S7E7 [EN]科技周会-Groq, AR valuation, DoorDash history and more
23:03 Share

So speaking of like small companies that has a rapid development and edge against all the competitors, recently we've also heard a lot of news about Grok. So just to clarify, this is not the AI model by Twitter or X. This is a hardware company that's specializing in manufacturing ARPUs.

So maybe Tracy, if you have any more information, please share right now, maybe. Yeah, so I also don't really understand the benefits of having a very fast response. So currently, I don't mind if I wait for a few seconds and the large language model gives me an accurate response instead of a fast but random one. I mean, of course, the faster the better.

So there is some scientific research. I think one is a paper published by my advisor. So they studied the latency of the response and how does that impact the user. And their user is data analysts. And they find that even a delay as short as 500 milliseconds, which is 0.5 second,

it will dramatically influence the user experience and data analysts lose engagement. So, since this is also the case for other interactions like browsing or interacting with the chatbot. So what do you think? Do you care about speed? I think speed has some like

speed is definitely important in some application scenarios. For example, the Rabbit R1, the AI hardware and some others like the AI ping released a few weeks ago, I think those since it's a hardware product. So you definitely don't want a very obvious latency on top of the hardware user experience. So I can see in that case,

speed or like the short latency could be very important in terms of user experience but from what i read is that the rocket seems to get rid of a lot of the like

inference or some of the more human-like processing in the modeling. So that's how it gets so much faster. So I definitely need to do more research on that, but it seems like

Maybe Grok is faster because it's kind of like deviating towards the traditional language model, but leaning towards more like the, I don't know, like the traditional Google search type of behavior.

I guess it really depends on the user's preference if you want a more natural language-like experience or a faster machine-like experience. Since you mentioned Rabbit, do you remember why they are so fast? I think they haven't said they are faster than the competitors,

From what I read, one of the ways they can ensure that at least for all the devices you are getting a pretty consistent experience is that for each hardware terminal you would have a dedicated virtual machine or a cloud machine running individually for that. So it's like a one-on-one cloud experience in that way.

that might explain like why uh the how the hardware slash software experience could be consistent for rapid it's not like everything is connected to a one single central server okay interesting yeah speaking of like uh hardwares uh i think uh recently there have been a lot of like uh

dramas and news regarding AR devices. As everyone probably has heard, whether you are like a tech enthusiast or not, Apple Vision Pro seems to catch everyone's attention, but that's already like a month ago. So the more like recent updates I've read is that people are starting to like return or at least many people are returning the

Neurovision Pro purchases. Some of the reasons I read are like well first of all it's just way too expensive like price is probably the most important factor that people consider when making a purchase decision and I've also read something like it's still like too heavy

wearing it like long time every day I've seen some like drawings online like what people have proposed like a vision air so it'd be more like a glass or like a goggle itself instead of a whole like helmet set up right now so I know Tracy you've considered a question like what should be the right like market value of AR so maybe could you elaborate a little bit on that

Oh yeah, so I just came across this tweet on X. So it claims that the major market of AR should be the productivity. So the marketing value should be the combined of all those screens, laptops, tablets, phones. So perhaps we can also start with enterprise users like the schools. They buy those AR helmets just like 40 years ago they buy Apple II.

for those kids for education usage. But I'm not sure what can this AR do that cannot be done on a computer. So can you imagine why a school wants to buy such expensive devices for kids?

That's a good question. Yeah, I've actually had some experience like working in school, like summer camps and like some other education activities. Yes, throughout the past like 10 years. So yeah, what I've observed is that school they usually like, if it's like a one single device in a classroom or in a school like a digital whiteboard or something like that, they will love to spend like

a big amount of money on that. But if it's an individual device for students, they usually go for the cheapest option, whether it's like a $200 Chromebook or if it's a tablet, then they would just buy the base model of iPad. You probably wonder why they are still selling the older models of iPad online. Well, that's mostly for the K-12 schools.

I think Vision Pro or like the later Vision or AR products by Apple or similar companies, if they're going to stay with a similar price like above $1,000, then it's probably just going to be a one single piece in the classroom where everyone takes turn to try it instead of, you know, every student gets one. Interesting. I thought schools should have received some subsidies from the government. So they are not that price sensitive.

