🤝 Meet Nick Frosst. He's the co-founder of Cohere, the Toronto-based AI company valued at US$7 billion that's positioning itself as Canada's champion in the global large language model race. Before starting Cohere, Frosst was Geoffrey Hinton's first hire at Google Brain's Toronto lab. And for good measure, he's also the lead vocalist of Good Kid, an indie rock band that has played Lollapalooza. We chatted with Nick about why he stayed in Canada to build an AI company, whether sovereign AI is realistic for smaller countries, and why he thinks AI will eventually be boring — in the best way.
What was the moment for you when you realized that AI was going to be a big deal? And then after that, why did you decide to start Cohere rather than going to the U.S. and working for a huge paycheque at one of the frontier labs?
I think one of the big moments was early on when Geoff Hinton released his ImageNet research. At the time (around 2012), it was so mind-blowing that I thought I was late to the AI boom. Of course, we now know that AI is capable of many, many orders of magnitude more than that. We started Cohere in 2019, after my co-founder Aidan Gomez worked on the seminal Transformer paper, because we knew enterprises had a lot to gain from this technology and we wanted to bring that capability and power to them.
How does Cohere fit into AI's competitive landscape? How do you position yourself with regards to the big American AI players?
Cohere is the only major AI provider exclusively focused on the enterprise and government market. We focus on security, privacy, and data control, and design our models and our agentic AI platform called North to run efficiently and privately, even air-gapped. This approach is especially valuable for enterprises in regulated industries and in countries that are looking for sovereign and resilient technology systems.
Does Cohere do the same sort of R&D as these American companies when it comes to training more sophisticated general-purpose models, or do you take a different approach when it comes to product development?
I'd consider our approach more practical and focused on real-world problems. Things like strong performance in many languages, fluency in industry-specific terms and problems, and customization are important for the global teams at the large enterprises and public-sector organizations we work with. We prioritize making our tech do what companies need, and making it efficient to serve and highly secure so they can use it for complex, sensitive work.
We read a lot of stories about AI businesses bleeding unfathomably large amounts of cash, but Cohere doesn't seem to be in that situation. Why is that?
Because we were clear about our market from day one. We knew enterprises could benefit from this technology, so we've been able to build really great models and products that are effective and efficient. Others are burning cash in pursuit of AGI, but we're building technology that works today.
With American and Chinese labs throwing so much money and compute at this technology, is it realistic for smaller countries like Canada to have their own "sovereign" AI capabilities? Or are we destined to be dependent on technology built and owned elsewhere?
It's realistic. I think Canada is already set to be a luminary of homegrown, sovereign AI. The country was an early leader in AI and still home to some of the greatest talent and research programs focused on this technology, and it's important we maintain that early competitive edge. I'm proud we're playing our part at Cohere, and proud to see several of Canada's most prominent institutions like RBC and Bell doing the same by using Cohere's made-in-Canada technology.
The federal government seems enthusiastic about building an AI industry here. What's the most important thing you think they could do to support that goal?
Continuing to champion innovation — through access to compute, through talent retention (helping ensure the people who build this technology in Canada have opportunities to work in the industry in Canada), and through ongoing investment in the ecosystem.
Most people I talk to about AI are worried about it, sitting somewhere on the spectrum between "it's going to replace me at work" to "it's going to lead to the end of the species." Any thoughts for them, reassuring or otherwise?
If the industry gets it right, AI will be boring. In the best way. It should fade into the background, power the technology and tools we'll use every day, and make work more enjoyable — like what we've seen with technologies that have become invisible enablers, like cloud technology or GPS. This will be transformative, but nowhere near the same way doomsday headlines claim.
Do you still have time for Good Kid? And on a related note, what album/artist have you been really into lately?
Yes, we're going on tour soon! I have been listening to a lot of Lia Pappas-Kemps recently. I think she is one of the best songwriters in Toronto right now.