you2idea@video:~$ watch LgyaAx-E7HU [1:17:54]
// transcript — 1026 segments
0:02 If you told me your health is at risk, >> here's your choices.
0:07 >> This one doctor is going to give you these three doctors are going to give
0:10 you their best advice. >> Uh or you can pick these three LLMs and
0:14 you can feed the information, same information you give to these three
0:17 doctors or these three LLMs. I'm taking the three LLM. >> That sounds crazy, I know, but I think
0:22 for almost every case I would do that. What would you do, Alex? Uh, I would
0:27 take talk to the physicians because >> I have a great number of them in my the
0:32 geographically local family here and I hear about what people bring to them
0:37 from LLM and what kind of advice they're getting not just GBT but the whole gamut
0:44 and I don't think that the a Jason the average person's not very smart so
0:47 they're not doing great research here. But I'm not. >> No, I'm saying for you for you though.
0:52 >> I even with all that. No. >> Okay. >> Yeah. And do you think in the coming
0:55 years it would change? >> Oh, five years from now. Oh god. I mean
0:58 I'm either going to be in a flying zuks or I'm going to be digging in the ground
1:02 with sticks. So I'll take the LLMs in five years. Yeah. >> Yeah. So I think it's then it's just a
1:06 matter of time. We're in agreement. Um >> I think most most AI questions is just
1:09 your time horizon. You know, like >> So you know what this sounds like to me?
1:13 It sounds to me like people with credentials are going to start suing the
1:19 LLMs for giving advice without a license. >> This week in Startups is brought to you
1:25 by Nexos. Stop shadow AI in its tracks with the unified platform for secure AI
1:31 adoption and productivity. Try it with a Pilot. Focus on your product. Let Pilot
1:41 handle the bookkeeping. Pilot provides the most reliable accounting, CFO, and
1:46 tax services for startups and small businesses. Head to pilot.com/twist
1:53 and get $1,200 off your first year. And Squarespace. Turn your idea into a
1:57 beautiful website. Go to squarespace.com/twist for a free trial. When you're ready to
2:03 launch, use offer code twist to save 10% off your first purchase of a website or
2:08 domain. >> All right, everybody. Welcome back to this week in startups. Yes, we're back.
2:16 It's November 3rd. Here we go. Not only is the summer over, but the fall is
2:21 ending, and we will soon be in full-on winter. But before then, uh I am in
2:27 Saudi Arabia. I'm in the Kingdom. I'm in Riad. Uh we had our first day of Founder
2:32 University today. And it went spectacularly. For those who don't know,
2:36 founderun university is a program I've done 11 cohorts of in the United States
2:41 is a pre-acelerator. It's where teams of two or three founders get together and
2:46 [snorts] build an MVP, a prototype. They could be incorporated. They may not
2:50 be incorporated. And then um we help turn your startup idea into a reality.
2:55 We accepted 350 people to the latest cohort in America. That was too many. So
2:59 we're going to bring it down to 250 in the next cohort. Uh but we uh decided to
3:04 do it with our friends at Sonobble which is the venture arm of the PIF uh which
3:11 is the uh sovereign wealth fund of the Kingdom of Saudi Arabia. And if you go
3:15 to mina.launch.co you will see our uh website there. And after a couple of
3:24 days here in Riad, Alex, I will go to Dubai. And then I will say goodbye to Dubai.
3:32 And I will get on a plane to Tokyo. And then in Tokyo, I will announce that
3:37 Founder University in partnership with Jetro, the Japanese enterprise economic
3:42 trade organization that we will be bringing Founder University. I'm just
3:46 going there to announce it. Uh will be coming to Japan in January. and we'll be
3:52 opening up applications for founders in Japan uh to start companies and join our
3:58 cohort there. So, we'll be doing cohorts every year in America, Saudi Arabia, and
4:05 Japan. the business class and the first class cabins. Now on these airlines, I
4:10 took uh American Airlines and British Airways combo from Austin to um Dallas
4:19 to London to Riad. So it's three flights, but >> man, this uh these nice planes are
4:26 unbelievable in business class now. And I was traveling with my friend who is
4:30 one of these, you know, really considered Austin biohacker type folks.
4:36 And I slept. I took some supplements, went right to bed, eye mask, sound, uh,
4:40 dampening, earplugs, all that kind of stuff. Slept like a baby, then stayed
4:44 awake, and then did the same thing. And so that's why, you know, I'm doing the show
4:48 here. You'd be like, "Oh, take a day off from the show." But it's 8:00 p.m. here.
4:52 I'm actually trying to stay up till 10, 11, so that I can reset my mind.
4:56 >> Yeah. You actually don't sound like [ __ ] I kind of I kind like no offense,
5:00 but I've had that like before, too. I kind of thought I was going to like
5:04 extra caffeinate and kind of run things today. Well, you >> limped along like a good sport, but you
5:08 actually seem >> 100% there. All right. Uh, speaking about people who might not be getting
5:13 enough sleep and are a little bit cranky, I think Sam Waltman kind of blew
5:18 up the internet recently by shooting back uh at one of your friends, Jason,
5:23 over at the BG2 podcast. Uh we don't usually like podcasts about someone
5:27 else's podcast, but in this case, Sam Alman kind of cut loose, let's say, uh
5:33 telling Brad Gersonner uh kind of to be quiet about concerns about OpenAI's
5:38 revenue versus its infra uh spend. Jason, Brad Gersonner says, "Hey, you
5:42 know, your revenues are about 13 billion this year. Sam, you have $1.4 trillion
5:46 in compute commits over the next four or five years. You know, 500 billion to
5:51 Nvidia, 300 million to AMD and Oracle. So, how are you going to pay for all
5:55 that? And that is when Sam kind of unloaded. So, here's his response to
5:59 Brad's question. >> And you've heard the criticism. >> We're doing well more revenue than that.
6:05 Second of all, Brad, if you want to sell your shares, I'll find you a buyer.
6:08 [laughter] I just enough like, you know, people are I think there's a lot of people who
6:15 would love to buy OpenAI shares. I don't I don't think you >> including myself. including myself
6:21 >> who talk with a lot of like breathless concern about our comput stuff or
6:24 whatever that would be thrilled to buy shares. So I think we we could sell you
6:28 know your shares or anybody else's to some of the people who are making the
6:30 most noise on Twitter whatever about this very quickly. We do plan for
6:35 revenue to grow steeply revenue is growing steeply. We are taking a forward
6:38 bet that it's going to continue to grow grow and that not only will chache keep
6:43 growing but we will be able to become one of the important AI clouds that our
6:49 consumer device business will be a significant and important thing that AI
6:53 that can automate science will create huge value. So, you know, there are not
6:59 many times that I want to be a public company, but one of the rare times it's
7:03 appealing is when those people are writing these ridiculous OpenAI is about
7:06 to go out of business and, you know, whatever. I would love to tell them they
7:09 could just short the stock and I would love to see them get burned on that.
7:12 >> I think it's a great place to end, Jason. >> Sure. Yeah.
7:16 >> People were a little surprised at his kind of f off response to to Brad. I'm
7:20 curious what your first thoughts were and uh with some more time, what did you
7:23 come up with? >> Yeah, I'd say two things. Number one,
7:27 he's obviously being a little jokey. Um, and I know Sam personally for a long
7:31 time, like 20 years, I guess, that we've known each other. So, I think he's being
7:34 a little cheeky there. I don't think he crashed out like full-on crashed out,
7:38 freaked out. >> I think he was kind of pushing back a little bit. And listen, on the All-In
7:46 podcast here, other podcasts, it's known for being a little bit more full combat,
7:49 right? >> Yeah. >> And I probably have a part in that,
7:54 obviously. So, I think he's being a little bit combative. Um, because that's
8:00 what podcasting is in 2025. Um, and he could also be just like I'm
8:05 on the road right now. Guy's been on the road making a bunch of deals. He could
8:08 be tired. He could be tired of answering that question. Uh, and just say,
8:11 "Listen, if you don't believe in us, sell the stock." Which is a completely
8:15 valid thing to say. And his other point, which is completely valid, is if you uh
8:21 when we're public, this will be easier for everybody to understand. Now, let's
8:24 take the other side of it. >> Okay, >> there is um a really big issue he does
8:32 have. They've been saying they have 1.4 billion dollars in commitments, right?
8:36 >> True or hundreds of billions of dollars. They've got a company that last we know
8:40 is making 13 billion. He said it's a lot more than that. So, I would say a lot
8:44 means I don't know three, four, five billion more. So, let's say maybe he's
8:48 got 18 billion in revenue right now. Maybe that's his run rate. You know, a
8:52 billion5 a month is incredible, right? So, this is a lot of money. A billion five a
8:57 month is like >> $50 million a day, right? >> Yeah. 30 billion a day, 50 billion. This
9:03 is a lot of money this thing's printing off here. So, um it is however very valid to say if
9:11 they're losing 10 billion a year, and I think that's the number. 8 n 10 billion
9:15 is what people say they're short every year. >> Okay. So, you start to put these numbers
9:20 together. There was a rumor that they would raise 60 billion in an IPO.
9:26 >> Yep. >> And if they're deploying 1.4 billion in
9:31 five years, 300 billion a year. He said four in this clip, Brad, but maybe it's
9:35 five years. Who knows? Four or five years it's deployed over.
9:38 >> We're talking about three 400 billion deployed a year.
