Latest / In the Loop AI / This Week in AI: $2T Market Crash, Anthropic VS OpenAI, SpaceX Merger

Transcript

This Week in AI: $2T Market Crash, Anthropic VS OpenAI, SpaceX Merger · In the Loop AI

0:00
Josh: There are some weeks where nothing happens and some weeks where it feels like
0:03
Josh: years worth of news happens.
0:05
Josh: This is one of those weeks. There is so much going on from the market crashing
0:09
Josh: to Anthropic and OpenAI beefing over a Super Bowl commercial.
0:12
Josh: There's a lot of crazy frontier stuff, but where I want to start in particular
0:15
Josh: is with the markets because my God, they're getting destroyed.
0:18
Josh: I think the total market cap of the S&P 500 is down $2 trillion over the last 48 hours.
0:25
Josh: Microsoft, the fourth most valuable company in the world or something like that,
0:29
Josh: they're down 20% in the last couple of weeks.
0:32
Josh: I mean, things are getting hammered, but this sits at conflict with what the
0:36
Josh: companies are actually saying.
0:38
Josh: We just went through Google's earnings report yesterday, and they reported record
0:41
Josh: blowout earnings across the board. And this is true for a lot of other companies.
0:44
Josh: So clearly, there's this mismatch happening between public sentiment and what
0:48
Josh: the actual companies are reporting.
0:50
Josh: And I'm curious, I think we're going to spend a lot of time this episode getting
0:53
Josh: into why and how to kind of navigate that as we move through the rest of earnings season
0:56
Ejaaz: Yeah if we look at the mag 7 stock price
1:00
Ejaaz: over the last week um they're down between
1:03
Ejaaz: five to eleven percent with
1:06
Ejaaz: this chart is horrific well yeah with the with the most valuable company
1:09
Ejaaz: nvidia being down the most um of the
1:12
Ejaaz: week which is just shocking right um so a lot of people are
1:15
Ejaaz: calling this the end of the ai bubble
1:18
Ejaaz: it's popping um you know that is over and um
1:21
Ejaaz: there's an argument that a lot of this is triggered by to your point the
1:24
Ejaaz: earnings reports that have been released over the last week we had microsoft
1:27
Ejaaz: and a few others last week we had google yesterday and we got amazon um today
1:31
Ejaaz: and what's crazy is like they're beating expectations but they're increasing
1:36
Ejaaz: capex spend so i think the markets are at odds with this and they're kind of
1:40
Ejaaz: like selling off out of fear that this might be an issue but
1:43
Ejaaz: The obliteration is like pretty significant. You mentioned Microsoft.
1:46
Ejaaz: It is down 21% over the last two weeks.
1:52
Ejaaz: NVIDIA is down at 11%. Tesla is down 15%. But there is one company,
1:57
Ejaaz: Josh, which you'll never guess, that is outperforming all of these guys over
2:02
Ejaaz: that same time span. This is insane.
2:04
Ejaaz: 7% over the last week. We have Apple.
2:08
Josh: Apple. Could you believe that? Apple, the company that failed to participate
2:12
Josh: in AI, is the winner of the AI market race this week, at least,
2:16
Josh: over the last, what is that, five days?
2:18
Josh: It's up six and a half percent relative to everyone else.
2:21
Ejaaz: Josh, why do you think this is?
2:22
Josh: Because they have no exposure. Can't hurt you if you're not in it.
2:25
Josh: And what is their exposure to AI? A billion dollar deal with Google.
2:29
Josh: And Google already pays them $20 billion for the search rights. So, I mean, Apple's up.
2:34
Josh: And that's the thing is like, sure, I'm sure this was not calculated,
2:37
Josh: but in a way, sitting on the sidelines did play a big role in the success this week.
2:41
Ejaaz: Yeah, pretty insane. But let's get into the reason why everyone is freaking
2:45
Ejaaz: out and the markets are dumping. It's capex spent.
2:48
Ejaaz: Google is the largest perpetrator of this. In their earnings report that they
2:53
Ejaaz: announced yesterday, they're targeting up to $185 billion of capex spend alone this year.
3:01
Ejaaz: That is a gargantuan number, which they could use to acquire pretty much all
3:05
Ejaaz: but 13 or 16 companies in the Fortune 500 index, which is just an insane stat.
3:13
Ejaaz: The guided estimate that analysts had for this capex spend in 2026,
3:17
Ejaaz: by the way, was $120 billion.
3:20
Josh: 30% more.
3:21
Ejaaz: 50%. No, 50% over. If there was 120, $60 billion over.
3:25
Josh: Yeah. Oh my God.
3:26
Ejaaz: That's a lot. Insane.
3:27
Josh: Yeah. And I think this comes as a surprise because of how staggeringly large it is.
3:32
Josh: I mean, basically doubling this year.
3:37
Josh: In a world where there was steady growth, there is now this exponential growth.
3:40
Josh: And that probably causes a lot of concern for the public stock market.
3:44
Josh: But when you listen to why these companies are spending it, I think the answer
3:48
Josh: becomes clear that the scaling laws still exist.
