00:00And this visual AI, I mean, this is not something that's been a priority, but you say it's time to
00:05get into it.
00:05Yeah, that's right. Visual AI has kind of been underserved.
00:09I would say there's so much going on in coding and math, and the models are advancing very quickly there.
00:15But in spite of that, visual reasoning is still working at kind of the level of a preschooler.
00:20So these, I would say, elementary school kids, they can do things.
00:26Three-year-olds, you say?
00:27Yeah, three or four or so.
00:29And just to be sure, when you're talking about visuals, we're talking about, you know, sectors like architecture.
00:34Yeah, architecture and engineering, you can imagine like CAD, CAM, understanding these like technical diagrams.
00:41Also, there's areas like construction, agriculture.
00:44A lot of actual, people say that white-collar work is going to go away, but a lot of that
00:49is actually visual work.
00:51Work on diagrams or work like looking at crops and understanding what's going on with some disease.
00:57But why is it important, and how would visual AI help expedite perhaps the whole process?
01:04Yeah, so a lot of these current use cases are being covered with traditional AI methods like CNNs or traditional
01:12kind of like visual classifiers, object recognition.
01:14So they don't have the capabilities of, say, reasoning, the textual reasoning that these current models have.
01:21So we are developing models that have visual reasoning, that can reason in the visual space, and bring in these
01:28reasoning capabilities to these traditional, more dumb CNN-type models.
01:33And really, what is your edge in this?
01:35Is it about proprietary data, or is it about training?
01:39What is it that you bring to the table?
01:41Yeah, great question for a startup.
01:43Well, to begin with, we have a great world-class team.
01:47So our team has decades of experience on frontier models.
01:50And how big is your team?
01:51From what I gather, you've had 12 people before?
01:55Yeah, now we're 17 or so people.
01:58We're growing rapidly with a ton of experience.
02:02So we're going to essentially build, we're essentially building proprietary data sets.
02:07We're sourcing them.
02:08We're generating them ourselves.
02:09But we're also making changes to the model architecture, to the RR algorithms.
02:14So we're building a specialized model for visual reasoning all the way through.
02:18It is a pretty niche area.
02:20I'm just wondering whether you see competition coming up.
02:22I mean, as you gain more customers, chances are people see the potential and they'll come on board as well.
02:28How do you view the competition?
02:30How difficult is it to go into this field?
02:33Yeah, the AI market is very competitive.
02:35As you know, there's open-source models and closed-source models.
02:38But so for us, it's very important to move very quickly.
02:41So we founded this company in December.
02:43We're already 17 or so people.
02:45And we plan to have a model that's state-of-the-art very soon.
02:50And that allows us to stay at the cutting edge.
02:53At the same time, we're talking to potential customers, potential partners, who can work with us closely to develop better
03:00and better models over time.
03:02Still in talks with potential customers.
03:03Have you secured any customers?
03:05And who exactly are you in touch with?
03:07So we're currently in touch with some video companies, some entertainment companies, and some robotics companies.
03:15Here in China?
03:16Where do they be?
03:17Oh, no, these are in the U.S., in California.
03:20And there we're talking to them about how we can do pilots with them, how we can do data exchanges,
03:26so we can really build models that are useful for their visual tasks.
03:30How soon do you think you might secure these customers?
03:32What are the indications so far?
03:35I think the indications so far are pretty positive.
03:37There's a lot of legal paperwork, especially for a new company, that we are working our way, waiting our way
03:44through.
03:45So hopefully in the next few weeks, we can have an agreement.
03:49Andrew, given that it is such a niche space and just wondering how you are in terms of talent recruitment,
03:57where are you with that?
03:58You're from 12 to 17, and as you look to grow, I imagine you need more manpower.
04:04Yes, so recruitment, I've found, is very different from being at a company like DeepMind, which I was at for
04:1012 years.
04:10Now it's a lot about selling the vision, selling the research vision, and I've had all that time of research
04:18experience, so people know my experience in the field, and they come because of that.
04:23But also I think there's a lot of people who are getting a bit tired of all the focus on
04:28coding and math, and they believe that AI can be more than just coding and math.
04:33And we should be making much bigger steps towards getting things at the level of human capabilities, and multimodal visual
04:41reasoning is one of those where there's so much potential, so much more work to be done.
04:45So has it been challenging getting the right talent on board?
04:48I'd say it's a mix. We've had a lot of success in the last few weeks after we came out
04:55of stealth.
04:55So we came out of stealth like a few weeks ago with $55 million in funding from some partners, Striker
05:03Ventures, Mendo Ventures, and Altimeter.
05:05And that has really helped us accelerate.
05:09So people look at that and they see that the conviction these well-known investors have in us, and that
05:13really helps bring them on board.
05:14I also understand that you've been approached for new funding, yes or no?
05:19Yes. Yes, I have.
05:21Okay, perfect.
Comments