9:41 >> Well, that's a lot of money. And even if it was deployed over 10 years, it would be
9:47 140 billion a year. If it was deployed over 10 years, let's say the numbers are
9:51 totally incorrect by a magnitude and it's like or or double. Um then if
9:58 they're making a and they got to 100 billion, if this revenue went from 18 to
10:05 36 to 72 and it just kept doubling like in a couple years they're making a h
10:07 100red billion but they're spending 140 billion they're losing 40 billion a
10:10 year. That means the IPO cash the 60 billion would get eaten up in but two
10:14 years and in this model would get eaten up in six months if it was in fact going
10:18 over four or five years. So this is important math. What we don't know and
10:22 we actually had this conversation you're like yeah we can't find this information
10:26 um and if we're wrong and you can find this information by all means tell us in
10:30 the chat rooms how guaranteed are these contracts and we will probably find out what
10:38 conditions there are on these contracts. what milestones, what outs there are.
10:44 You can typically cancel or change or defer a contract at different times. So
10:49 let's say they did buy 300 billion in compute from Oracle. Is that the number?
10:52 300 or 400 billion from >> 300 >> 300 from Oracle. Let's say that contract
11:00 says it's up to 300 billion and it's intended to be deployed in six years, 50
11:07 billion a year, but the buyer can reasonably extend that in six month
11:13 increments 10 times. In other words, for another five years.
11:15 >> Yeah. >> You know, there it's not uncommon for
11:19 those kind of things to be in contracts. So maybe they're on the hook for the
11:22 whole 350 billion. Maybe they're on the hook for a the first hundred and they
11:26 can cancel the last 200. Maybe they can cancel by a certain date. All of those
11:29 kind of conditions haven't come out. Squarespace has been a friend and
11:32 partner of This Week in Startups for over a decade. And it's because they are
11:37 one of my favorite startups of all time. You know Squarespace, the all-in-one
11:41 platform that's going to help you make a beautiful website to help establish your
11:45 brand online and scale your business with you as you grow. Doesn't matter if
11:49 you're selling content or a course. But what if you want something super
11:52 customized? Well, you can use their AI powered blueprint feature that works
11:56 right alongside you as a creative partner. You tell it what vibes you're
12:00 going for, what the aesthetic is, and it makes something totally beautiful and
12:04 personalized to you and your business. And new features keep coming from our
12:07 friends at Squarespace. You remember they did SEO, you know, they've been
12:10 doing that for a decade now. They're doing it for AI. Yes, AI search. People
12:14 are starting with AI and then you see all those citations. They lead to
12:18 Squarespace websites because they do AI search optimization. Go to
12:22 squarespace.com/twist to get 10% off your first website or domain purchase.
12:31 So maybe the 1.4 billion in in all these just mind-boggling commitments that have
12:37 been made are not guaranteed. And maybe the fault that we have here is
12:43 that we're all thinking they're guaranteed. The people who are making
12:47 the money on the other side are indicating to the markets it's
12:52 guaranteed which makes their stock go up which is another can of worms. And they
12:57 are probably going to report in the next quarter or two because these all
13:00 occurred in the last 3 to six months. They're probably going to report soon
13:04 what the details of those contracts are or maybe analysts will start asking them
13:08 and when they do maybe it turns out the 1.4 4 billion in commitments is he only
13:13 owes 10% of that or 20% of that actual number. >> Yeah, I don't think that number is going
13:17 to sink the company because much like you, I think it's a little bit flexible.
13:20 The only thing I'll say is you can only extend certain deals so long in terms of
13:24 repayment periods, Jason, because if you buy the GPUs, they begin to depreciate
13:28 not only if you're using them, but also if you're not because new technology is
13:32 coming. So, there is kind of a a ticker on this. It's not like a bridge we can
13:35 just extend it lifespan by five years. this asset is just non durable in a way
13:41 that some fixed cost investments tend to be. So I I'm with you. I'm not that
13:46 worried. And just to add more fuel to this fire, OpenAI isn't worried. They
13:50 just announced today a 38 billion deal with AWS. So okay. >> Okay. Yeah. So if they're not
13:58 guaranteed, but now let's take the other side. >> Okay.
14:04 There is also the idea uh that people have been speculating about on X.com
14:09 that um there's speculation that if this if they can't make these commitments and
14:17 the company stutters >> or has headwinds and if there's
14:22 headwinds what would happen to these commitments? In other words, if they couldn't make
14:27 these commitments and they broke these commitments. Okay. >> So that means Oracle stock I believe
14:33 went up 20% or something after that announcement. >> Huge.
14:38 >> All right. So what would happen in that case? What would happen to Microsoft if
14:43 their 27% ownership of OpenAI wasn't worth $270 billion at a billion dollar
14:49 IPO, but the company became worth 300 billion or 200 billion and that 27
14:54 billion was only worth 50. So you get the idea. We could be in a situation
14:59 where there is a contagion. >> Yes. >> Let's just say there's a 1% chance Chat
15:03 GPT fails. A 2% chance. >> Sure. >> 2% chance it fails. 2% chance it gets
15:08 caught up in legal, you know, things that cause a problem. the Elon lawsuit,
15:12 the New York Times lawsuit, whatever it is. Um, whatever reason they don't win
15:16 big and they don't become a trillion dollar company, which a company with 13
15:21 to 18 billion in revenue is not worth a trillion obviously, but everybody's
15:24 marking this as a trillion dollar company now. If it is not worth a
15:27 trillion, then there would be a contagion. And who would get hit by it?
15:31 Nvidia would lose hundreds of millions in revenue. Oracle 100 millions in
15:35 revenue. Looks like uh Microsoft would lose hundreds of millions in revenue.
15:40 And now Amazon Web Services would lose close to 40 billion in revenue.
15:45 >> A lot of stocks on the same day in the case of a chat GPT failure
15:49 >> or even a flattening >> it would be a black Monday. It would be
15:54 like 20% draw down in these stocks 30% draw down. It could crash
15:58 >> the AI boom. >> Okay, so a couple notes about that one.
16:02 Agreed across the board there. But for context, if you lost 20% of the stock
16:05 market, Jason, that would take us back to what, like December 24. Like I
16:08 >> correct. >> Yeah. It's not the end of the world.
16:12 Terrible day, but I don't think crashes the tech industry. The other thing is I
16:16 read through all the big tech companies earnings, and one thing that Satia said
16:23 a lot on their uh calendar Q3 fiscal Q2 earnings call was that this compute
16:26 they're building out is is fungeable. And by that, I think he means you can
16:30 kind of swap out first and third party demand. So if OpenAI comes to them and
16:34 says, "Hey guys, listen. We only need 80% of that $250 billion marginal Azure
16:40 commit we made." Well, I think Microsoft can just use it. So I think that that
16:44 helps a little bit. But you're right. A lot of things are depending on OpenAI
16:49 Jason actually as you said on on X and you're a little worried about
16:51 competition I think. >> Yeah. So this is my next like scenario.
16:57 So just to give you one scenario which is chat GPT fails and you could include
17:04 in that what if LLMs are overhyped. >> Yes. >> And all at launch they hit a ceiling and
17:09 for two years they don't get better >> and they're as amazing as they are
17:14 today. They don't really change the world the way we think they are. They
17:18 don't replace as many jobs. >> They don't the agentic stuff doesn't
17:22 work. They don't replace all the call centers. they don't replace all the
17:27 salespeople, they don't replace all the, you know, uh, cab drivers, etc. Well, if
17:31 that were to happen, then you could have like a serious contagion where,
17:37 >> you know, chatpt were to fail and then like Domino's, all the other companies
17:40 would fail as well, right down the line. Yep. >> If like and and when I say fail, they
17:44 just wouldn't reach their potential is actually probably the way to say it. So,
17:48 we're pricing in perfection and that these things are going to have a
17:52 dramatic impact on productivity and they're having a dramatic impact on
17:54 productivity now. But if it stops and plateaus, that would be a problem. But
17:57 here's the other risk. This because the risk I just explained like the either
18:02 the entire sector doesn't deliver or chatp doesn't deliver. I put that at
18:06 less than 5% combined. >> Okay. >> One in 20 like so if you're playing
18:12 Dungeons and Dragons with a one in, you know, uh >> if you're rolling a d20, you can
18:16 >> is there a d20? Yeah, I know it's a D30, but you know, it not zero. Yeah,
18:19 happens. >> Sure. So, it's it's sort of something to that effect. Now, here's one that's more
18:25 likely. Competition. >> Okay. >> Um because the company is priced to
18:31 perfection. In other words, based on 13 or 18 million, they're trading at 30 to
18:37 50 times their price to sales ratio, right? 30 to 50 times their revenue. uh
18:43 at 500 billion and at a trillion they would be at almost you know double that.
18:48 They would be at 60 times 70 times 80 times uh their revenue. Now
18:55 what if Claude uh's API because I saw some notes that Claude's API was beating Chat GPT's
19:02 API. I just saw a chart go by that was getting more popular. What if Cursor's
19:07 new LLM for coding is just much better than Chat GPTs? What if Grock or Gemini,
19:12 you know, start taking some of the consumers away from it? What if Apple
19:18 does a partnership with Meta or with Google and Gemini gets all the Apple
19:22 phone users? In other words, what if they don't get all the market share?
19:26 What if they lose the market share? Chat GPT's market share was at a 100% of LLM
19:30 traffic >> of AI answers traffic. They were 100% at one point. Now they're down to like 80
19:38 or 85% I believe. um which is still a massively dominant position. But what if
19:42 they go down to 50% or 30% eventually and there's five major players.