3:51
Josh: And so long as you can put dollars in and get intelligence out at a rate that
3:56
Josh: makes sense, then they're going to continue to do this.
3:58
Josh: And for the first time ever, I mean, Google is basically going to have no free cash flow.
4:02
Josh: All the dollars that come in that they make are going right back out into scaling this.
4:06
Josh: In fact, I read this crazy statistic that the amount of CapEx spend Google is
4:10
Josh: going to have this year in 2026 is equal to that between 1998 to 2021 combined.
4:17
Josh: This is a tremendous amount of money being spent.
4:20
Josh: And I think the market could view that as like a pretty high risk vector because
4:24
Josh: they're spending all of the money that comes in.
4:26
Josh: And what if these scaling laws don't continue?
4:30
Josh: What if there is a glut somewhere in the supply chain? Like what type of downstream effects that have?
4:34
Josh: How much can it hurt a company who's betting all of it on the fact that they
4:37
Josh: can throw more money at the problem and get better intelligence?
4:40
Ejaaz: This might be a hot take, but I think they're absolutely right to be spending
4:44
Ejaaz: this amount of money. And I think the market is wrong.
4:47
Ejaaz: Look at some of the highlights from their earnings reports.
4:51
Ejaaz: Search, who everyone called the death of, is up 17% year upon year.
4:58
Ejaaz: Gemini, monthly active users, is up to 750 million monthly active users.
5:04
Ejaaz: Now, I just want to highlight that that is 100 to 150 million short of OpenAI's
5:09
Ejaaz: current monthly active user.
5:11
Ejaaz: So, you know, they're making billions, tens of billions from YouTube.
5:15
Ejaaz: Their cloud revenue is up. By the way, that's the greatest signal that their
5:19
Ejaaz: CapEx spend is justified.
5:20
Ejaaz: The fact that most people are using their cloud infrastructure,
5:23
Ejaaz: their TPU infrastructure and stuff like that to train or inference certain models.
5:27
Ejaaz: So clearly, Google sees the demand for AI products.
5:31
Ejaaz: And Sundar actually went on record to say that the thing that's keeping him up at night is...
5:37
Ejaaz: The top question is definitely around compute capacity. And he talks about power,
5:41
Ejaaz: land, supply chain constraints.
5:43
Ejaaz: He can't get enough compute out there. So there's a conflict of opinion between
5:47
Ejaaz: market investors and what the Mac 7 are actually saying.
5:50
Josh: The difficult thing to wrap your head around is any way that you look at Google's
5:54
Josh: earnings report, it is an incredible earnings report.
5:56
Josh: I mean, it's a record across the board. It shattered expectations and estimates by a large margin.
6:01
Josh: And I think people are really scared about this capex. But the reality is it's
6:05
Josh: not just Google who's participating in this.
6:06
Josh: We have this post on screen that shows a total of $680 billion between some
6:12
Josh: of the top hyperscalers that they're planning to spend just in this year of 2026.
6:17
Josh: What does it say? Almost $300 billion or 1% of GDP growth in CapEx from just these five companies.
6:24
Josh: That is insane. And it's funny when you hear Jensen Huang on his earnings report
6:29
Josh: say, we're expecting half a trillion dollars of revenue.
6:32
Josh: Well, suddenly it kind of makes sense. I mean, this is where it's coming from.
6:35
Josh: What are they going to spend that money on?
6:37
Josh: Mostly gpus mostly new chips mostly the infrastructure to support the build
6:41
Josh: out of these chips and the idea is that by the end of this year they'll have
6:44
Josh: intelligence so strong that it will offset a lot of the other downsides that can come from this
6:50
Josh: And I hope that they're right because, I mean, we've seen the effect on Google search.
6:54
Josh: Everyone thought this was the pinnacle of the innovator's dilemma.
6:56
Josh: But the reality is, is that search is up because Gemini now is sitting right
7:01
Josh: at the top of all your search results and it yields pretty amazing results.
7:04
Josh: So I find myself using Google a lot. They're winning.
7:07
Josh: The market is continuing to prove that these scaling laws apply.
7:12
Josh: And so long as the scaling laws continue to grow, dollars in intelligence out,
7:15
Josh: this is probably going to be a trend that we continue to see.
7:18
Josh: And the market is a little nervous about it.
7:20
Ejaaz: To kind of put myself in the shoes of like an average investor,
7:24
Ejaaz: I can anticipate that they're probably scared at the numbers that are being
7:30
Ejaaz: spent and the percentage of cash reserves that are being spent by their darling companies.
7:35
Ejaaz: Like these investors have sat pretty comfy for the last decade, right?
7:38
Ejaaz: SaaS company, they look at the rev chart every quarter and they're like,
7:42
Ejaaz: okay, cool, earnings per share is up, like let's invest, it makes more sense.
7:45
Ejaaz: Now to see those margins compress, You know, it's undisputedly like obvious
7:51
Ejaaz: why they're going to get scared.
7:52
Ejaaz: They looked at Meta spending all that money on the Metaverse vision way back
7:56
Ejaaz: when, and that ended up pretty badly.
7:58
Ejaaz: So they're thinking this is probably going to wind up for everyone else.
