Loading...
bot.auto
Speaker 1
A technical issue as always. You see, this is actually going to be the thing that I'm going to talk about. We always talk about, when the technology works in the lab, it works really well. Yeah, there we go. But when it works in reality, there's always problems here and there. But the key problem for that, you know, in fact, I think this is a great example to begin with. The key problem for that is who is going to pay for this real world problems, right? And that's why I think the real world problem is really killing this business. And we certainly we are here to defend the business, not making to be killed. So that's why I'm here to talk about the new business model in autonomous ready. Next page, please. I'm not sure if this works. Okay, there we go. What's a market pain point? Capacity. So for autonomous trucking, I think the biggest incentive for me to do autonomous Trucking is because we don't have capacity. The entire US market is having 80,000 truck driver shortage, and we need more drivers. And in fact, we need more capacity. But on the other hand, does a tech can offer demos? So I think this is really the problem that happened in the past several years, all that the entire industry is providing this demo, and nobody's providing the real capacity, like a meaningful capacity that can solve the problem. So in the end of 2021, after half a year of literally, like, no spare time, or like, I only have seven calendar days off, my last company, we delivered the driver out demo. But again, it's still a demo. And after the demo, I we were celebrating amongst each other, and I was very happy for about 50 minutes. Why? Because I know that this is not the finish line. The finish line is that we need to have a plan. We need to have a plan that market needs, that is to answer the capacity problem, and this is legit come to my second point, a legit business makes profit, not revenue. In the past, it's very easy to make revenue if you subsidize your autonomous driving fleet. You can make it work. You can just offer it every day, and you can probably say we're commercializing autonomous driving, but in the end, you're losing money in every mile, and that is not something that's sustainable, and the market needs something better. So I know there are several different business models. People are talking about SaaS software as a service, or in the autonomous driving world, some other people saying hardware as a service, as has but for my new company, bought auto, what we wanted to do is transportation as a service, because we wanted to answer the question of where the capacity is. And yeah, you can, as you can imagine, if I'm making this transportation as a service, in front of many investors, first question is that, oh, you're asset heavy. Asset heavy is actually a taboo somehow in the Silicon Valley, but I'm glad I'm here in the east coast to talk about it. My opinion is that I think asset heavy, asset light may be a pending question, but operation heavy is something that we have to admit that for autonomous Trucking, we have to go through the operation heavy as a way to deliver, to provide capacity. I think in the end, whatever approach it provides, the capacity wins. We have to bite the bullet for admitting that we are an operation heavy business. Because you know what, if your operation light, someone else will be operation heavy, and who will be the most deserving party to be operation heavy? I believe the autonomous driving companies who build this very complicated technology should be the one that are asset sorry, Operation heavy. And of course, there will be some other ways that, for example, we can can use more clever finances to make sure that we're asset light or or alleviate our asset pressure. But in the end, Operation heavy is really the way to go, the only way to go. And when it comes to auto driving, in order to make a business that makes money, makes profit instead of making revenue, we need to think about three things. This is what I think about the value of a product. Thinking about autonomous driving, especially for autonomous trucking, the drivers cost. Yes, by removing the drivers, we save some money. Therefore, I really don't feel that any model that says like, we're facilitating the driver, we're making the drivers, like, happier, easier, this kind of level two system, they can save money because the driver is still there, the driver. You cannot argue with the driver, because the presence of the level, level to assist driving system, I'm going to pay you less. The driver wouldn't allow that. So in order to really create value, the driver cost must be removed, and that's the first part of the value. But by removing the drivers costs, there are two other things that can be added, that will be added on. One is the operating cost, just like the you know, in the beginning of this presentation, that's a great example. It's operating costs can happen everywhere. I think people are talking too much about corner cases, like if you see an alligator in the middle of the road, how do you interact with alligator? Politely, for example. But the problem, the real problem that I have in my my my eight years of autonomous driving, is that the real problem is always trivial, but it's always there, and it buys you every day. I'm making an example. I consider myself a pretty healthy person, but if using a healthy standard, if I'm taking a physical exam, if you take the standard of sending me to the International Space Station, no, I'm not at all healthy. I probably have 25 metrics or index or being abnormal. And this is like autonomous driving. If you're thinking about a human, operable driver, it's still someone behind the wheel for driving the vehicle. A human can actually do a lot of things, but when it comes to autonomy, we're actually asking for an absolute reliable system that can always guarantee functionality whenever a single point of failure has occurred. So in order to operate in this super complicated system, we have a lot of issues. So that's operating cost. And lastly is 