19:44 >> Yeah. >> Then that could also mean that they can't reach that $ 1.4 billion
19:51 commitment. What if the utilization of these chips like we saw the uh research
19:57 on deepseek and coming out of China that they were using OCR to you know store
20:03 data and process data like 10 times more efficiently? What if they don't need all
20:07 that compute because the open source community gets it right and it takes 10
20:11 times less to provide the same thing? What if the price becomes so
20:16 deflationary that, you know, paying $20 a month is not necessary because Gemini
20:20 decides it's free? Google just says it's going to be free. Apple says it's going
20:23 to be free. >> Uh, and what if Amazon just says, "Yeah,
20:26 you know what? Screw it. We got to protect the AWS franchise. we're going
20:33 to just put up open- source uh LLMs and, you know, we'll just make this so cheap,
20:37 it's close to free, and there's no margin. Well, then you're going to have
20:43 to price Chat GPT like a SAS company. And oh boy, a six times multiple of
20:49 earnings or, you know, a 10 times multiple of sales might not look as
20:53 good, right? It's hard to become a trillion dollar company if you're being
20:58 traded at 20 times earnings or 10 times sales or five times sales. So that
21:02 margin compression could also cause a downdraft. I think that's a 50% chance
21:07 happening. So I think it's 45% right now that OpenAI continues their dominant
21:12 run. 5% chance the whole sector andor chat GPT hits a major, you know, plateau
21:18 and then there's like a 50% chance that competition makes it impossible to spend
21:21 that 1.4 4 trillion and the technological risk. Probably the best
21:25 part of being a founder is watching a project that started as just an idea
21:31 blossom into a full-fledged startup. The worst part, it's bookkeeping. It's
21:35 boring. It's detail- oriented. And it takes time away from building your
21:39 product and hey, growing your team. It's time for you to let Pilot take care of
21:43 all this bookkeeping so you can get back to running your company. Pilot's the
21:47 largest accounting firm and it was built just for startups. They understand the
21:51 technology industry and the landscape. They know the stakes and they're already
21:55 trusted by amazing companies we love like OpenAI, Scale, Air Table, and
21:59 beyond. And you know what? Pilot is about so much more than just getting
22:03 your taxes done on time. Pilot has a smart dashboard that puts all the
22:07 essential metrics your startup is tracking at your fingertip. And they
22:11 have helpful adviserss who can walk you through crucial financial decisions
22:14 you're going to need to make in the coming years. Plus, when it's time to
22:19 raise your next round, Pilot COO and CFO services are there to help. So, it's
22:24 time to focus on your company and let Pilot handle the books. And Twist
22:29 listeners get $1,200 off their first year. Just go to pilot.com/twist
22:37 You know, if if it ends up the case that a lot of these companies don't need as
22:42 much infro as they think or that demand doesn't surface, other companies are
22:45 going to really struggle. I mean, Cororeweave whacked. XAI, which is
22:49 spending a lot of money on building footprint, is Anthropic.
22:56 >> Like, shout out to them for not buying GPUs, I guess.
23:00 >> Yeah. And here's the other thing. it when people worry about this stuff. I
23:03 don't worry too much because I think we're in the situation where it doesn't
23:09 really matter uh if these people lose a little bit of money. The people who are
23:11 putting up the money can afford to lose it. If true >> Google spend and if Google and Facebook,
23:18 you know, blow their earnings a couple years building out this infrastructure,
23:21 it's not the end of the world. So, carry on everybody. And I think it can give
23:25 Sam a little grace. Um it wasn't a good moment for him. I think people might be
23:29 overreacting to it a bit. Oh, he's handling this so much better than I
23:32 would. If I was him, going around the world jetlagged, exhausted, trying to
23:35 literally save a >> He's on a global He's on a global 650.
23:40 He's good. Don't worry about Sam. He, unlike me, he's got a global 650
23:45 minimum. You might have a Glo G800. You might have the 800 at this point.
23:48 >> Uh Sam, do you want to come on the show and tell us what you're flying in? We'd
23:51 love to know how you're handling jet lag and if you too are big on Whoop.
23:55 >> All right. What else do we have here on the docket?
23:58 >> All right. Uh, next up, Jason, I want to talk about OpenAI and legal and medical
24:02 advice. This is all over X this morning. People are saying, "Oh my gosh, have
24:06 they just neutered chat GPT?" Because if they did, back to your point, that could
24:09 limit usage and so forth. I'm here to tell everybody that they have not loized
24:14 your favorite chatbot. Instead, what they have done is changed the terms of
24:17 service a couple of days ago, Jason, and now Chad GPT will effectively offer you
24:22 information on medical and legal questions, but it will not offer advice
24:27 to you. And in the new version of the OpenAI kind of terms of service, there
24:33 is a line that says uh you cannot use our services for uh the provision of
24:37 tailored advice that requires a license such as legal or medical advice without
24:41 appropriate involvement by a licensed professional. This to me reads like C
24:44 YA. >> Yeah. So I guess there were a couple of accounts on Twitter that were saying
24:51 there it's chat GPT is no longer going to give health advice and people were or
24:55 legal or financial and immediately people said well that's what I use GPT
24:59 for most >> that's the use case. Yeah. [snorts] >> Uh so like literally I take a picture of
25:06 all the supplements I'm taking and I say what else should I take? What are these
25:11 known for? And I will take a picture of my creatine, my electrolytes, my protein
25:18 powder, and my collagen when I make like a shake in the morning. I'll put it all
25:20 together and say, "Hey, >> yeah, >> what is this? What impact is this going
25:25 to have on my workout today, etc." And uh, you know, I'll put it in my little
25:29 health tracker. So now they're going to instead of saying, "Hey, you should add or you should
25:34 consider adding these things." I guess they would say a doctor, right, or the
25:39 FDA says you should have this many grams of protein. Peter Aia says you should
25:46 have this many. So this does feel like they are concerned about practicing
25:51 without a license and I think this reflects how good they are. It turns out
25:57 there was a study um which we'll have somebody look up right now. I think it
26:01 was John Hopkins Hodgekin >> Hopkins >> uh Hopkins or it might have been Harvard
26:08 did a study on just general practitioners GPS doctors versus chat GPT and large language
26:15 models giving advice and they they sort of tested all of them and it turned out
26:20 you know head-to-head the LLMs did better than the doctors. Then uh they
26:25 did bedside matter and it turned out Jack GBT like uh or all these services
26:31 did like much better than the doctors because they're patient with patients
26:35 and they have great bedside manner. >> So just Jason backing you up there there
26:39 is a John's Hopkins University study from uh mid 2023 entitled the doctor is
26:45 out but it's okay chat GPT can answer your questions. uh found that Chad GPT
26:48 outperforms human physicians in quality and empathy. And then there was also
26:52 Times coverage of similar studies in November of 24 saying that quote a small
26:56 study found Chad GPT outdid human physicians when assessing medical case
27:00 histories. So essentially we are seeing doctors take this stuff up as well but
27:03 they can interpret they can better understand and find obvious errors. Um
27:10 >> I can they find errors? I you know I I would actually say if you told me
27:16 um if you told me your health is at risk, >> here's your choices.
27:22 >> This one doctor is going to give you these three doctors are going to give
27:25 you their best advice. >> Uh or you can pick these three LLMs and
27:29 you can feed the information, same information you give to these three
27:32 doctors or these three LLMs. I'm taking the three LLMs. >> That sounds crazy, I know, but I think
27:37 for almost every case I would do that. What would you do, Alex?
27:41 >> Uh, I would take talk to the physicians because >> I have a great number of them in my the
27:47 geographically local family here. And I hear about what people bring to them
27:52 from LLM and what kind of advice they're getting, not just GBT, but the whole
27:58 gamut. And I don't think that the a Jason, the average person's not very
28:01 smart, so they're not doing great research here. They're not for you. For
28:04 you though, >> I even with all that, no. Okay. >> Yeah. And do you think in the coming
28:09 years it would change? >> Oh, five years from now. Oh god. I mean,
28:12 I'm either [clears throat] gonna be in a flying zuks or I'm going to be digging
28:15 in the ground with sticks. So, I'll take the LLMs in five years. Yeah.
28:19 >> Yeah. So, I think it's then it's just a matter of time. We're in agreement. Um,
28:22 >> I think most most AI questions is just your time horizon, you know, like.
28:25 >> So, you know what this sounds like to me? It sounds to me like people with
28:31 credentials are going to start suing the LLMs for giving advice without a
28:36 license. And you did have this suicide and I believe the the suic there's a
28:43 serious suicide case that Chat GPT was involved in. Not the person who worked
28:48 for Chat GBT who was the whistleblower. That's the Joe Rogan clip we could play.
28:52 Uh which was >> crazy. Uh on top of the Tucker Carlson
28:57 one. Um and we can get into that. But I think the you know more relevant one is
29:04 >> uh Adam Rain who died by suicide in April and he was uh deep into
29:08 conversations with uh CH GBD and if you recall everyone listening this is one of
29:13 the things that led to OpenAI releasing data about people using the service
29:17 while under uh Jason heavy mental duress let's say and also they changed how they
29:22 handle uh outreach to uh crisis response call lines and also how they respond to
29:27 people who are um as I might say losing it while chatting to the uh the
29:29 [clears throat] chatbot. >> So, according to the lawsuit that you
29:35 just mentioned with Reigns, um the lawsuit says, quote, "Chacht actively
29:41 helped Adam explore suicide methods." >> Wow. >> This is the one when it said it
29:44 couldn't, but if you ask me to do it in a story world building setup, I can. And
29:49 he took that easy workaround. >> Yeah, I guess. Um [sighs] this is some
29:55 dark stuff. So, I I will give Chat GPT some credit here. I'll give OpenAI some
30:01 credit for maybe trying to get ahead of the how great these things are giving people
30:09 a false sense of security that they should blindly rely on them. And so,
30:12 maybe they'll just start giving more disclaimers. And I I was talking to some
30:16 friends over dinner about this when the story broke and I just said
30:19 >> all they have to do is have anytime anything health related, finance related
30:25 or legal comes up, they say legal is a serious, you know, practice, we can give
30:29 you general information from this experimental software. Please do not
30:34 rely on it. And they should just give that paragraph and then you should have
30:37 to check a box that says click here to sign you have read this.