8:01
Ejaaz: CapEx spend isn't the only reason why investors are scared.
8:06
Ejaaz: I'm showing you this tweet from an official report from JP Morgan this week,
8:10
Ejaaz: which talks about enterprise software.
8:12
Ejaaz: And the argument that they make in this investment thesis is that coding models,
8:17
Ejaaz: specifically ClaudeCode, which is quoted like 50 times in this report,
8:21
Ejaaz: is going to eat away at any traditional industry that could just simply be a
8:26
Ejaaz: plugin into ClaudeCode or one of these coding models.
8:29
Ejaaz: And the reason why they're making this point is the adoption rate for CodeCode
8:33
Ejaaz: over the last two months is up more than like 350%. It's just gone completely parabolic.
8:40
Ejaaz: It's overtaken OpenAI's codex, and we have more news on that later.
8:43
Ejaaz: But the point is, it's automating about 50% to 60% of job roles for firms like
8:50
Ejaaz: law firms and SaaS companies.
8:52
Ejaaz: And the stocks are kind of responding in a very similar way.
8:56
Ejaaz: The top law firms, I think, are down $250 billion this week.
9:02
Ejaaz: And then SaaS companies on average, the top SaaS company index is down 10% to
9:06
Ejaaz: 15% over the last two weeks.
9:08
Ejaaz: So we're seeing a lot of this market reactionary stuff from updates from these AI models.
9:14
Ejaaz: But one person kind of stands against the tide.
9:17
Ejaaz: We've got Jensen Huang and a clip here from a recent appearance at the Cisco
9:21
Ejaaz: AI Summit, where he basically says that the market is getting this very,
9:24
Ejaaz: very wrong. Like, if you think about it, coding AI models aren't going to replace these functions.
9:30
Ejaaz: They're going to use the tools that a lot of these SaaS companies have built.
9:34
Ejaaz: And the market hasn't realized that.
9:36
Josh: Yeah. So we're hearing two conflicting opinions. One is through the public market
9:39
Josh: sentiment, which is clearly down only.
9:41
Josh: And the other is through the actual builders who are creating the future,
9:44
Josh: which are saying, hey, guys, it's actually going to be a pretty great future.
9:47
Josh: We're building towards something pretty amazing.
9:49
Josh: And I guess we'll see that play out over the next couple of weeks.
9:51
Josh: But now it's time for some drama because...
9:54
Josh: Yesterday was full of fireworks on the timeline, mostly due to Anthropic and
10:00
Josh: their war that they're waging with OpenAI through the Super Bowl commercial
10:04
Josh: or the series of Super Bowl commercials.
10:06
Josh: They published, I think, three, four, maybe five of these things.
10:09
Josh: And every single one is coming at OpenAI for their use of ads.
10:13
Josh: They're really funny videos.
10:15
Ejaaz: Let's show a clip. How do I communicate better with my mom?
10:19
Josh: Great question. Improved communication with your mom can bring you closer.
10:24
Josh: Here are some techniques you can try.
10:26
Josh: Start by listening. Really hear what she's trying to say underneath her words.
10:31
Josh: Build conversation from points of agreement.
10:34
Josh: Find a connection through shared activity. Perhaps a nature walk.
10:39
Josh: Or if the relationship can't be fixed, find emotional connection with other
10:44
Josh: older women on Golden Encounters, the mature dating site that connects sensitive
10:49
Josh: cubs with roaring cougars.
10:51
Josh: Yeah, so I think you probably get the idea.
10:54
Josh: Um, it's coming at them for ads and not only their ads, but how their voice
10:59
Josh: advanced voice mode sounds, which is kind of robotic and not very human-like.
11:02
Josh: And this started a war on the timeline.
11:06
Ejaaz: Yeah. The basic thing that they're getting at here is OpenAI needs to add ads
11:10
Ejaaz: and ruin their complete user experience and Claude, or Anthropic rather, never needs to.
11:16
Ejaaz: Um, listen, I watched these ads and I laughed my head off.
11:19
Ejaaz: It's hilarious. There's another ad where this guy is like doing a pull up and
11:22
Ejaaz: he asks this AI, presumably ChatGPT, like, how do I get a six pack in a week?
11:28
Ejaaz: And the guy is just like, sure, I can do this for you. And obviously that's not realistic.
11:32
Ejaaz: I don't like very much that they're taking pot shots. But me personally,
11:36
Ejaaz: I find it pretty hilarious.
11:38
Ejaaz: The other kind of like subtle thing that I don't think a lot of people picked
11:41
Ejaaz: up on here is it's a very clear statement from Anthropic that they're making
11:45
Ejaaz: enough money to pay for all the capex spends and costs that they have,
11:50
Ejaaz: which I don't think any other AI lab,
11:52
Ejaaz: independent AI lab, might I say, that doesn't have like, you know,
11:55
Ejaaz: search revenue coming like Google.
11:58
Ejaaz: It's pretty impressive. And Darryl is projecting $70 billion ARR by 2027 and profitable in 2028.
12:05
Ejaaz: So I don't know whether this is an egotistical thing, but it's certainly a confident one.