So from what you talked, they still very care about the price. Yeah, I think once you have like thousands of students in like one class or like one school year, then yeah, the cost like quickly adds up, I think. Yeah, so that's why they still go with like the cheapest laptop or tablet they can get. Makes sense.

all right so just to let's see wrap up the ai section so there's also been very recent news like probably two days ago on the cloud uh new newer cloud models so they've released like two separate models uh

The most powerful one called Cloud Opus is slightly better than the GPT-4 and I believe the middle one called Cloud Sonnet is about the same as the GPT-4, probably better in some areas but worse in others. I will attach the official comparison chart they produced later on.

Yeah, and it seems like what they claim is the newer models are more like energy or computational efficient compared to the GPT-1s. And I've also heard rumors that the GPT-5 and I believe the LAMA-3 by Meta, so those two will be the most powerful ones. They are going to be released sometime in the July, middle of the summer.

Since there's been some great outlook on the development of models, what's your view on the never-stopping train of AI here in Tracy?

That's a good question.

Well, I haven't paid for Perplexit yet, so do you know what their price is? $20. Oh, wow. So one way I would look at this is maybe here, first of all, I think the $20 price is just they didn't consider profitability at all. I'm pretty sure everyone is still losing money at this point.

It's more like you want to have the same price as the other competitors so that people think at least your product has about the same performance or functions as the others. I think the price here is more like the price is one of the features instead of a very objective metric or a measurement of the whole package. So that would be my guess.

I feel like at this point, even if you're setting the price at $30, you may still lose profits. So I feel like people are just trying to keep the same price as the others just to show that our product is about as powerful as the others. So that's just my wild guess. I just don't feel the enough motivation for me to switch from ChargeBT to cloud.

So since you've already tried Cloud, how do you feel? Can you tell the difference between it and ShareGPT? Yeah, good question. I just tested

So I gave both Cloud and ChatGPT like a chart, it's like a business chart and the answers they gave are pretty similar. So this is testing on the newest cloud models, the mid one versus ChatGPT 4.0. So yeah, they're supposed to be perform about the same. One good reason I could think of that people might pick one over the other is just purely like visual aesthetics.

Cloud, I think it has more like traditional newspaper-like or you could say like a consumer visual design. It has this brownish or pinkish background color with a, I don't know what the exact font is, but it's kind of similar to Times New Roman, so it's like a newspaper font.

I feel like a lot of people, especially English-speaking world might prefer that. It also makes the product seems like more human-centered. So that would be my own personal view. Interesting. Yeah, I did not think like URI will play such a big role. Yeah, that's...

Also goes back to our initial discussion about perplexity. I think their UI is even one step ahead of chat GPT in terms of how geeky or how futuristic it looks like.

Maybe that's the reason a lot of people actually prefer that, especially those people working within the business and spend a lot of time on social media. They probably prefer the tech-looking user interviews. I see. Yeah, but I cannot afford more than one subscription. Imagine if you have Chagipity, Perplexity, and perhaps another one.

That's a lot. That's crazy. Yeah, that reminded me of the earlier days, like over 10 years ago, you remember like when first there was only one like streaming service like Netflix, so it's cheaper just to pick that over the whole table. It's like $100 per month, but Netflix is only like 10 bucks. But later on, like everyone's releasing their streaming service and

they have their own exclusive content so to get a whole thing in order to subscribe to like five different services again so kind of a similar situation yeah you also remind me that i subscribed to mid journey perhaps i should cancel it yeah mid journey yeah i think uh unless they do something really truly like groundbreaking i feel like otherwise all the more general models are going to like catch up eventually yeah

but currently it still generates more photorealistic images than chat gpt so i keep it for a while yeah all right so let's do our latest uh last two news here so those two are all like kind of chinese related so first one is the start of uh dd like the ride sharing app and the second one is i think a very recent one probably just happened like today uh i think google

or not Google, but like a former Google employee got arrested today in California. So maybe you could provide some details first on those. Yeah, so which one? So the Didi one is not news. It's just I read an old book and it tells the starting story of Didi. So can you imagine if you designed Didi software and you don't have passengers, you don't have drivers, how do you start?