30:40 >> Yeah. like you do for your terms of service or maybe even like a docuign
30:46 sign it and say I release all liability from chat GPT Grock Gemini etc for any
30:53 responsibility for me using this health information to make health decisions.
30:58 >> Yes. I think also they're trying to frontr run regulation
31:01 >> because if they didn't do this sort of thing as the most popular consumerfacing
31:07 AI tool uh they're catching all the all the fire like you mentioned earlier the
31:11 API between anthropic and uh open AI how anthropic is winning it's true anthropic
31:14 business is growing very quickly on the back of enterprise sales whereas open AI
31:19 is majority chat GPD subscriptions so they're the tip of the spear here and if
31:23 they screw this up and then Congress comes down with you know regulation from
31:27 hell. Everyone's going to suffer. So, they're kind of doing I mean, God's work
31:32 in a way by limiting us from, I would say, misusing these tools as oracles
31:37 versus what they are, which is large language models. And I don't mean that
31:40 as a disc. It's more of a statement of fact. They're not omnicient. They don't
31:43 have consciousness. And I do think people spend a little too much time
31:48 trusting them here. Uh that said, whenever OpenAI tells me you won't do
31:52 something, Chad GBT, I scream. I'm like, are you kidding me? I'm not four. Do it.
31:56 So, I'm on both sides of this one, Jason. >> There are so many amazing AI tools out
32:02 there that can make your team faster and more effective. Any founder would be a
32:06 fool not to embrace these tools. But a large team making use of a diverse array
32:11 of cuttingedge AI apps can also pose a privacy and security nightmare for your
32:17 organization. But now there's Nexos AI, the workspace that will help your team
32:21 up their game with greater efficiency without sacrificing your data or your
32:26 peace of mind. Nexos AI is going to give your teams a simple browserbased
32:31 environment to work in, integrating all the latest and greatest AI models and
32:35 replacing the various potentially problematic tools and subscriptions your
32:40 staff can't even keep track of. Anyway, you're probably paying for two or three
32:42 of things over and over again redundantly. Finally, your admin team
32:47 gets full visibility and oversight into what's happening. So, your team can
32:51 enjoy all the productivity gains of AI without the data leaks, and your
32:55 platform remains compliant while your data stays safe and secure. But don't
32:59 take my word for it. Try it yourself. Go to nexos.ai/twist for a 14-day trial. That's
33:12 Next on the docket is a find from the production team. We have talked on the
33:16 show quite a lot about how startups might be reporting gross revenue instead
33:21 of net revenue or contracted ARR versus real ARR. Well, what's a great way to
33:25 find the source of truth, Jason? It's linking your Stripe account to a public
33:29 database so everyone can see how fast you're actually growing. So, to that
33:33 end, Marcus and I have found for you trustmr.com. Check this out. This is a
33:37 list of 100 companies that have connected their Stripe and you can
33:41 literally just see how much revenue they have and how quickly they're growing.
33:45 Take a look. [snorts] >> I love this. Right. >> So, the number one number one company is
33:50 this and you can see its revenue over time. >> This being data expert tech creator has
33:58 3.4 million in all-time revenue, 168,000 in the last 30 days. 4589.
34:04 Yeah. There was a company called uh Buffer that was doing social media
34:08 posting. Yes. >> And Buffer was part of that build in public group.
34:13 >> So what's great about this is hey most people are not going to put their
34:17 content up here but this is alltime revenue. If we just check that down box
34:21 there and I see here you just put your restricted I API key in but instead of
34:26 all time let's do this month or 30 days. >> Yeah. So the 30-day winner is hype
34:30 proxies. Um >> well now every VC is coming here to set
34:37 up meetings with these companies. Um so the upside here is you've just got the
34:43 entire investment community to come here and go talk to you about your company.
34:48 Um the downside is that you've just inspired people to create a version
34:52 [laughter] of your software and charge half as much or turn your software into
34:57 a feature. And this goes to the should I build my startup in public discussion
35:01 that I keep having with people and you got to be careful. Um sometimes giving a
35:05 revenue milestone can get you a little bit of action and press and information
35:10 but you are educating people and it's a very competitive world. One thing I've
35:14 seen over and over and over again, not to single out Y Combinator, is failed Y
35:21 Combinator companies, uh, who got the 500K from Y Combinator, the 125K and
35:27 then the convertible note, um, just literally ripping off other companies in
35:31 the marketplace because they can't think of an idea, but they have some money and
35:33 they want to make a go of it, so they just take another startup's idea. And
35:37 there's been some gnarly cases where young entrepreneurs do stupid things.
35:40 And it's not just Y Combinator, but I think it happens in Y Combinator because
35:44 you have this combination of a startup with builders who can build quickly who
35:49 have been selected because they're kind of hackers and they like to break rules.
35:52 Remember that question they asked? What's an example of you breaking rules
35:55 in your life? >> Social hacking or something. Yeah. >> Yeah. There's some like rule there. You
36:00 have rule breakers who are known to build fast who have some money in their
36:05 pocket and then they fail and instead of coming up with their own idea they come
36:09 here. So imagine if you know there's 200 people in a Y combinator class okay 10%
36:13 of them can't figure it out. So you got 20 people with 500k and two developers
36:19 three developers whatever three builders what should they do they should go right
36:24 to this list and steal these ideas. And that's the dark side of this and why I
36:30 would recommend not doing it. >> Not doing it. That was kind of my
36:32 question to you is should V should startups do this because I don't know I
36:36 went through this list. I'm not a VC but I'm keeping track of companies for us
36:39 and I was just like who do I know on this list? Who is not who do I not know?
36:43 Who is news? Um as a small addendum Jason there is a there's a fake version
36:47 of this now. If you go to fake-mr.com uh it's the exact same thing but done in
36:51 a whimsical format. So if you want to see how totally real company and uh
36:57 Walmart Walmart are too. Um people are having a lot of fun which I which I
36:59 appreciate. >> I mean it's an interesting concept. Um and in a world without
37:09 people stealing each other's ideas and so many overfunded teams in the world in a peak we were in
37:16 peak zer now we're peak AI bubble. Yeah. >> So if it wasn't peak bubble or like
37:21 we're kind of We're getting to peak bubble. I would say when the whole
37:24 discussion of the first half of the show is some lunatics spending$1.4
37:31 trillion dollars, not billion, trillion dollars. There's a startup spending$ 1.4
37:36 trillion dollars on infrastructure. Like it's pretty heavy out there, folks. So I
37:41 would be careful inspiring people to go steal your business idea.
37:47 >> For 1.4 34 trillion is is 1,400 billion which is just
37:50 >> it's 1400 unicorns. >> Yeah. Yeah. >> Yeah. It's crazy.
37:54 >> Jason, uh before we move on, I have a thing I want to ask you about. So, uh
37:56 Jason L, >> you could give 14,000 startups $100,000. $100 million to you could fund 14,000
38:04 startups with a hundred No. Is that right? Wait. >> Yeah, I think that's right.
38:09 I I I mean >> you' be a billion dollars to 1400, which
38:14 means a h 100red million to 14,000. >> Yeah. >> I mean, it's the numbers are so big. I'm
38:19 like actually not believing the back of the envelope math I'm doing from halfway
38:23 around the world with jet lag. >> That's the problem with, you know, I
38:27 mean, a thousand thousands a million. A thousand million is a billion. A
38:30 thousand billion is a trillion. And so it it gets >> big really really quickly.
38:34 >> Very big. >> All right. Next up, Jason. StarCloud is
38:39 uh finally in space. They went up on the latest SpaceX mission and dropped their
38:43 first uh satellite into space. Would you like to see a Nvidia GPU take flight
38:49 from uh SpaceX? >> I know Jensen does. >> Yes, he does.
38:56 All right. So, here is the clip of Okay, here we go. I guess that's
39:09 >> dark cloud one. Separation confirmed. I 3 separation confirmed.
39:15 >> And that was our final payload of the bandwagon 4 mission manifest.
39:20 >> I thought that was super duper cool. Um, we've been talking
39:22 >> Wonder what that's attached to. Is that attached to the to the, you know, like a
39:27 Falcon 9 or something like that? Or is that their unit now attaching H100s?
39:32 Um, >> so this this is the first uh kind of demonstrator device from the company.
39:40 And when we had uh Philip Johnson on the show earlier this year, he said
39:43 >> so 2073. >> Yes, this is a 1 kilowatt, so not very
39:49 powerful, 50 kilogram satellite that I believe has a single H100 on it. So this
39:55 is a let's get it up in space. I mean, you have to deal with radiation. It's a
39:58 different, you know, level up there. Cooling's different and so forth. Uh,
40:01 but the cool thing is now that they have this up into space and working, by the
40:05 way, they didn't lose connectivity to it. So, that's fantastic. Um, they're
40:09 working on their next satellite, which is going to be, I believe, called
40:12 StarCloud 2. The company rebranded from Lumen Orbit earlier this year, but
40:15 doesn't really matter. Um, it's going to have about a hundred times more GPU
40:20 capacity. So, the company is now uh, you know, Stellar, and then soon it's going
40:24 to be uh, a fleet. And I I don't know, Jason, I just I this company was one of
40:28 those you see, YC demo day, you're like, "Sure, data centers in space. Why not?