12:09
Josh: I think it's probably a lot of ego. I'm not sure the, I mean,
12:12
Josh: the amount of money these AI companies are spending relative to what,
12:15
Josh: a $50 million YouTube spot is probably fairly small.
12:20
Josh: The ads were received very well. People really liked these. I unfortunately didn't.
12:26
Josh: The Super Bowl is this opportunity, right? You have 150 million concurrent viewers
12:30
Josh: and what they've done is decided to spend all of this money and awareness on
12:33
Josh: taking down other people instead of introducing
12:36
Josh: introducing new users to ai in a world where it's
12:39
Josh: mostly met with resistance i think a lot of what we're seeing
12:42
Josh: in the public market is people getting nervous because they think
12:44
Josh: it's just this one big bubble that's going to harm a lot of like the local economies
12:49
Josh: through the data center build-outs and it's not actually going to generate enough
12:51
Josh: value and it's an opportunity to introduce the technology in a way that helps
12:57
Josh: bring those people on board but instead they're just kind of tearing each other
13:00
Josh: down through this like trolling ad thing so
13:05
Josh: Not my favorite way of displaying ads, but I mean, nonetheless,
13:08
Josh: it will certainly get attention. And it attracted some attention,
13:12
Josh: right? From who is this OpenAI CMO?
13:15
Ejaaz: Yeah, Katie Rauch here claims that the anthropic ads are kind of childish and
13:20
Ejaaz: inaccurate. And in her defense, she's kind of right.
13:23
Ejaaz: OpenAI has explicitly said that they're not going to bake in sponsored advertisements into responses.
13:29
Ejaaz: And any kind of adverts that are placed into their responses will be highlighted as such.
13:34
Ejaaz: So it specifically won't be in a voice feature like what these Super Bowl ads show.
13:39
Ejaaz: So it's kind of deceiving. And I see Katie's point, but it's also kind of,
13:43
Ejaaz: it comes across as insecure from OpenAI.
13:46
Ejaaz: I wish they had just kept quiet and maybe posted their recent like codec signups,
13:51
Ejaaz: for example, which just hit like a million. I don't know, Josh,
13:53
Ejaaz: do you have a different opinion?
13:55
Josh: Yeah, I'm just so disappointed in both of these companies. My God,
13:57
Josh: they have such an opportunity to create a positive looking platform for AI in
14:03
Josh: a world that really needs it.
14:04
Josh: It's like crypto went through this phase where it was the most hated thing in
14:07
Josh: the world. And it still hasn't quite reached that adoption because they haven't
14:10
Josh: convinced people why it's good for the world.
14:12
Josh: And AI very clearly is a tool made for good. And this is such an amazing opportunity
14:17
Josh: to introduce so many people to that idea.
14:19
Josh: And instead, they're using it to fight against each other. And it's so exhausting.
14:24
Josh: It's so tiresome. And it feels reflective of companies that don't have leadership
14:28
Josh: with clear values or vision.
14:30
Josh: Because I think about old school Apple and their marketing department.
14:34
Josh: And granted, they're the biggest winner, not because they're doing the right
14:36
Josh: thing right now, but because they've actually missed the boat.
14:39
Josh: They're washed up and they're not what they used to be. But in a previous life,
14:42
Josh: Given the opportunity to have a platform to reach 150 million concurrent viewers,
14:46
Josh: it's hard to believe that they wouldn't use that as a platform to inspire new people.
14:51
Josh: And to get them excited about the new products they were offering that allowed
14:54
Josh: them to leverage this technology that's indistinguishable from magic.
14:57
Josh: And all we're seeing is...
14:59
Josh: Like these companies kind of tearing each other down and it's exhausting,
15:02
Josh: it's zero sum, and it's a testament to the focus of these companies,
15:06
Josh: which is being better than the other one.
15:08
Josh: And I guess signaling that you are better than the other one in exchange for
15:12
Josh: earning more dollars so you could pay off your debt so you could continue to scale.
15:15
Josh: And I don't know, OpenAI, they have a Super Bowl commercial that they teased,
15:19
Josh: that Sam Altman teased, and it seems to be more directionally what my preference
15:24
Josh: would be, which is just optimistic about the tool and the tools and the builders
15:28
Josh: and all the things associated. with that.
15:30
Ejaaz: I think OpenAI has had a shift of culture and direction over the last three months.
15:35
Ejaaz: They called it Code Red, what was it, two months ago when Google released Gemini
15:39
Ejaaz: 3 and it was like marginally better than ChatGPT.
15:43
Ejaaz: So Sam decided to focus on two main things, making a better generalized AI model,
15:49
Ejaaz: an LLM, a chatbot, and the new best coding model.
15:52
Ejaaz: And we're starting to see the emergence of that this week. We have a new Codex
15:59
Ejaaz: model app specifically that can run on your desktop, similar to Claude Code's desktop app.
16:06
Ejaaz: But it's cool for a few different reasons.
16:09
Ejaaz: Something that I can pull off the top of my head that I thought was awesome
16:11
Ejaaz: is you can now create a bunch of coding agents and run them in parallel.