customer acquisition costs. If you're building a software as a service or hardware as a service, then you also need to sell it to other companies. Maybe the first 101st, 100 items can be easy to sell, because they're always Innovation Fund, innovation department of a big company, but to solid 10s of 1000s of items, no. And I think the all the industry needs to focus on one thing. Years ago, we talked about MPI, but we all know that MPI is like your MPI may vary type of thing, because every company can have their own determination of what is the safety related intervention, and many of the interventions are actually discounted by the company themselves, who reported data to, for example, the California DMV about the FBI. Therefore, we need a new metric, and the metric is the consensus of cost per mile, and we can make it work, because the cost per mile is actually a very auditable accounting term. Basically, we need a count. We need a countdown billboard for all saying, like, how many dollars have you spent in terms of operating this vehicle, in terms of interfacing with the hardware, in terms of maintaining, for example, maintaining a map, maintaining the remote operation. And, of course, the vehicles cost, the hardware cost, the camera costs, lighter costs everything. Putting together the totality of that is what we call cost per mile. And we believe this is the time for us to re establish a new, new focus for all of the company to think about and maybe to report that is the CPM metric. And then lastly, I wanted to talk about the check bubble. We know that scaling law does not apply for time driving, because this is edge computing. This is 
not something we can ask for very large scale, parallelized servers to be in the vehicle. And also, there are a lot of things about end to end learning. I wanted to give you this so I just feel that this year, end to end level, at least, starting in this quarter, end to end learning is not the most term in the world. People are talking about world model, but I guarantee you, next quarter, the world model will become obsolete as well. So what I what I mean here is that I think there's a so called technology, techno, technological ideologization term happening on people are talking about concepts, using these concepts, considering them as the silver bullet to solve the problem. But in the end, what really we should be focusing on is focusing on the objective, rather than focusing on the approach. Because, you know, in every quarter, you will have a new approach, but objective is something will last longer. That is the CPM, okay, think I'm still one minute.
+2
customer acquisition costs. If you're building a software as a service or hardware as a service, then you also need to sell it to other companies. Maybe the first 101st, 100 items can be easy to sell, because they're always Innovation Fund, innovation department of a big company, but to solid 10s of 1000s of items, no. And I think the all the industry needs to focus on one thing. Years ago, we talked about MPI, but we all know that MPI is like your MPI may vary type of thing, because every company can have their own determination of what is the safety related intervention, and many of the interventions are actually discounted by the company themselves, who reported data to, for example, the California DMV about the FBI. Therefore, we need a new metric, and the metric is the consensus of cost per mile, and we can make it work, because the cost per mile is actually a very auditable accounting term. Basically, we need a count. We need a countdown billboard for all saying, like, how many dollars have you spent in terms of operating this vehicle, in terms of interfacing with the hardware, in terms of maintaining, for example, maintaining a map, maintaining the remote operation. And, of course, the vehicles cost, the hardware cost, the camera costs, lighter costs everything. Putting together the totality of that is what we call cost per mile. And we believe this is the time for us to re establish a new, new focus for all of the company to think about and maybe to report that is the CPM metric. And then lastly, I wanted to talk about the check bubble. We know that scaling law does not apply for time driving, because this is edge computing. This is +2
not something we can ask for very large scale, parallelized servers to be in the vehicle. And also, there are a lot of things about end to end learning. I wanted to give you this so I just feel that this year, end to end level, at least, starting in this quarter, end to end learning is not the most term in the world. People are talking about world model, but I guarantee you, next quarter, the world model will become obsolete as well. So what I what I mean here is that I think there's a so called technology, techno, technological ideologization term happening on people are talking about concepts, using these concepts, considering them as the silver bullet to solve the problem. But in the end, what really we should be focusing on is focusing on the objective, rather than focusing on the approach. Because, you know, in every quarter, you will have a new approach, but objective is something will last longer. That is the CPM, okay, think I'm still one minute. Speaker 2
Thank you so much. Xiao di I think that online side presented such a like a large touch the humorous style but underneath it, there's really quite deeper seriously in this right talking about realistic operation of autonomous track and business, thank you for offering that to perspective to us right, and that's why I invite Professor Henri, Leo, who's the director of MCD and Professor, University.