40:31 But in like two years, they've gone from founding to actually having a GPU in
40:36 space that works. So, I'm blown away. >> Which says a lot about SpaceX's cost of
40:40 putting stuff in space that Nason startups can actually get to space,
40:44 right? Without SpaceX, that wouldn't happen. So, that's really interesting.
40:48 Um, I guess the question is do do these satellites want to connect to it and
40:54 actually offload a job to it or uh like so that would be their customer
40:58 is there's a satellite up there and do they dock with that satellite or how
41:03 does a satellite get data from a satellite to the H100? Is there a
41:09 standard for that? Like does Wi-Fi work in space? I know this is a very stupid
41:12 question, but I would assume Wi-Fi doesn't work in the vacuum of space or
41:16 it works really well. >> Well, uh given my my deep um you know,
41:21 physics PhD that I have, the way that he explained to me was uh lasers
41:24 essentially in between different satellites to connect uh them in orbit
41:29 compute is going to be one of their markets, but I think more uh near term
41:34 they're going to be working with you basically using lasers to send data up
41:38 and down from the planet. And he says that there is an advantage to having
41:43 optical for that versus connecting to ground centers. You're gonna have to
41:45 help me out with the terminology here, Jason. The the the places on the on the
41:48 Earth where we talk to satellites traditionally using radio waves and
41:50 such, >> that's much lower bandwidth. But with their second demonstrator, they are
41:54 going to have the ability to beam stuff down with the laser, which is much
41:57 higher throughput. But right now, I I I just think they're trying to do like the
42:01 kindergarten version of their future projects to make sure that they can do
42:04 it. Um, the company though is not derisked. I don't want to overly uh sell
42:08 this. They definitely still need lower launch costs to get the amount of solar
42:12 panels and GPUs and stuff into space to make their idea function. So,
42:15 >> of course. Yeah. >> But I don't know. I'm just I was so
42:19 jazzed for them. Like, do you ever just get excited for a founder that you meet
42:22 and you really like and then something good happens? That's why I wanted to put
42:25 >> Absolutely. Yeah. And so, I see here just asking producer Claude to help us
42:30 out at claude.ai and we'll give you a promo code in a second. Radio frequency communication.
42:36 Most common method spacecraft use radio transmitters and receive with
42:40 directional antennas. Common frequencies sband 2.4 GHz. That sounds like your
42:46 Wi-Fi at home. Data is modulated onto radio waves and beamed between
42:49 spacecraft. Requires line of sight between transmitter and receiver.
42:53 Optical laser communications, which is the newer, faster technology, uses laser
42:58 beams to transmit data at much higher rates than RF. can achieve gigabits per
43:04 second versus megabits for RF radio frequency. Requires extremely precise
43:09 pointing since laser beams are narrow. NASA LCRD laser communication relay
43:14 demonstration uh demonstrate this technology is less susceptible. Wow, that's really
43:20 interesting. Um yeah, and I guess there's distance and power limitations
43:24 on these, but that's how data is sent in space. Now everyone knows, but we have
43:28 startups that are working on building space tugs, that are working on in orbit
43:32 refueling, that are working on trash cleanup, that are trying to put GPUs
43:36 into space. We're talking about in orbit scientific experiments from Varta, uh,
43:41 in orbit factories. I, if I was starting a VC fund today, I would just raise as
43:44 much money as I could and give it to all these awesome companies because I'm
43:47 really, really bullish on the space economy in the next 10, 20 years. there,
43:51 [clears throat] you know, with this whole space economy though, there needs
43:56 to be a customer other than the space industry. So, it feels like we're picks
44:01 and shovels, but what is the actual benefit to humanity? And I think the
44:05 best thing I've heard is that mining certain things off of asteroids, etc.,
44:10 and the collection of some of those minerals, rare earths, you know, like
44:14 maybe they're rare on Earth, but not rare on asteroids type thing. That those
44:18 might be the big win. That's what I'm waiting for is like what's the big win?
44:22 I mean, scientific discoveries are great, but you know, are some of the
44:27 moral victories or intellectual victories that don't actually translate
44:32 into a monetizable revenue stream for a product or service to sell to other
44:36 humans and businesses on the planet. That's my only concern with these
44:40 businesses. I know that there's a great business in, >> you know, sending stuff up there, but
44:44 you know, Elon found a business a lot of people are willing to pay for,
44:52 >> which is SpaceX. People want um to have Starlink. >> Yeah. Oh, Starink. I mean, if if they
44:56 broke out Starlink, how much would that company be worth, Jason?
44:59 >> I think it's a hundred billion dollar company out of the gate. Yeah. I don't
45:03 have enough information to know, but I mean it's >> I think there'll be I think there'll be
45:08 a billion people who are paying for Starlink eventually. >> Yeah, I I can see that. Uh to answer
45:12 your question though, what is the actual like use case? By the way, this echoes
45:16 to me like crypto conversations and also early AI to some degree. It's kind of
45:20 fun that we end up doing this kind of on repeat. um picking a company from the
45:25 bucket. Uh Albido Space is doing very low Earth orbit um imaging of the planet
45:30 uh with some special tools and techniques they're doing with that.
45:31 They're another company that I've tracked for a while. Uh but to make
45:35 their technology work, they need to have very very low satellites, which means
45:38 that they have high drag, so they burn a lot of fuel, so they need inorbit
45:42 refueling. So I think we're starting to see some applications that will use the
45:45 Pix and Troubles that we're building, but you're right, we're not there yet.
45:48 Um and >> I do believe space stations um we're going to announce an investment um
45:56 shortly and I think space stations are going to become like the next
46:02 um card to turn over in this. you know, people are going to want to have a
46:05 permanent home in space. And there's going to be a lot of providers,
46:09 non-governmental providers who are going to provide, you know, I don't want to say Airbnb of
46:15 space, but there's going to be a need to have permanent installations in space,
46:20 whether it's for a data center or for people or for shipping stuff. So, it's
46:25 it's coming, folks. It's going to get really interesting in the next
46:29 >> Axiom Space is a one name in the uh private space station game. And also, if
46:33 you want to look out asteroid mining, uh Trans Aastra and Astroforge are two more
46:38 names to add to the mix. All right. >> And what is our code for Claude? I just
46:41 did a Claude promo. We shouldn't throw in the We missed to throw in the code
46:43 there. >> Oh, yeah. No worries. claude.twist if you want to save 50% off your first
46:50 three months of paid Claude access. We use Anthropics Claude here at Twist each
46:55 and every day. It powers every show. We live and die by it. Come join us if
46:57 that's your jam. >> Yeah, it's uh it's awesome. Well worth
47:04 paying for. All right. Um you know, I I I'm on the road right now and I've seen
47:10 there's a lot of uh poly markets going on that I just want to check in on.
47:15 >> Oh, okay. Hit me up. I wanted to check in on the New York City mayor election
47:20 and I have an announcement after we see what's going on on poly market with this
47:24 New York City mayor election because I been hearing that uh Cuomo is coming
47:29 back um and that these there's some polls. I saw this on social media but I don't
47:35 trust polls and stuff I see on social media anymore. So when I want a version
47:41 of the truth I just go to Poly Market. It's got the Zohan. The Zoron is the Zon
47:47 the Zoron is at 94%. >> Yeah. Uh and even more so, Jason, look
47:52 at the trend. Plus one versus Quuomo down one. Now, uh for folks out there
47:55 who don't know what we're talking about, this is the New York City mayoral
47:59 election. Uh no, this is not for president. It may feel like we talk
48:02 about it as much as a presidential election, but because New York City is,
48:05 oh, I don't know, Jason, one of the capitals of the world and a financial
48:10 titan, um it gets outsized media and press attention. Well, and also he's
48:15 30-some years old and a socialist and uh I guess you know I I don't know
48:23 how to say this. Uh anti-ionist, pro Palestinian. I don't I don't know
48:28 exactly how he frames himself, but my Jewish friends in New York are very
48:32 concerned about his stance on uh Israel. So New York >> I mean [laughter] a lot of Jewish
48:37 brothers. >> There's a lot of Jewish people in New York City. Absolutely.
48:42 >> Yeah. And they're like, "What's going on here?" And there's a lot of capitalist
48:46 there, too. But here we are, folks. Uh I I think Poly Market gets it right. So,
48:50 >> what uh what else did you want to check on, Jason? >> Um
48:55 >> I was I'm keeping track of the government shutdown myself.
48:58 >> Oh, yeah. Is that Yeah, it's very interesting because
49:02 I guess what we're trying to figure out is when it ends. >> Yeah. So, the thing that I wanted to
49:06 highlight uh for everyone today on this play market, a couple things. One, I
49:09 like poly markets that have lots of volume. This one has nearly 6 million.
