Skip to main content
Nick Frosst

Nick Frosst

Computer scientist and musician

Nov 18, 2026

5:03
Why?
5:04
Well, A, we have the evidence to back that up for a little bit, right? Like we have had a few years of large language models being deployed.
5:11
And in the areas in which they're deployed most, we've actually seen an increase.
5:14
So coding is a good example.
5:16
Programmers really like to use large language models.
5:19
And I think there's a few reasons for that.
5:21
One, large language models are really good at coding.
speaker_5SOUNDBITE_SPEAKER
8:12
The Wall Street Journal reports that AI found a bug that puts up to 70% of the world's websites at risk.
10:51
Have you seen those kinds of things at Cohere?
10:55
No, and that's because we deploy them for a very different reason.
10:58
We deploy our models into a company's environment securely so it can operate on their data without sharing that to people, connected into the right data source so that somebody at a company can actually ask a question about their company in specific and have the correct answers.
11:14
And we deploy them efficiently so that they actually get ROI.
11:18
All of that means that people are not coming to our language models to talk about their feelings.
11:23
They're coming to our language models to help them do their work.
11:25
So we don't have those use cases at all, nor do I think those are particularly good uses of language models.
12:53
Here's Nick Frosst again.
7:17
Like, I guess with OpenAI, do they get a Joolst because of the consumer slant? And you guys are
7:26
B2B, and it's trickier to raise money.
7:29
Well, we're only B2B.
7:31
So that means that we're not interested in making the biggest model possible.
7:35
We're interested in making a model that is efficient to deploy.
7:38
And that when you deploy it privately into a customer's environment...
7:42
it adds enough value to what they're doing that it warrants paying for it.

34 MINS LATER

42:09
Totally.
13:04
And I think there are a generation of people growing up now all their experience with technology, all of their digital experiences are terms of service controlled.
13:13
Yeah, no, I mean, I definitely grew up as a technological optimist.
13:18
And I grew up hanging out in the hack lab of Kensington in a little hacker collective that was running there.
13:24
And I was hanging out there every now and again.
13:26
was really into mesh networks and, like, really thought that through the ubiquity of technology, people would get more independence.
13:35
And that...
13:37
I'm very disillusioned with that now.

30 MINS LATER

43:53
Anyone who criticizes that is the enemy.
26:58
Is this, like, another problem for creativity that, like, maybe we don't want data, maybe we just want the USS Enterprise to talk to, and, like, our little hologrammatic fantasies? I mean, isn't that another sort of threat to creativity?
27:12
Well, let me pick up on a few pieces.
27:13
Like, yeah, I don't want data.
27:15
I'm not interested in making artificial people.
27:17
I'm interested in people.
27:19
It's not obvious to me that society wants or needs data.
27:25
We need and want better computers.
28:35
[laughs]
9:53
[laughs]
9:54
Uh [laughs], and, and I really only worked with him closely when I was at Google.
9:58
So I was a, a research, a researcher at Google Brain, and I worked with him at, for...
10:04
I, I was, uh, working out of Waterloo for a little bit, and then I found out that he was working in Toronto, and then we started to work together, uh, and then I helped start up the Toronto Brain Group with him, uh, and worked there for a few years.
10:14
And it's during those three years, four years that I learned most of what I know about research and machine learning and neural nets, uh, and I learned it from him.
10:23
So I learned a huge amount from him, but I, I'm, I'm not hi- I don't have a PhD or a master's or anything.
10:27
As for his contribution, uh, it really can't be understated.

13 MINS LATER

23:40
Okay.
Claire Duffy
Claire DuffyCORRESPONDENT
14:32
Nick explains why he thinks the industry should not be focusing so much on AGI.
14:37
There are ways in which language models can be detrimental to society.
14:41
Misinformation is a big one.
14:43
Things like trust and safety, data privacy, are much better conversations to be having for policymakers and researchers.
18:12
What is the company's mission? If AGI is not your North Star, what are you building?
18:18
Yeah, I mean, the mantra we say in Cohere a lot is like ROI, not AGI, which really just means like we want to build something useful with the technology today.
18:31
Like large language models can be doing so much more than they are doing now, but they are blocked by things like data privacy.
18:39
They're blocked by cost.
18:41
They're blocked by retention.
18:43
They're blocked by performance.
18:44
They're blocked by not having good enough interfaces.

6 MINS LATER

24:58
And I'm curious for your take on this, especially as somebody who is working on automating processes within companies, do you think that we could see a jobs apocalypse because of this technology?
15:16
Oh my God.
15:17
You know, when, when, when, (laughs) when the world was new.
15:19
A- a- any task you wanted to do with a neural net was b- The best neural net you were gonna get was training a model on that task.
15:26
So, if you wanted to make a task that was gonna, like...
15:28
You know, a neural net that was gonna identify pictures of cats and tell you how many cats were in an image, you were best suited to train a model on identifying pictures of cats.
15:37
That was like, the best.
15:38
And the, the world of machine learning in the first half of that decade was all about, here's a problem, make a dataset, train a new model from scratch, or maybe take like, SIFT features or maybe take like, whatever, some base model, but pretty much fine tune...

17 MINS LATER

33:03
You wouldn't say, "Hey, make it a funny tone," or, "Hey, add in a light personalized style that also is sincere."

We value your privacy

We use cookies to understand how you use our platform and to improve your experience. Click “Accept All” to consent, or “Decline non-essential” to opt out of non-essential cookies. Read our Privacy Policy.