundational models, uh, are not, right? Like, we don't have a consumer app. We're not trying to get anybody to spend $200 a month on something for their personal lives.... were singularly focused on working with enterprises and businesses and making sure that they get to production with AI. I have my whole... I, I'm constantly telling people like, "Not AGI, ROI. ROI, not AGI." Yeah, um, and, you know, it turns out there's a lot of work that still needs to get done there. And it turns out there's a lot of companies that tried to go to production with AI by using something-

    4. HS

      But do you, do you think then that OpenAI and Anthropic will just cede enterprise?

    5. NF

      I think, like, right now, you know, those, both those companies have a, are, are pretty cool. They've both made good consumer products. Um, I think where this technology, like, adds the most value is in work for, like, personal reasons. Like, that's where I see this technology being the most useful. Um, I don't know if they'll, if they'll start working on that. Um, I know that making models that work in that environment is pretty different than making a model that works in a consumer environment. In a consumer environment, you can make the biggest model possible. You can have, like, complicated switches to tell you to go to this model or that model because you're h- you're just posting it on a huge amount of GPUs. You can be, like, losing a ton of money on every inference call, but you know, you're getting, you're getting users and something, and so, like, that works. The types of models you have to build to succeed there are different. Um, I know the work you need to do on the interface, like, we know we've, we've announced North, which is our agentic framework. It's privately deployable, customizable, you know, for, uh, knowledge work- workers within an enterprise. It's pretty diff- it looks pretty different than, uh, some of the consumer applications, right? Like, a big one is, like, our model doesn't generate images. Nobody in a workforce is really wanting to generate images as part of their work, for the most part, right? It doesn't... But as a consumer, it's very fun. It's very cool to be like, "Oh, here, give me a picture of this or something." So, we... The types of models we, we train are different, and the interfaces we make are different. Um, I don't know if that's... if they'll be interested in that at some point. I think, like, we stay focused on talking to customers and adding value.

    6. HS

      How do you price?

    7. NF

      Um, that's entirely dependent on what the customer wants to do with us. So, we do have some customers where we give, like make a custom model for them and give them that model.

    8. HS

      So, do you have forward-deployed engineers?

    9. NF

      We do, yeah. Yeah, we have-

    10. HS

      Okay.

    11. NF

      ... forward-deployed engineers. Yeah. And they're very... Yeah. So, they're a crucial component of how we, like, go get a company up and running and into production with us.

    12. HS

      Do you think everyone will have forward-deployed engineers in a future AI world in a way that Palantir has glamorized?

    13. NF

      Yeah. I mean, I think forward, forward-deployed engineers are a good idea, right? Like, you're selling technology to somebody, it makes sense to have some engineers who ha- come and help them get it set up and work with them to, like, make sure it's actually delivering value. You know, I think that's a good idea. Um, I don't know if that's true for every business, you know, but...

    14. HS

      Does FDEs not just allow for poorer technology?

    15. NF

      No.

    16. HS

      And what I mean by that is like-

    17. NF

      No, no, I think that there's this, there's this idea, there's this idea sometimes like, "Oh, yeah. You can just make the thing, and for every business, it'll be, it'll work perfectly and require no engagement." And like, that's not the way a- that's the way some technology works. That's the way a lot of consumer technology works. That's not the way a lot of enterprise technology works, right? Like, there's... You're selling things to people, um, that are, that have to be, like, matched to the way their business is set up, right? And so having engineers go along with it and say, "Cool. Here. Here's, here's a model. Here's what we can do to make sure that that's perfect for you in your specific use case," is helpful.

  12. 46:17 – 56:15

    Do Enterprise Companies Trade at Lower Multiples?

    1. NF

    2. HS

      Given that you sell to enterprises, I, I'm an enterprise investor and-

    3. NF

      Mm-hmm.

    4. HS

      I love enterprise. Revenue quality is much higher, much-

    5. NF

      Mm-hmm.

    6. HS

      ... stickier. But growth is slower-

    7. NF

      Mm-hmm.

    8. HS

      ... because you're working with large enterprises.

    9. NF

      Mm-hmm.

    10. HS

      Do you think you had an, an, have an enterprise discount applied to valuation because of revenue growth being slower because of enterprise?

    11. NF

      Hmm. It's an interesting question. Um, maybe. Yeah. I mean, our, you know, I'm, you know, I'm very proud of what we've done and what we've created.

    12. HS

      What was the price on the last round? It w- it was public. I think it was like 6.7, wasn't it?

    13. NF

      Yeah.

    14. HS

      Yeah.

    15. NF

      Yeah. 6.8.

    16. HS

      6.8?

    17. NF

      Yeah. Yeah, it's- it's- those, these are all staggering numbers. These are all numbers that are impossible for an individual to conceive of. Like they were, were, were so far into that, you know, as a, as a single individual, this is well beyond the realm of what you can reasonably engage with in your life. Um, so yeah.