49:15 Uh, but it's the decline of optimism. So, if you're on the audio version,
49:17 we're looking at a chart right now of just when will the government shutdown
49:22 end. And the furthest out option, Jason, is November 16 plus. And it just keeps
49:27 getting higher, which to me implies we could have at least another two weeks of
49:31 this shutdown, making it, I think, the longest shutdown in history. And it
49:35 doesn't seem that anyone who's making hedging bets here thinks this is gonna
49:39 change. And that to me is just an indictment of um my nation, our nation's
49:44 current political good luck. >> Yeah, what a disaster. Um uh November
49:50 16th plus >> is 54%. [snorts] >> And I as you can see there was a ton of
49:55 volume. See this is actually the interesting thing of a timebased one. If
50:00 you scroll down just to teach people the poly market uh how poly market works you can see the
50:08 resolved. So if you scroll back up you'll see like uh yeah so you see view
50:13 resolved. So what you can see here is that if you had been betting on this and
50:20 wagering and you know uh you could have bet before 15th, before October, before
50:26 the 19th, before the 20th, those have been resolved. So this is like continues
50:30 to resolve itself. So >> I guess the November 3rd one resolves,
50:34 you know, later this week, November 7th over the weekend, etc., etc.
50:41 >> Man, this is um really interesting. I I kind of like the government being shut
50:44 down because I think we're spending too much money. >> We're not saving. We're not saving much.
50:49 >> Yeah. Like, doesn't all the money just go back out anyway? Like, everybody just
50:53 gets their paycheck. So, the only people suffering here >> are people who are civil servants. So,
50:59 >> unless they're the ones in charge and they still get paid.
51:03 >> There's there's a there's a disconnect between uh business and government that I think
51:08 people get a little bit mixed up. So, um, in a business, Jason, what's the
51:11 number one expense most of the time? Humans, right? >> People. Yeah. Humans.
51:15 >> Yeah. It's a it's a it's much much smaller fraction of the government
51:18 spend, uh, then compared to corporations. And so, when people show
51:22 up and try to like cut staff as a way to like dramatically reduce the federal
51:25 outlays, it doesn't work because most of it is entitlement programs and earned
51:28 benefits. >> So, I don't think the shutdown actually saves us much money. I just think it
51:33 makes the country worse. >> Yeah. Yeah. And those 600,000 furoughed
51:38 workers, um, those folks are furled, which means when they come back, they
51:41 get their money, I guess. And then there are a bunch of them, God bless them,
51:45 who've been working now for a month. >> Yeah. >> Without pay.
51:48 >> Yeah. >> Which brutal is is brutal. And then people aren't I mean, I understand like
51:55 the food stamps program is abused massively. You know, who knows 10 20 30
52:01 40 50% who knows what the abuse is. whatever it and you can buy soda on it.
52:04 You could have tons of complaints about it, but there are some people whose food
52:09 is, you know, like they're I don't know if they're going to starve, but there
52:12 are people who are going to struggle and so I just wish these people would work
52:17 together in a more functional way. >> all around. But at least we have poly
52:23 market to keep us entertained, Jason, as we fiddle while watching Rome burn, you
52:27 know. Yeah, I mean well no what actually is really great about it is there's
52:32 another poly market I saw which is will it be the longest in history
52:37 >> and I didn't know this but you know that's one of the things I like about
52:40 poly market is I can start to build mental models of what's going on in the
52:42 world >> and I didn't actually know what the longest shutdown was it was December
52:50 2018 to January 25th 2019 >> so according to quad that was 35 days
52:54 and that was during the first Trump administration and this was the fight
52:58 over the southern border wall. Um, but we're at 34 days now. So, if the
53:01 government doesn't radically immediately open by tomorrow, well, congratulations
53:05 everybody. We've done it. A new all-time worst. Uh, would you like some good news
53:09 though, Jason, to put a little smile on your face? All right.
53:12 >> Uh, another Twist 500 company is in the news. This time it is Lora. It's a legal
53:18 AI startup or maybe an AI legal startup. Swedish offices around the world just
53:22 raised 150 million in a series C led by Bessemer. Now, a unicorn 1.8 8 billion
53:29 valuation and its customer count um I think effectively doubled while it added
53:33 another 20 markets since its last round and the company's CEO Max Unistron says
53:38 quote we are seeing astronomical demand for our product legal professionals
53:42 across the globe are adopting AI into their work at an unprecedented rate so I
53:46 was curious what does that mean for other legal AI companies I went to the
53:51 Twist 500 I pulled all of them and it turns out they've all raised money in
53:56 the last month more or less. So I think what we're seeing here is a really
54:01 impressive locus of market adoption and demand for generative AI products in a
54:06 new industry that is not software development. But Jason, I'll just say
54:10 structured language, huge corpus of information, historical record to pull
54:14 from. Those are things that LLM tend to do well. So maybe it's not a surprise
54:17 that we're seeing legal AI companies do so well. But I was just blown away by
54:22 the amount of capital going into Eve, Nexel, Spellbook, Harvey, Even Up, and
54:27 Lora in the last like 30 days. >> So, yeah, there's and there's a couple
54:31 of things that lawyers do. Um, you know, of course there's research like you're
54:36 researching a case. Um, and you know, research has had WLAW and Lexus Nexus
54:42 and services like that, but being able to use AI to do it and do natural
54:46 language queries. You used to have to learn how to use Lexus Nexus really
54:49 well. And there were some people in the firm, your law firm who knew how to use
54:52 it really well. So, you just go to them and you would pay like buy the page, buy
54:57 the surge. It was not cheap. And then you would just pass those like 5 cents a
55:02 page, 25 cents a page of research on to your clients anyway. So, I remember it
55:07 was like, you know, you might be doing thousands of dollars worth of legal
55:11 research on those services. Now, you're going to have all the that corpus into
55:15 an LLM and the ability to, you know, not pay for it. Really interesting. Um but
55:22 drafting documents is something that um people do all the time and they have
55:27 document management systems but to be able to draft a document using AI is
55:32 really powerful because if you said I need um a non-disclosure agreement for
55:39 Alex and Jason they both work at launch you know launch has its own template
55:43 they have the names of the company you have your name your social whatever
55:47 information a lot of that drafting could go a lot quicker. Um, and then the third
55:52 thing that lawyers spend a lot of time doing is reading documents, right? They
55:56 have to process them. >> And document review is the other thing
56:01 you get charged for. So, if you get, >> I don't know, um, if you get sued, okay,
56:07 you have a human resources case. Now you got to go find all the documents
56:12 associated with that manager, that employee, uh when they started, all of
56:17 their emails, all of their Slack messages, right? In the case of
56:20 discovery, you're researching those documents. And then there could be legal
56:23 documents. Oh, we're doing an M&A deal. We did an M&A deal. We're being sued.
56:29 You hire a new council to, you know, do the um to do this. You have to go get
56:36 binders of documents, folders and folders of signed documents and
56:39 summarize them and read them. Like you're saying, research, drafting, and
56:44 reviewing, those things are just so custommade for an LLM. Now, the problem
56:50 is you can't make mistakes. >> No. >> So, this software, I think, is going to
56:57 need time to fully bake. So, and then people get lazy. >> Yes.
57:01 >> And they don't check their work. So, there was a case of somebody who drafted
57:06 a document and sent it to court with fake stuff in it. We covered that two
57:10 years ago. >> Yeah, I was going to say I've seen that recently, but then my second thought was
57:15 I haven't seen that only once. >> Yeah. >> Uh people are sometimes quite lazy and
57:19 if you give them a tool, they will just kind of copy and paste it in and they're
57:24 going to get into a lot of trouble. The thing though is what I was surprised by
57:27 as well, Jason, is the diversity of business plans for these companies. It's
57:32 not all Harvey clones. Uh just for example, uh Eve and Up, which raised 150
57:38 million that Bessemer led, is working on just the personal injury legal space.
57:43 Eve wants to help people that are at the plaintiff side of legal cases. And uh
57:48 Nexel is actually building kind of like GTM software for for legal firms. So
57:53 we're seeing a kind of a multivariant approach here. I don't know if all these
57:57 will end up being companies versus uh features, but it's not just the same
58:01 idea and then nine clothes, which I think is very encouraging.
58:05 >> You know, I feel like talking to editorial director Lon Harris. I I've
58:09 been seeing that this Warner Brothers, which I sold out of, I took a bath on my
58:14 J trade. >> Hey, Lon, how are you doing? Good to see you uh from around the world. Tell us
58:20 what's going on with Warner Brothers. >> Sure. Well, so Warner Brothers, they
58:23 sort of announced a few weeks ago that they are considering offers, basically
58:27 putting themselves up for sale, announcing to the world they they are
58:32 planning to uh sell the company or possibly to split up the company. That's
58:36 what it's more and more looking like. So, we've had Paramount, the newly
58:41 merged Paramount/SK Skyance, the Ellison family of course took that over. Uh so,
58:45 now they have run Paramount. They were thinking about trying to take on Warner
58:48 Brothers Discovery. There was some concern that this would mean two of our
58:54 largest sort of news organizations, CBS News, which is owned by Paramount, and
58:57 then CNN, which is owned by Warner Brothers Discovery, would both be in the
59:02 hands of the sort of Trump aligned or at least loosely Trump aligned Ellison
59:07 family. But for so far, Warner Brothers has rejected those offers. They're too
59:11 low is what the trades are saying. So now Netflix has hired an investment bank
59:16 and sort of is sort of circling Warner Brothers and trying to see if there's a
59:20 deal. Now Ted Sarando's kind of talked around this during their last earnings
59:24 call last week. He's suggesting the interest of Netflix is entirely 100% IP.
59:30 So Warner Brothers owns I'll give you a brief list. DC Comics, all of the HBO
59:36 shows, Harry Potter, Barbie, Looney Tunes, The Monsterverse, that's that
59:40 Godzilla King Kong franchise. it, the Stephen King franchise, Lord of the
59:45 Rings, Mortal Kombat, Final Destination, Bladeunner is also in the Warner
59:49 Brothers. So that's what Netflix is looking at and uh licking their chops.
