00:00SenseTime, when it first started, it was really well known for the computer vision, the facial recognition, smart city business.
00:07That was almost the entirety of your business.
00:09But more than a decade on, you have pivoted away from that.
00:13Tell us about how your company has evolved since then.
00:18SenseTime has been pondering a core question.
00:20Where will its intelligence come from in the next stage?
00:23Our answer lies in how to effectively utilize the cross-connections between visual information, what the eyes can see, and
00:31textual information,
00:32and how to use cloud connections to form a paradigm we call native multimodal, thereby raising the upper limit of
00:40our intelligence.
00:40So in recent years, we've actually moved from visual to what we call native multimodal applications.
00:46You talked about your core model, SenseCore and SenseNova.
00:51If I were just to go out on the street and ask people about what models they're familiar with,
00:56people will probably talk about Alibaba's Q1 or Tencent or ByteDance's model.
01:00I think not many people are aware of SenseNova.
01:03How do you differentiate yourself from all the other big players out there now?
01:12I think a crucial point here is a choice of path.
01:15Our understanding has always been that if we can use the boundaries between language and text,
01:20the connections between language and vision, images and videos, to improve our model's capabilities,
01:26then it will reach a new upper limit.
01:30SenseTime is committed to using our strengths, that is, vision,
01:34to raise the upper limit of our model's capabilities in the next stage.
01:39Therefore, I hope that our SenseNova can achieve some breakthroughs within this upper limit of intelligence.
01:45What's your profitability looking like based on your current pricing model?
01:51Because of the current scarcity of computing resources and the need for optimization,
01:56even with significant resource investment, the current pricing and profit margins are very high.
02:01Therefore, it can still be considered a very profitable software business.
02:05You talk about the strength of your computational capabilities.
02:09I know when you started back in 2014, I believe,
02:13the export controls are not in place yet, and you were relying on NVIDIA's chips.
02:17So since the export controls, how have you diversified your supply chain?
02:22Are you still relying on NVIDIA's chips right now?
02:29We've long known how to better connect larger clusters to solve these problems.
02:34With the changing geopolitical landscape, we've indeed done a lot of promotion of this hybrid business model.
02:40We have many domestic GPU partners, and we have accompanied them on their development journey,
02:45exploring how to compensate with better software.
02:49What is the availability of these domestic GPUs now?
02:52If you want to buy them, do you have to wait a long time?
02:55Is there a supply constraint right now?
02:57To a certain extent, some chips do have supply cycles.
03:01But because of the many partners we have, the overall industry development remains relatively healthy.
03:06I do want to ask about the geopolitical risk that you're facing,
03:09because you've been placed on the Commerce Department entity list,
03:12so that restricts your access to U.S. technology.
03:16How have you been able to navigate that?
03:18And what's the journey been like since then?
03:24Many years ago, we actually closed our offices in the U.S.
03:28Firstly, we felt that geographical factors were unavoidable.
03:31For our company, the key is to do what we can do best.
03:34We do believe that making such differentiation between tech companies
03:38based on geographical proximity is a biased approach and not a very correct path to take.
03:44But for us, we will focus on doing what we are capable of.
03:47For the company, the company will be able to do well.
03:49We can do well.
Comments