16:16
Ejaaz: The other second cool feature is that it has a bunch of automated functions
16:21
Ejaaz: that you can run in the background. So even whilst you sleep overnight,
16:24
Ejaaz: you can now have this coding tool do a bunch of stuff for you.
16:28
Ejaaz: Now, it's worth saying that this is something that Claude Code can kind of already
16:31
Ejaaz: do, but I'm happy to see that OpenAI is finally investing in the things that
16:36
Ejaaz: I think are super important. I think they're a little late to the game.
16:38
Ejaaz: I think they got distracted with social media stuff and hardware stuff and chip
16:42
Ejaaz: stuff, but it's nice to see that they're currently focusing and shipping right now.
16:46
Josh: Yeah, I think there's two schools of audience for this, for the people who are
16:50
Josh: listening. One is like if you are a casual user of AI who wants to experiment,
16:55
Josh: the Claude desktop application is probably your best bet.
16:58
Josh: It has chat it has agent and it
17:01
Josh: has code all under one roof if you are
17:04
Josh: serious about coding or serious about creating projects or
17:07
Josh: developing new software codex is a pretty compelling product and i would strongly
17:12
Josh: encourage downloading it one because the model is fantastic and the app is actually
17:16
Josh: pretty good but two because they doubled the usage limits so one of the problems
17:20
Josh: with using cloud code is if you're not paying 200 a month even if you are you
17:23
Josh: run into the wall of token limits pretty quick.
17:26
Josh: Codex, the limit is much higher and the outputs are much stronger.
17:29
Josh: So if you're someone who does actually write code for a living,
17:31
Josh: I think you'll find a lot of success and a lot of pleasure that comes from using Codex.
17:36
Josh: And it's reflected in the downloads, right? Like they have a huge amount of downloads.
17:39
Ejaaz: I was about to say, this tweet says there's been over 500k Codex app downloads since Monday.
17:44
Ejaaz: It's actually more than that right now. They just hit 1 million Codex downloads
17:49
Ejaaz: and users right now, which sounds very impressive, but I think is less than like 0.1,
17:55
Ejaaz: not 0.1, but like 0.5% or less than a percent of their entire monthly active user base.
17:59
Ejaaz: But to your point, Josh, not everyone's going to be using this thing.
18:03
Ejaaz: This is for a specific type of audience that wants to kind of code either seriously or vibe code.
18:09
Ejaaz: Now, a final point that I want to make on this is I've spoken to a bunch of
18:13
Ejaaz: really talented software engineers recently, and they repeatedly claim that
18:17
Ejaaz: Codex is a better coding agent than ClaudeCode.
18:21
Ejaaz: They say ClaudeCode produces more bugs in serious software development,
18:25
Ejaaz: but it's a much better orchestrator.
18:27
Ejaaz: So the setup that a lot of these engineers have is they set up Claude Code and
18:33
Ejaaz: they get it to spin up instances of OpenAI's codecs and manage it.
18:37
Ejaaz: So in a weird way, although these companies are at odds with each other,
18:40
Ejaaz: although they're releasing Super Bowl ads that are fighting each other,
18:43
Ejaaz: they're kind of working best synergistically.
18:45
Ejaaz: And I just, the winners are the software engineers coming out of all of this.
18:49
Josh: Yeah. I mean, as always, the user wins big time. And I think...
18:53
Josh: Doubling the usage limits is a perfect example of that where they
18:56
Josh: want a compelling reason to go over and that's what
18:59
Josh: we're getting is double the amount of token usage an interesting
19:02
Josh: thing that's happened with chat gpt recently or open ai products in
19:05
Josh: general is there's a lot of downloads to be had there's like
19:08
Josh: the chat gpt app there's sora there is their new research
19:11
Josh: app there is now codex app they've really unbundled
19:14
Josh: a lot of their offering and segmented it into separate
19:17
Josh: silos which i think is probably confusing and difficult uh again
19:21
Josh: the claw desktop app has everything under one roof
19:24
Josh: it's very convenient i think we'll start to see the actual
19:26
Josh: user experience of these things matter and that's one way in which it very clearly
19:29
Josh: matters for the average user having it all under one roof is a nice perk but
19:33
Josh: now going to outer space news we earlier this week covered spacex and xai merging
19:39
Josh: together spacex acquiring xai to turn it into a single company prior to IPO-ing.
19:44
Josh: This is going to be the biggest IPO of the week. I highly suggest watching that
19:48
Josh: episode if you want to get up to speed. But there are some new developments
19:51
Josh: in this story, right, AHS?
19:52
Ejaaz: Oh, yeah. So it was announced, I think, a few weeks ago that Elon Musk or SpaceX
19:59
Ejaaz: specifically had filed to the FCC,
20:01
Ejaaz: which is basically the Space Bureau, a request to launch one million satellites
20:07
Ejaaz: into lower orbit in order to
20:09
Ejaaz: set up a constellation which will make the first AI data center in space.
20:15
Ejaaz: The idea with this constellation is that it'll harness the energy directly from
20:19
Ejaaz: sunlight, churn and transform that energy into compute, which can be used to
20:24
Ejaaz: train AI models, presumably the newer Grok models that come up.