59:53 Like they want to make Batman shows, they want to make a Harry Potter movie,
59:56 they want to make all of this stuff. Now, what Sarandos is suggesting they're
60:01 not interested in is the cable TV business and all of that infrastructure.
60:05 So, in addition to CNN, Warner Brothers, Discovery owns Discovery, TBS, HGTV,
60:11 Food Network, TLC. This was all that stuff Zazlab was so excited about, like
60:15 the 90Day Fiance verse they were launching on HBO Max. That stuff Netflix
60:19 not interested in. They don't want all they want to make their own reality
60:23 shows. So, they're thinking of maybe they would buy the Warner Brothers movie
60:29 and TV studio and then spin off the cable networks into their own something
60:34 else. They should go obviously they should buy the iPad. It's a no-brainer
60:38 and they can outbid everybody except for Apple, Amazon. And where is Apple and
60:44 Amazon and all this? Like Apple owning HBO is the obvious uh best for it because Apple likes
60:50 prestige. >> Yeah. When you could see these like hit Apple TV shows would be right at home.
60:56 They're trying to make HBO content anyway. Stuff like Severance, Silo, Slow
61:01 Horses, those are all just HBO shows that ended up on Apple. Yes, very
61:05 precious, very manicured, very well thought out. Very orur, I guess would be
61:10 the best word. Very orur silo was good. Those books are fantastic and they did a
61:14 pretty darn good job bringing it to the screen. Yeah. >> Yeah. So, I don't No spoilers, please.
61:20 So, I just here's what I First of all, I just want to say my JRA got crushed.
61:25 >> I lost $24,000 on this effing stock because I sold it at
61:30 >> eight bucks. What is it at right now? Oh, what's Netflix today? Well, they
61:34 >> No, not Netflix. I kept my Netflix Warner Brothers 22.
61:37 >> I bought it at 16. >> I bought it at 16. I sold it at 8
61:41 because I was so disgusted. I had lost money on Disney. I had lost most money
61:46 of all my J trades. The biggest loser was Warner Brothers >> and I love Zaz Love and I love these
61:51 assets. But I was just watching this thing for two years go down and down and
61:55 I said, you know what? I have all these other stocks going up and up. I had to
61:59 sell some winners and I had a big profit. >> So I was like, "Okay, let me use some of
62:05 these losses to offset it." So I sold it. >> But if I had held it instead of losing
62:11 $8 a share and 24 grand, you're telling me I would have made $6 a share and I
62:15 would have made like 20 grand. >> But it doesn't matter because I was
62:18 offsetting losses. It's all good. Um, they should keep CNN. This is the one
62:24 piece that's a wild card. I predict they keep CNN. Why? There are people who love
62:32 CNN and having CNN as part of Netflix would give you a reason to keep it open
62:35 all the time. >> Yeah. >> And it's global. >> Now, Netflix has a global business.
62:44 There's CNN in every country around the world. And if there isn't, whatever
62:48 pockets where there isn't CNN, what an amazing opportunity to expand the
62:52 franchise. Now, the reason Ted might not do it is because of politics and
62:58 toxicity and stepping in it. But if they went back to CNN just doing straight
63:04 reportage and not having like a bunch of emotional, which they've tried to do,
63:10 Zazlov has moved CNN in the Trump 2.0 era to really trying to get back to a
63:18 centrist, non opinionated. and they they hired um that right-wing guy who just
63:24 dunks on everybody on when he's on panels and they've kind of deprecated
63:28 having it just be all >> yeah, >> you know, Trump derangement syndrome,
63:32 whatever. Um and they they tried to just be more even killed, but I think maybe
63:38 they don't want to be involved in anything as toxic as politics and news.
63:45 I think it's a mistake. I think you play the long game. You got three years to
63:49 navigate the end of the Trump era or seven, but I'm gonna go with three. See,
63:54 polyark. Does polyart have Trump take a third term? I'm on it.
63:57 >> Oh my god, [laughter] >> that's the one I want to put money on.
64:02 >> Now Trump is he he was he was saying he was doing all of that sort of stuff
64:05 around with the hats or whatever, but now he's sort of saying he was just
64:09 joking and he doesn't really think he's going to do a third term.
64:11 >> So he's definitely doing it then. >> Well, that's what Trump is saying this
64:14 week. But who knows if you can believe him. 4% on poly markets presidential
64:19 election winner 2028 poll okay Donald Trump 4% Marco Rubio 5% to give you how
64:23 much faith the market currently has in a third Trump JD on top Gavin second AOC
64:27 third >> I think a now wait I can't bet against AOC no I can bet no on AOC you can you
64:33 could buy >> you can bet yeah or no on >> there's no way AOC is winning president
64:38 in 2028 right that's not happening >> it doesn't feel super likely to me I
64:42 mean like we the Democrats that are really finding a lot of enthusias IASM
64:46 for the base. They are do tend to be sort of left, you know, like Mandani,
64:50 Bernie Sanders, AOC, they're the ones that have a lot of hype around them
64:53 right now, but that doesn't necessarily mean that doesn't mean going to win a
64:57 40. No, I don't I don't think I don't think that >> you have to win the This means win the
65:00 nomination and win the presidency. You got to make it to the finals,
65:04 >> win the general, and then you >> Yeah. Yeah. It's not It's not happening.
65:06 >> Yeah, I don't. >> So, what do you think of my idea here?
65:10 >> Who's Ivanka? >> Me. >> Are you on here?
65:14 >> Uh, no. I don't think so. >> All right. So, what do you think of my
65:18 idea with CNN? >> Well, I think what's so interesting about it is just if you remove the
65:24 political angle entirely, CNN is one of the most established live brands in the
65:28 world. And that's exactly what Netflix has been trying to position itself. They
65:33 want to be you watch live events on Netflix, not just BOD, Squid Game,
65:37 whatever. They want you to watch boxing matches. They want you to watch
65:42 football. So like live news, 24-hour news seems like a perfect complement to
65:46 that. So I think it it is an interesting idea. >> You should buy CNBC as well.
65:51 >> Well, I I think the the idea that you're going to step in and say we don't want
65:55 all media to be in the hands of a few people. It kind of there's a hero
65:59 narrative in this for Netflix as well. Like we're going to step in and save
66:04 American journalism. It was all gonna go to this one family was going to control
66:08 all these news networks. But now we're gonna take it and we'll step up and the
66:12 other option that we >> is the Ellison who already own Paramount
66:17 Sky Dance. They already own CBS. So they get CNN too. That's a lot. And actually
66:21 David Zassab has spoken to this pretty specifically. Last week he was at the
66:25 Simon Weezenthal Centers humanitarian dinner giving a keynote and he said
66:30 quote when the government controls the news that is the end of democracy. So,
66:35 he didn't name names, but everybody sort of feels like that's what he's saying,
66:38 that he's being pressured. And there has been, now none of this is confirmed, but
66:42 there has been some suggestions in the media. CNN reported on this actually,
66:48 that the Trump administration has been signaling that they may not give
66:53 approval, regulatory approval to any acquisition that isn't the Ellison's
66:58 taking CNN. Like, they wouldn't, the FCC would shut down a merger between Netflix
67:02 and CNN or whatever. That's just the room. We don't know if that's true.
67:07 >> Uh yeah, I think it's a pretty good point. And today, uh in a transcript of
67:10 the 60 Minutes interview that Trump did, which is now part of CBS, which he
67:13 helped facilitate the transaction with the Olson and so forth, um he
67:16 specifically praised the new editor-inchief of CBS and so forth. So,
67:20 I think we're getting into Hungary territory. >> We're going full orbin,
67:24 >> right? And I think I think we can sort of see that the Trump administration
67:28 would prefer that CNN end up in the hands of an allied sort of group and you
67:33 know Netflix or another loose cannon company. Although not not that Netflix
67:37 is necessarily Trump's worst enemy. I mean I don't think any of these
67:41 companies would go at him necessarily. But anyway, it's an interesting
67:45 >> shout out to Simon Whisinthal. Uh not people don't remember but
67:51 >> he was like the greatest Nazi hunter. >> Yes. of all time brought,00
67:56 Nazis to justice. [applause] While I have you here, >> best Nazi hunter film
68:05 >> or film thread line >> or TV show? >> I mean, I would say
68:09 >> I have my own, >> but I want to hear yours. >> Yeah. I mean there is one of my favorite
68:17 movies of all time is a 1969 French movie called Army of Shadows uh by a guy
68:21 named Jean Pierre Melville is the filmmaker and it's just a French
68:24 resistance. It's a bunch of different stories interconnected about members of
68:29 the French resistance during the Nazi occupation and it is incredible and it's
68:34 like it's a great movie to watch like right now because it does have that kind
68:38 of like it's very motivating and it's very exciting and it's very compelling
68:42 um just like following along on these little missions and how they sort of
68:45 went about sabotaging and what they would do if they got captured and all
68:48 this stuff. really great movie that's about uh I guess Nazi evading and Nazi
68:54 thwarting more than Nazi hunting. I do really like that. I'm trying to remember
68:57 which one it was. I think it's >> The boys from Brazil is the classic.
69:01 >> That's a great one. That's a great >> The boys from Brazil. You ever see that?