Okay, suppose there are 10 drivers on the app now. Do you think a passenger will download an app if there are only 10 drivers? So how are you going to get them? That's a good question. Yeah, they probably just, it's a very simple, I would imagine, maybe just like a link to a phone number or something. Because I've read like about the DoorDash, like the delivery app, one of the largest one in the United States. When they first started, they also don't have a,

So they're just like having a, probably like one single phone number on their app and you call them and they will like figure out something on their own, like without the use of like a database or something. So would that be the same case for Didi? I think DoorDash is slightly better because the restaurants are already there and customers, they know those restaurants. So DoorDash just deliver those orders. Of course, the more dashers, the better, but

But if they run out of dashers, they can always find someone or they just deliver those orders themselves. I believe those founders did that at the very beginning. But for daily, I feel the problem is slightly harder because if you don't have enough passengers, then those drivers are not incented to download your app. And if there are not enough drivers, you're not going to have any passengers. So it's kind of like a deadlock. So how do you break the lock?

Wow, good question. Well, I don't know much about like ride sharing business. What I read mostly in the history, tech history is on like meal delivery. So I don't know, I would imagine they start with some like very good promotion or just be my guess.

Yeah, kind of. So they try to get some drivers on board and then they hire someone to just hire taxis all the day. So make sure those drivers get money from the app and then perhaps they will recommend more drivers. Wow, yeah, that's amazing. Yeah, here I would definitely recommend a TV show. I believe it's called...

super pumped if I remember correctly about the earlier days of Uber so the US counterpart to Didi from what I watched in the show and also what I read about Didi it seems like those were like the golden days of VC backed startups you just have like endless streams of cash to burn to start like a business platform like this from ground zero so

But nowadays, it seems like if you want to do some startups in AI, we'll probably just have to figure out the revenue model from the very beginning to make sure we're profitable. So yeah, definitely probably a better case for the founders because we can have a more healthy, healthier growth model.

And yeah, let's just spend maybe the last two minutes on the last news here, the Resman. I read a little bit today. It seems like it's pretty, the funniest thing I found out is like he was, I think, doing a startup pitch, maybe in China. And yeah, they took some photos of the presentation. And yeah, it might be the, like the simplest PowerPoint slides that I've ever seen at a startup pitch. Just like some,

default fonts of black and red color like large Chinese characters describing very high level information just seems pretty interesting. Interesting. So you mean you saw the presentation slides? Yeah, I think someone, I think it was a public

like a group startup pitch at like formerly known as like YC China now like TG Chuang Tan I think yeah it was just like large almost like poster like found just saying things like we are the only like China Chinese startup that's on par with like Google well that's probably because you know he got all the information and technology from Google yeah

Yeah, that's crazy. I think it's too confident. Yeah. Do you have any additional information? So first, I think we should definitely

condemn his behavior is unlawful and if he really wants to build his startup perhaps he should resign from google first and wait until the non-competitive agreement expires i'm not sure if they have that at google and and then he can go back to china and build his startup i think there's nothing wrong with that but i think this guy has copied those internal documents

and participate in all those venture capital events while he was a Google employee. So I read from Wall Street Journal that he even had someone swipe his card into Google's building to pretend he's still in the US. That's crazy.

Yeah, that's pretty bad. That also reminded me of, I think there was a former employee at Apple Car Division, which is non-existent right now, but it happened a few years ago. He stole a bunch of critical information documents and went back to

one of the largest like EV maker in China. I believe it was XPEN and kind of helped their like automatic driving division. So yeah, this is definitely not the first case and probably won't be the last one. Yeah, so yeah, I think

Here, I will just be a friendly reminder to our listeners. Yeah, definitely don't attempt any of this kind of behavior. It's not just like a massacre, but also it could lead to like very serious legal consequences. Jail time is definitely like within reach. Yeah, that's so sad. Yeah.

All right. So I think that's pretty much all we need to discuss. I will just do an Easter egg section on my own as I try the sweet green experience from this into a podcast episode, tech podcast episode. Yeah, but that's all for our February and March recap. So we'll release the segments like gradually, probably weekly on our

all the channels we have and definitely looking forward to recording again with Tracy probably in early April. Yeah, so that's all. Thank you everyone. Thank you. Bye.