    18. HS

      Is it?

    19. NF

      For a regular person who grew up working as a cook, y- like, yeah. Right? Like, my first job was burgers, you know? (laughs)

    20. HS

      (sighs) Okay.

    21. NF

      But yeah.

    22. HS

      (laughs)

    23. NF

      But yeah, that's a great... You know, these are all crazy numbers. Yeah.

    24. HS

      Do you care about money?

    25. NF

      Yeah, I certainly. I think everybody cares about money. Um, and everybody's motivated by money.

    26. HS

      With being motivated by money, y- you're an incredibly acquisitive asset, and it's been a very strategically important thing for large players to do. Have you had M&A offers across the journey?

    27. NF

      Oh, yeah. We have at times. Yeah.

    28. HS

      How's the decision-making gone, though? I always want to be in the room. I always picture it-

    29. NF

      It's-

    30. HS

      ... kind of like thundery nights-

  13. 56:15 – 1:05:12

    Should countries fund their own models? Is model sovereignty the future?

    1. HS

      hilarious. Uh, uh, one, one area we haven't covered, which is interesting, is the area of like sovereignty. We're sitting in London now.

    2. NF

      Yeah.

    3. HS

      And, uh, Mistral is in Paris, and oh, I get in so much trouble these days, Nick, 'cause I just... Uh, my mother says that I have Asperger's, but I just call it kind of freedom of speech. Um, but it's just like, we all say that for Mistral, it's like the Europe play, and that's why it's funded and it continues to be funded. Um, do you think that we will see sovereign models and usage because of geography?

    4. NF

      I think this technology is, is a lot like infrastructure, right? Like I think building a... Like having a language model that speaks the, the language of your country is like building infrastructure for the people of your country. Um, so I think that's broadly a good idea. I think the past like 20 years, longer, of technological history has been very defined by Silicon Valley. It's been very defined by, by California and America. And I think a lot of people are not very happy about that, and I think a lot of people are rightfully upset with some of the developments, right? Like I used to be a real technological optimist. Like I used to love the way technology was built and be like, "Oh, it's so exciting," or something. And I, I, I wouldn't describe myself as a technological optimist over the past 10 years.

    5. HS

      Wow, why? What happened to change that?

    6. NF

      Oh. Uh, well, wait, sorry. Let me, let me answer that. First, let me get back to the, the sovereignty thing before I go off on this tangent. Um, so yeah, I think there's a lot of people who are interested in building that infrastructure within their country and having the technology for their economies. Um, and I think just using a model that is built, you know, by China or built within America might not set your com- your country and your economy up as well as having a model that is, uh, understands the context built in that language, in that dialect, in like, you know, has the, has the, the cultural fluency, um, needed to empower the people of the country. So I think that's like a good idea. What that ends up looking like, like I, I don't, I'm not exactly sure, you know.

    7. HS

      Geopolitics obviously influences a lot. Do you think geopolitics has influenced customer decisions around sovereignty of models in the discussions that you see?

    8. NF

      Um, I think us being Canadian is, is an asset, right? I think, um, that's helpful for people.

    9. HS

      Uh, do Canadian companies wanna buy you more?

    10. NF

      Uh, I think coun- companies around the world are interested in talking to us. And in part that's because we're Canadian. Um, look, like, you know, over the past few years, y- y- you know, America has shown that they're willing to like turn off access to tech based on political reasons, right? Like, you know, we've seen, uh, uh, uh, connections between American tech and the American government is like l- less clear as time goes on.

    11. HS

      What does that mean?

    12. NF

      What does that mean?

    13. HS

      Yeah.

    14. NF

      Uh-

    15. HS

      I'm a Brit. It means like Trump in- influences US tech companies?

    16. NF

      Seems to be, yeah. Yeah. Seems to be. Right, I mean, I mean, even it was like last week or something they announced they're taking a 10% stake in Intel, right? Like that's an interesting development. I, I, I'm, I'm not an, I'm not an economist. And I don't know if that's good for the country or not. Like I'm, I'm not... That's... But it is an interesting, um, development, right? Um, so I think there's a lot of coun- companies in Canada and around the world that are interested in working with non-American tech companies.

    17. HS

      Mm-hmm.

    18. NF

      And I would say that's been an asset for us, right? Like that's...

    19. HS

      Do you think governments should fund sovereign models? Like is j- is it a European imperative for us to have Mistral as an asset for Europe?

    20. NF

      I think it's a good idea for countries to have infrastructure within their countries. Like I think it's a good idea for people to have power plants in the country, you know. I like that Canada has several nuclear power plants and has several water power plants. Like, you know, that- that's great. Yeah, I think language models are not that dissimilar from infrastructure.

    21. HS

      Do you think our primary, like, input device will still be a phone in five years time?