69:04 >> Sure. >> No, I know. Line [laughter] >> sounds honestly I mean I'm not going to
69:11 lie. I I don't think I've even heard of this. Uh, but here is
69:16 >> Boys from Brazil is a 1978 obviously in Glorious Bastards. Um,
69:20 the cover of the Boys from Brazil. >> The Boys from Brazil is amazing. I was
69:24 say there's also Lawrence Olivier. >> There's that great X-Men first class
69:29 sequence about Magneto. Magneto being a Nazi hunter. Yeah. >> Was so great when he hunts them down in
69:36 their like Bavarian village in Brazil, >> right? Yeah. He finds him in South
69:39 America. Yeah. >> Yes. I forgot about I was >> great scene which I think is inspired by
69:44 boy with the coin. I think that's inspired by Boy Zimmer. >> For sure. It for sure was. Yeah.
69:50 >> I just think there's a big TAM for uh Nazi hunting videos and movies. So I
69:53 think some startup should make more of these because apparently we can just
69:56 there's a lot of good ones. >> Well, the great thing I mean and
69:58 Spielberg talked about this when he made Raiders of the Lost Arc. The great thing
70:02 about making Nazis your villain is you could do anything that people don't
70:05 care. You could blow away a hundred of them. James Gunn actually has been
70:08 having fun with this on Peacemaker, the HBO Max show. Like if you set up that
70:13 your bad guys are Nazis, you can gun down a whole field of them and
70:16 everybody's like, "Hooray! Bravo! Good job." >> Melt their faces with the Ark of the
70:18 Covenant, >> right? Nobody feels bad no matter what gruesome, awful thing you want to do to
70:24 them in your movie because they're nuts. >> All right, folks. I think I think it's
70:29 time for me to hit the hay. >> But before I do, maybe I should do one
70:33 founder question since you're here, Lon. >> Oh, sure. Do you have a great founder
70:36 qu? Because I literally spent the whole day talking with founders. I did the
70:39 founder university presentation that you helped uh punch up.
70:41 >> Sure. Yeah. Yeah. >> So I did that for an hour this morning.
70:46 I walked it was great. People loved it. So great job on the presentation.
70:49 >> It's a good presentation. Yeah. We went through it. Kabir and I spent a long
70:53 weekend going through that uh slide by slide. >> Well, you guys worked over the weekend
70:55 for me. Thank you for that. That's right. >> Yeah. Kabir Bianca, too. Bianca helped.
71:00 She's been very busy in in the kingdom, but she helped out as well.
71:02 >> Yes. >> All right. All right, so I do have a good founder question here from warm
71:08 salesale7908 on r/startups. Uh is they're asking founders who've
71:12 raged angel or seed rounds. How do you track runway and burn rate and how
71:17 painful is it creating investor updates? This was such a good question because
71:21 it's so it it it's so directly a founder university question that we hear from
71:24 all the time. They wrote, "I'm researching how founders manage runway
71:28 tracking and investor updates after raising angel or seed rounds, trying to
71:32 understand if there are real pain points here, if most people have figured it
71:35 out. Would love to hear how you handle it, what tools you use, what's annoying,
71:39 what works well, and so forth." So Jason, what do you think is the best or
71:44 simplest way for early startups to track their runway and burn?
71:48 >> It's very simple. There are tons of accounting firms out there who do this
71:53 kind of work and uh they'll send you every month a P&L and they will send it
71:58 to you somewhere around like 10 days after the close of the month, 15 days
72:02 after the close of the quarter and you have to have your accounting right. The
72:07 good news is in year zero or one you may not turn on revenue or if you do it's a
72:10 pretty simple revenue. So, and your costs are usually limited to your Austo
72:16 account, like what your salaries are or and like you bought some computers and
72:21 maybe you got some credits from, you know, Google or Amazon and, you know,
72:25 maybe you're paying for some cloud as well. So, you essentially take your last three
72:34 months burn, you average it. Okay? So, let's say you burned 50K and then you
72:38 had a big month because you bought some laptops and it was 70K and the next
72:42 month you made a little more money and it went down to 40K or 30K. So, you got
72:47 50 7030K in losses in a quarter. Instead of doing it based on one of those three numbers,
72:53 you just blend the three together. You say based on the last three months burn
72:59 rate, which is revenue minus spend, positive or negative number. You're if
73:02 it's negative, it's a burn. If it's positive, it's uh your your runway is
73:07 infinite. And it's always great to get that email infinite runway. So when you
73:12 go to $1 in profit over your spend, you're now infinite runway. But if you
73:19 had a burn of 50k a month and you have 1 million in the bank, you have 20 months
73:23 of runway. 50,000 your average burn for the last three months. So you say based
73:26 on our average burn the last three months, we have approximately 20 months
73:32 of runway. And I think a short overview at the top of what you do. Hi, we are
73:37 Uber. We are everybody's private chauffeur. We operate in three cities.
73:43 We're opening up two more cities. Um, and then you might have team, product,
73:48 customers, you know, uh, fundraising or corporate activities. And you just list two or
73:55 three sentence of narrative. If you have a chart, you put in a chart. And then at
73:59 the bottom, it's always good to put requests. Why do you put it at the
74:01 bottom? You want to see who gets to the end of your rider. Famous story like uh
74:07 take the brown M&M's out of the >> Van Halen, I believe is that story.
74:10 >> Yes. >> And so they had at the bottom of their
74:13 rider, take out the brown M&M's. Why? They wanted to make people make sure
74:17 people got to the end of the rider, the rider being the conditions under which
74:21 they would play in an arena. So here you put at the bottom things you can do to
74:26 help. One, we're looking to get press in the legal press for our legal LLM
74:31 solution. This includes the following journalists and whatever. Two, we're
74:34 looking to fill these three positions. You can retweet that we're looking my
74:39 tweet here or my LinkedIn post. Number three, we're going to be raising in the
74:43 fall. We're looking for connections to these series A firms. Here's a list of
74:46 the companies. Here's a list of the firms we've met with. Please put your
74:50 name on it if you know anybody at the firm and who you know. So you like a
74:53 call to action, something for your investors to do. Monthly is great. If
75:00 you do it monthly, you will have a deep, meaningful relationship with your
75:03 investors. If you do it quarterly, you're going to have a great
75:05 relationship with your investors. If you do it less than quarterly, your
75:12 investors might feel, unless you're just having massive success and they're
75:15 seeing you do crazy fundraising and reading about your press, they might
75:20 feel like maybe you're ungrateful. Um, and the only time you do email them
75:25 is when you've sent updates. So, a lot of times we'll get an email, hey, we
75:28 have three months of runway and I haven't heard from the founder in a
75:30 year. >> So, we literally, you know, try to train our founders to do this. We put it in
75:37 our documents even like as part of our agreement. We have information rights
75:41 and we'll get a quarterly update and we like define what we're looking for.
75:46 Not because we need it as much as you do, but because as the
75:53 founder does, but because we want to see the founder succeed.
75:55 >> Yeah. >> And if you're not a good person with
75:58 sending updates, like I'm not because I'm so pulled in different directions
76:03 for my updates for my venture funds, I deputize Jackie and our team, a partner
76:08 and our company to write them and send them. And if I I have a time to edit
76:11 them, if I don't have time to edit them, they go out from Jackie and I'll read it
76:15 when other people read it, which is actually in some ways like even more
76:19 honest because I didn't see them. So I like the idea of even deputizing
76:22 somebody on your team or two people on your team. You're responsible lawn for
76:28 the product and this and Alex, you're responsible for this and this. And then
76:32 if you're not the type of person who stays on top of these things, you
76:36 forget, deputize your other co-founders or somebody on your team and say, "I
76:40 want you to write this update in this format by this date. Have all the
76:44 numbers in and then I will write an opening paragraph. If I don't write the
76:47 opening paragraph by this date, please send it and just send it from you, the
76:51 VP of operations or the general manager of the startup or the VP of product."
76:56 Um, they're really important. You don't need software to do it. You can find a
77:00 million templates online, but short is better than long and short is much
77:05 better than nothing. [laughter] And no numbers means no numbers means bad news.
77:11 >> Yeah, >> numbers typically mean good news. So
77:15 when you don't give numbers, people just assume, oh man, this startup is screwed.
77:19 Or you don't hear from people. It's a 90% chance, as I wrote in my book, like
77:23 if you don't get updates, it's a 90% chance it's going to zero or 10% chance
77:28 it's Uber or Robin Hood. But in truth, Travis sent regular updates. Even Travis
77:33 sent regular updates. Why? Things were going so well, he kind of wanted to
77:37 connect with investors and he had requests of us of things we could do to
77:40 help and he was putting us to work. So, put your investors to work. Great
77:43 question and I hope that's helpful. >> All right. >> I'm sure was.
77:45 >> Well, Jason, let's get you to bed overall there in the Middle East. lawn,
77:48 you and I are going to get back to work, but this has been another This Week in
77:51 Startups. We'll see everyone live on Wednesday. Bye. >> Bye.
$

Sending Data Centers into Orbit! | E2203

@thisweekinstartups 1:17:54 18 chapters
[AI agents and automation][hardware setup and infrastructure]
// chapters
// description

👉 Register here to join the Founder University Japan kickoff! https://luma.com/cm0x90mk * Today’s show: *Starcloud has started sending AI chips into orbit. So is this where the datacenters the power AI will one day live? And does wifi work in the vacuum of space? We’re asking the TOUGH questions. PLUS Jason and Alex are doing a deep dive on all this week’s OpenAI news. Will they ever be able to pay back all these GPU purchases? What happens to the wider economy on the off chance they don’t t

now: 0:00
// tags
[AI agents and automation][hardware setup and infrastructure]