Invisible Technologies on Future of AI Adoption

Bloomberg

Watch Invisible Technologies on Future of AI Adoption - Bloomberg on Dailymotion

Transcript

00:00We've been talking about circular deals, there's concern about how investments have been plowing

00:05through the AI ecosystem, and talk of froth, a bubble. What are you seeing in terms of demand?

00:12Yeah, so it's interesting. I think if you ask the question of a bubble, you can think about it two

00:15ways, the technology itself or valuations, and I think they are separate questions. I think if you

00:21look at the technology itself, there's no doubt over the last four years we've seen an absolute

00:24step change in model performance, the usage of those models in a lot of different areas.

00:29So I think the technology paradigm shift is very real. You then get to the question of

00:33valuations, and I think that's actually a more complicated question. You can definitely say

00:38valuations are high, but I'll just give you a couple stats that I think about a lot.

00:41You have basically the big tech companies whose earnings are very material, so it's hard to argue

00:45that in their cases it's a bubble given that performance. And then if you actually look at

00:49the AI space, it's something like three companies are 30% of the total funding, and about 100 companies

00:56are 70% of the total funding. So what you really have here is about 100 private companies with a

01:02lot more demand than capital access at the moment, and none of them really are public. And I think

01:08that is the interesting thing. This is not a retail dynamic. This is not a situation where, I mean,

01:12an interesting fact I saw the other day is that public software multiples today are one half what

01:17they were in 2021 today. And actually, if you look at the number of IPOs, it's still only about

01:22two-thirds of what it was, or actually a third of what it was in 2021. So it's not a public markets

01:26dynamic. I think what you have is capital scarcity around a smaller number of AI companies.

01:30Not for you. I mean, you've managed to raise, what, 100 million back in September, and now you're

01:35telling me more than 200 million?

01:37130. Yeah, we did get, after that, we had a lot more interest in, so we actually did it

01:41upsizing the last month.

01:42What are you doing with that capital?

01:44We are investing all of it in our core technology software platforms. So I think, you know, if you think

01:48about the history of our business, we've trained all the large language models. We're involved in

01:51statistical testing and what's called reinforced learning human feedback, where you basically

01:55test and validate the large language models. What we found is that that cycle is where the

02:00enterprise is going over the next 10 years. So if you're a big enterprise company and you're

02:03looking to deploy a model, let's say, for credit memo generation, let's say you have a bunch of

02:08credits and you want to generate 10-page documents to do reviews of those credits, you need to do all

02:14the same stuff that we've been doing kind of at the Model Builders over the last couple of years,

02:17where you create a rubric for what does the output look good, you score it, you test it, you train it.

02:22And so a lot of our capital now is going to enterprise-focused and training all these similar models.

02:27How difficult is that to have the adoption at the enterprise level, and how does it differ from consumers?

02:32It is difficult, and I think this is actually, you asked about kind of the question of a bubble,

02:36if you look at actually the uptake so far. So KPMG had this report, interestingly, that 60% of people

02:43worldwide, I think it's across a panel of 80 countries, use Gen AI once a month. So you have enormous adoption

02:50on the consumer side. On the flip side, on the enterprise side, MIT released a study that 5% of AI projects

02:56make it into production right now. And it's hard. I think you need to have clean data, you need to have

03:01testing and validation, you need to know that, you know, I always give the answer, if you had to generate

03:051,000 memos, and you had to bet, you know, your annual bonus that those memos would be right,

03:11how would you know? How would you train that? How would you test it? How would you get comfortable

03:14that these memos are right? And I think that's really the journey of, the enterprise is looking

03:19for extreme precision and kind of human equivalence in output. And I think that's harder than people

03:24expect it. When it comes to data, I mean, this idea of rubbish in, rubbish out, right? So the cleanliness,

03:29to your point, is important. But what do you see in terms of the scale of the market in data

03:35labeling? I think some estimates put it at 2 billion last year, quadrupling by 2030. And how

03:41will your company play in that ecosystem? Yeah, you know, I would say we're in the first inning of

03:47that. I think if you extrapolate what I just said to the enterprise, you actually have all of the big

03:51enterprise companies over the next 10 years as they start to label things like contact center

03:57conversations or video image data. Basically, the enterprise has barely started in the context of

04:03kind of data cleanliness. The other thing I'd say is we're talking about unstructured data in the

04:07context you just brought up. But a lot of the pain for the enterprise has actually been structured

04:11data as well. So like if you think about I have 12 ERP systems, a CRM system, I have this huge

04:16fragmented landscape of data that underpins my AI. And I've got to get that all organized before I can do

04:22any AI. My favorite joke is that when good AI meets bad data, the data usually wins. And I think

04:28that is the story. When it comes to enterprise segments, where do you see the most opportunity?

04:34It's an interesting question. So I think the investment so far in the enterprise has predominantly

04:38been in the sectors you'd expect. So financial services and healthcare. I mean, they tend to have

04:42the largest technology organizations, so they've put the most capital into it so far. But there are a lot

04:47of really interesting spaces where we are seeing a lot of opportunity. To give you a few examples,

04:51agriculture. So we think about like if you think about crop yields and herd safety and all the

04:56different dynamics of, you know, there's a lot of things you can do with video and image data in

05:00agriculture, for example. I actually look at some of the areas we've played in. So Swiss gear,

05:06like Swiss Army, the luggages, we've done a lot of analytics for them around inventory forecasting.

05:10So most consumer goods companies, for example, are still pretty basic in the information they bring

05:14together. And we saw a lot of opportunity there. My favorite one is sports. So I actually think

05:19everyone loves sports. If you think about the potential to like, look at the spatial movement

05:23patterns of players and who you might want to draft. We did that for the Charlotte Hornets.

05:27That's been another big area. Super quickly, how do you differentiate with competitors,

05:31scale, surge AI? Yeah, I think, look, there's kind of three to four of us or so that have done this

05:36for the last five years in the AI training space with model builders. I think the difference for us is

05:40that's really only half or a third of our business. A lot of what we're doing also is on the enterprise side.

05:44I think we're a tech-centric platform. And I think on the enterprise side where we serve,

05:50I think, eight different sectors now, I think that's a pretty different part of our business

05:54than the rest of those players.

Category

Transcript

Be the first to comment

Recommended