20:28
Ejaaz: And the reason why this is happening is the entire pitch or thesis of SpaceX,
20:32
Ejaaz: which you should check out in the episode, which is, you know,
20:35
Ejaaz: they think that it's going to
20:35
Ejaaz: be the cheapest and most efficient way to train AI models in the future.
20:39
Ejaaz: But the major update from this, Josh, is that it got approved so legally.
20:44
Ejaaz: Legally, we can now launch these million satellites and we're off to the races.
20:48
Josh: It's going to happen. There's no doubt in my mind that this is going to happen.
20:51
Josh: And it'll happen before the end of the decade.
20:53
Josh: In fact, Elon just went on record again saying that within the next 36 months,
20:56
Josh: he strongly suspects he'll be able to get Starship working at scale at a level
20:59
Josh: required to start putting these AI data centers in space.
21:02
Josh: But I also saw a headline that seemed a little confusing to me.
21:06
Josh: And we talked about briefly before the show, but I really want to dig into it
21:09
Josh: now because I don't know if I can get behind this one.
21:11
Ejaaz: I wondered if you saw me sneak this one and Josh Shoy to surprise you.
21:16
Ejaaz: But yes, it's reported from Reuters that Starlink might be launching their own phone.
21:23
Ejaaz: Now, you and I got into a bit of an argument as to what phone means to me.
21:28
Ejaaz: It's a cell phone. It's like, I'm not buying the latest iPhone.
21:32
Ejaaz: I'm buying the new Starlink smartphone.
21:34
Ejaaz: But you had a different idea, Josh. Walk me through it.
21:37
Josh: There's absolutely no way SpaceX is making a smartphone.
21:40
Josh: I don't know how more clear I can be about this. It's not going to happen.
21:43
Josh: Don't get your hopes up. They are not making a smartphone.
21:47
Josh: What I suspect happened is some rumor got out that they were making some sort
21:51
Josh: of mobile device, and that was interpreted as a smartphone.
21:54
Josh: The complexities required into making a smartphone are so high,
21:58
Josh: particularly in a world in which they need to start learning how to make chips
22:01
Josh: through this Tesla fab, I don't think there's enough resources in the Musk industry
22:05
Josh: universe to allocate to building a smartphone, nor are there an incentive to actually do so.
22:09
Josh: What I think is more likely is this mobile device is probably something similar
22:14
Josh: to an old school mobile hotspot.
22:16
Josh: The newest opportunity for Starlink is direct-to-cell connectivity.
22:19
Josh: So as they deploy these million satellites, a lot of that cluster is going to
22:23
Josh: be Starlink V3 satellites, which have direct-to-cell capabilities,
22:26
Josh: meaning if you have a smartphone or a device with a capable chip,
22:30
Josh: you can get Wi-Fi or a cellular connection anywhere in the world that Starlink reaches.
22:35
Josh: There are no dead zones. A lot of devices will not have support for this.
22:38
Josh: So what is going to need to be the bridge?
22:40
Josh: Some sort of mobile hotspot device that allows you to connect to this direct-to-cell service.
22:46
Josh: And I think that's probably what this is. For them to make a smartphone,
22:49
Josh: it doesn't make any sense at all.
22:50
Ejaaz: Yeah, I have to. Listen, I'm bullish about a lot of ideas, and especially if
22:55
Ejaaz: it has Elon at the helm, there's a high chance that it's probably going to get like real.
22:59
Ejaaz: I don't buy this one. I think if it does become a consumer hardware device,
23:04
Ejaaz: it's going to be something that is used or distributed in like kind of like the world nations,
23:09
Ejaaz: kind of replacing the average old Nokia brick phone to give people basic access
23:14
Ejaaz: to the internet and apps.
23:16
Ejaaz: And of course, it'll come pre-installed with Grok and a few other of XAI or
23:20
Ejaaz: SpaceX or whatever the entity is called going forwards.
23:23
Ejaaz: There is actually a final bit of news on the SpaceX front, Josh,
23:27
Ejaaz: which I don't know if you saw, but for the first time in, I believe, I think it's 50 years,
23:33
Ejaaz: ticker X became available.
23:36
Josh: I think it was 125 years, actually. It was a huge amount of time.
23:39
Ejaaz: 125 years. Oh my God. And then literally a day later after that news broke,
23:44
Ejaaz: it had recently got acquired by a secret entity.
23:47
Ejaaz: Now I'm speculating here, but if I had to bet, it's going to be SpaceX, essentially.
23:53
Ejaaz: I think it's going to be the universal ticker.
23:56
Ejaaz: Elon Musk saw the vision when he acquired, or rather when he founded PayPal.
23:59
Josh: Yeah united states steel was the previous
24:02
Josh: company that owned it and um it appears that they
24:05
Josh: have gotten acquired or they got merged it was the uh
24:08
Josh: yeah first time in 124 years after u.s steel so
24:11
Josh: it seems poetic that now is the
24:14
Josh: time that they can claim that handle and go public with the x handle and then
24:18
Josh: who knows maybe spin them all together but for those of you who bet on that
24:21
Josh: on polymarket congratulations you are looking like a pretty hefty payday now
24:26
Josh: we have some sci-fi news right this was this was a really cool thing for me
24:31
Josh: do you want to introduce this topic
24:33
Ejaaz: Well, I want you to introduce it because I'm used to Palmer Luckey tweets being all about U.S.
24:38
Ejaaz: Weaponry and bioweaponry. Why is he tweeting about mouse brains?
24:41
Josh: Okay, this is Palmer Luckey, the founder and CEO of Anduril,
24:44
Josh: which is the largest private defense contracting company in the U.S.
24:48
Josh: And they normally make missiles and rockets and drones.
24:52
Josh: But Palmer is talking about this biological computer that was used in one of
24:57
Josh: these racing games that they play. So there's a racing competition where some
25:01
Josh: sort of rules take place. But the noteworthy thing is that one of these racing
25:05
Josh: devices was not like the other.
25:07
Josh: And it's because it used a biological computer.
25:10
Josh: And what is a biological computer? Well, it's exactly what it sounds like.
25:13
Josh: It takes actual brain cells from a living thing and trains them similarly to
25:18
Josh: how you would train a large language model, which is really interesting.
25:21
Josh: So in this case, they took the biological brain cells of a mouse, they put them on a chip.
25:26
Josh: And they train those neurons the same way that you would train an LLM.
25:30
Josh: Meaning when presented with an obstacle in front of you, they were to train
25:35
Josh: those neurons to fire in a way that maneuvers away from the obstacle.
25:38
Josh: And through this reinforcement process, through this training of what's good, what's bad,
25:44
Josh: giving values to each one and slowly iterating through the progress,
25:48
Josh: they were actually able to make it work in a way that is kind of indistinguishable
25:52
Josh: from a silicon-based LLM.
25:55
Josh: And to me, I think it's really fascinating because it shows that there is
25:59
Josh: More possibilities when it comes to building compute
26:02
Josh: there's more efficiency gains to be had because when
26:05
Josh: you think about the optimal state of memory compression
26:09
Josh: right and the ability to like solve cognitive problems
26:12
Josh: the brain is far more efficient and far more compressed and just capable than
26:18
Josh: any sort of silicon chip and although it's kind of weird that they're taking
26:21
Josh: mouse brain cells and maybe soon human brain cells i did find it super interesting
26:26
Josh: just to see that there is experimentation happening with biological computers,
26:30
Josh: which is very different than silicon-based ones.
26:32
Ejaaz: Yeah, I mean, I'm at odds in my head because for years now, I believe that like
26:37
Ejaaz: silicon is the only way to process all of this stuff quick enough.
26:40
Ejaaz: But then you have startups like Neuralink, which are aiming to combine both
26:46
Ejaaz: silicon and brain tissue matter, right, into one kind of like synonymous function.
26:52
Ejaaz: And this is kind of like evidence or a sign that it might actually work.
26:54
Ejaaz: My bet is I think Silicon actually does a great job of executing the commands,
26:59
Ejaaz: but less of a great job of anticipating and being intuitive.
27:04
Ejaaz: This is like my guess, by the way. I don't have any proof about this.
27:07
Ejaaz: So it's really cool to see.
27:09
Ejaaz: And I wonder if something like this kind of plays into other stuff that Palmolaki kind of builds.
27:15
Ejaaz: I just find it weird that he's tweeting about this specifically when he typically
27:18
Ejaaz: talks about like kind of the weaponry stuff.
27:20
Ejaaz: Maybe there's a drone angle here where like a soldier with a helmet can basically
27:24
Ejaaz: think things and the drone does what he thinks um so there's some really cool
27:29
Ejaaz: sci-fi opportunity here um but i want to switch to this this next crazy story
27:34
Ejaaz: josh i don't know if you saw this but,
27:36
Ejaaz: Someone launched an app called rentahuman.ai.
27:42
Ejaaz: And the basic theme or function here is as a human, you can go on and volunteer
27:49
Ejaaz: your time for an AI agent to pay you to do a specific task. Now,
27:54
Ejaaz: I'll give you a very clear example.
27:56
Ejaaz: An AI agent came onto the site and paid someone $150 to go out on the street,
28:03
Ejaaz: hold up a sign that said, an AI made me do this.
28:07
Ejaaz: And then in brackets underneath it as a subtitle, and my pride is at stake,
28:12
Ejaaz: but I don't care. This is insane.
28:13
Ejaaz: And the first example of an AI agent kind of using human tissue and human muscle
28:19
Ejaaz: to do their bidding in the physical world.
28:21
Josh: Well, what's the through line through these two stories, right?
28:23
Josh: One is that, well, it turns out that biological brain matter can be trained
28:28
Josh: kind of similar to the same way we're doing for synthetic.
28:30
Josh: And this new story is, well, actually...
28:34
Josh: Since AI is superior, we might as well just offload the manual labor to humans
28:39
Josh: and we'll just pay them to do the tasks that we can't do quite yet because we don't have bodies.
28:44
Josh: And I mean, it's both are funny, forward looking, sci-fi-esque.
28:49
Josh: One is very real, one is not.
28:51
Josh: But I find this so funny that this is the second order effect that happened
28:56
Josh: only what a week after CloudBot went viral is already. They're offloading their
29:01
Josh: labor to humans and humans.
29:02
Josh: Of course they want to do it. And AI figured out how to make money and it's
29:05
Josh: ready to give it away. So why wouldn't you want to capitalize on that opportunity?
29:09
Ejaaz: I'm just scrolling the site right now. It's, I mean, a lot of it is like random
29:13
Ejaaz: esoteric Asian language posts.
29:16
Ejaaz: Which leads me to think that there's some like Chinese AIs involved in here,
29:20
Ejaaz: but like, hey, get $10 for joining Discord.
29:22
Ejaaz: So people are using it as kind of like a go-to-market kind of scheme.
29:25
Ejaaz: A lot of these tasks seem to be reproduced a lot. So I don't know how much of this is actually real.
29:29
Ejaaz: A lot of joining Discord add-ons, But a very interesting concept to see nonetheless.
29:35
Ejaaz: And then for our final story, some update from Google here. It looks something
29:39
Ejaaz: like a RuneScape battle, except it's AI models. What's going on here?
29:43
Josh: Oh, this is so cool. I'm so excited about this one because benchmarks stink, right? They're so lame.
29:48
Josh: Benchmarks can be gamed when it comes to models. Benchmarks,
29:51
Josh: I mean, it's like how good a model is at a specific task, but you can train
29:54
Josh: the model to be great at a specific task.
29:56
Josh: What you might have a more difficult time doing is training it to be good at a dynamic game.
30:01
Josh: So the video we're seeing on screen is an actual game
30:04
Josh: made by this company called Kaggle in this
30:07
Josh: this game arena Which allows you to sign your
30:10
Josh: model up for the game and then actually play against others in dynamic 3d worlds
30:14
Josh: So this is kind of similar to the trading simulation that we made a couple Videos
30:19
Josh: on in the past where you can't game the benchmarks It's a dynamic series of
30:24
Josh: events that a model must be good at and because you can't train it on the specific outcomes
30:30
Josh: it really proves which models are stronger than the others. And here we're seeing
30:33
Josh: an example of like a card playing game.
30:35
Josh: The other one is a role playing game. And I just thought this was really exciting
30:39
Josh: as a way to compare models because, I mean, benchmarks are lame.
30:43
Josh: You can't see them. But I would love to see these two fight against each other in an arena somewhere.
30:47
Josh: I think that just the gamified version of this is really exciting to see.
30:50
Ejaaz: I have always thought that taking exams and getting a percentage score to measure
30:56
Ejaaz: your intelligence as a human being is the worst possible way to do it.
31:00
Ejaaz: So I'm a massive fan of like having practical examples that everyday people can understand.
31:05
Ejaaz: And games is one of them. Trading is another.
31:08
Ejaaz: Video content creation. Like if someone can look at the thing and say,
31:12
Ejaaz: wow, that's impressive, then that's the test of AGI, not some of these esoteric random stats.
31:17
Ejaaz: So I'm very bullish on this. And I think that brings us to the end of the docket
31:21
Ejaaz: for this week's episode.
31:23
Ejaaz: It has been a crazy week, and you might be surprised to hear that we're actually
31:28
Ejaaz: waiting on more news today. It just hasn't come out yet.
31:32
Ejaaz: Anthropik's supposed to be dropping two new models, one which is going to be
31:36
Ejaaz: their latest coding model.
31:37
Ejaaz: OpenAI is meant to be dropping a new, I think it's GPT 5.3, which has been tested
31:43
Ejaaz: behind the scenes and apparently is amazing.
31:45
Ejaaz: So February apparently is the month for launches, and no one told me this.
31:48
Ejaaz: So we're getting sleepless nights. We released a bunch of episodes this week
31:52
Ejaaz: on Maltbook, which is the AI-only social platform.
31:55
Ejaaz: And as we mentioned earlier, the SpaceX and XAI merger, we have loads of juicy details in that.
32:00
Ejaaz: But yeah, thank you so much for listening to this episode. If you are listening
32:04
Ejaaz: to this on Spotify or watching it on Spotify or on YouTube, please subscribe. Please give us a rating.
32:10
Ejaaz: Leave us comments. We've been loving your comments and trying to respond to
32:13
Ejaaz: all. It's just been crazy.
32:16
Josh: Yeah, by the time you're listening to this, I'm reading a Sam post.
32:18
Josh: He says, big drop for Codex users later today.
32:20
Josh: So I assume by the time you're listening to this, two new models have already
32:23
Josh: dropped. So prepare for the episode impending and coming your way.
32:28
Josh: And yeah, as always, please share with your friends if you enjoy this.
32:30
Josh: I saw a few people who actually did that and their friends subscribed and that
32:33
Josh: was so cool. So thank you for joining us for another amazing week.
32:36
Ejaaz: We want to give you a thumbs up.
32:38
Josh: Please. Yeah, and yeah, we'll be back to cover all this new model releases next
32:41
Josh: week. We'll see you guys.