    22. NF

      I do think langu- like I know language is gonna be a more important part of it, right? Like I think fundamentally the way we should be interacting with computers is using language for the majority of it. Not all of it. There are times when like language is actually not the best way of interacting with a computer. It's much better to have a graphic user interface if you're, like, doing something. I know like last year there was a f- there was like the Rabbit R1, there was those like, the Humane PIN, and like none of... They didn't really get it right. But I think there's something cool there about like, hey, how do we use...... how do we use a language model to work with a computer better? I haven't seen it done right yet, and I don't know if it will, and I don't know if that's because, um, people don't ... Like, one of the things, like, going back to the technological optimism thing, like, you know, I was really excited when Google Glass came out. I thought that was really cool.

    23. HS

      Yeah, so was I.

    24. NF

      Yeah, and then I-

    25. HS

      I had it- I had it as a profile picture.

    26. NF

      Yeah. Okay. Yeah, and then I got on a bus one time and somebody was wearing a Google Glass, and they were delicate, they were like... And suddenly, everybody saw it immediately. Like, poof. You know, it, it clocked it immediately. And I think, you know, I was really excited about VR for a while, and then I realized I actually don't want to strap a computer to my face. I'm not interested in being disengaged from the world more. I want to be engaged in the world more than I am. I don't, I don't want more things removing me, and I don't think many people do.

    27. HS

      Do you just worry about this is so m- messed ... Like, this is the state of the world in terms of depression, in terms of loneliness. You know, the biggest p- pandemic, epidemic, whatever we want to ... I, I never know the difference between pandemic and epidemic.

    28. NF

      Wow, this is such a podcast. I haven't done many podcasts. This is the most podcast podcast ever. (laughs)

    29. HS

      I'm just, I, I, again, I can just do it.

    30. NF

      No.

  14. 1:05:12 – 1:14:14

    Why has Sam Altman actually done a disservice to AI?

    1. NF

      young.

    2. HS

      Okay, we're gonna do a quick fire.

    3. NF

      Okay.

    4. HS

      Yeah? So, if you were Sam Altman today, what would you be doing that he's not doing?

    5. NF

      I don't think Sam Altman has done a service to the world by talking about how close AGI is. I think he has made several predictions now that are wrong, um, and that were obviously wrong at the time he made them.

    6. HS

      Which one is most prescient?

    7. NF

      Oh, that, like, AI is gonna kill the whole world in two years. Or like, he's made, you know, he's made allusions to things, like, he did a world tour where he spoke to every major leader the world over to tell them, "Hey, this technology is gonna, you know, is, poses an existential threat." And I think that was academically disingenuous, and I think did a disservice to the technology he loves, you know?

    8. HS

      Do you not see a correlation between the words that one has with regards to the future of AGI and AI, and their requirements for funding?

    9. NF

      I don't, I don't know what it is.

    10. HS

      Do you see what I mean, though?

    11. NF

      I, I see what you mean, yeah. Yeah, yeah.

    12. HS

      Which is like Demis and Zuck for a long time do not need funding-

    13. NF

      Yeah.

    14. HS

      ... and they are much more balanced, neutral.

    15. NF

      Yeah.

    16. HS

      And then other people who do need funding have to be much more provocative and out there because they need your fucking dollars.

    17. NF

      Yeah. Yeah, I don't know if that was the strategy. Uh, uh, the correlation you're pointing at exists. Um, I would say that, you know, we're a, we're a venture capital funded company and we need funding, um, and we don't say that. <|agent|><|en|>

    18. HS

      What worries me, though, actually is even the rhetoric from your Demis and your Zuck-

    19. NF

      Yeah.

    20. HS

      ... has changed. Even their aggression towards the changes that are coming has, has, has flipped-

    21. NF

      Yeah.

    22. HS

      ... which does make me worry.

    23. NF

      Makes you worry because ... For what reason?

    24. HS

      For the reason that actually we are far closer than we think-

    25. NF

      Mm.

    26. HS

      ... to very material shifts in labor patterns, workforce behaviors.

    27. NF

      Mm-hmm.

    28. HS

      When even Zuck, who does not need the money from anyone, or Demis, who doesn't need it from anyone, is going, "Oh, shit. The changes are real."

    29. NF

      Well, yeah, well, there are some real changes, right? Like, I'm not, I don't wanna dem- You know, this technology, fundamentally transformative, the same way-... the personal computer, fundamentally transformative, the Industrial Revolution, steam engines, the printing press. Those are all big technologies. Like, I really think that. Um, so there's, there's very, there's tons of legitimate things to talk about, and I'm glad people are talking about them. Uh, there's also a whole lot of not legitimate things to talk about that those people are spending their time on.

    30. HS

      What is your founder ritual after closing each round? (laughs)

Episode duration: 1:14:38

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode Sw2chzwWLbQ

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome