Latest / In the Loop AI / THIS WEEK IN AI: Claude Opus 4.7, Allbirds AI Rebrand, Leopold's Bet Pays Off

Transcript

THIS WEEK IN AI: Claude Opus 4.7, Allbirds AI Rebrand, Leopold's Bet Pays Off · In the Loop AI

0:00
Ejaaz: Anthropic just released their most capable model yet, Opus 4.7.
0:03
Ejaaz: It crushes GPT 5.4. It beats Gemini 3.1. It's amazing at coding and agentic
0:09
Ejaaz: reasoning. But there's one massive problem.
0:11
Ejaaz: It sucks compared to Claude Mythos, which Anthropic previewed and announced just last week.
0:18
Ejaaz: So it begs the question, why is Anthropic holding back their best model?
0:22
Ejaaz: There's a number of different reasons.
0:23
Ejaaz: It's a cybersecurity risk. But this just seems like an intermediary step.
0:26
Ejaaz: But that hasn't stopped Anthropic at all. They've managed to release four new
0:30
Ejaaz: features this week, and secret IPO rumors are valuing it at a $850 billion valuation.
0:36
Ejaaz: Now, there's many other stories that we want to cover in this episode.
0:39
Ejaaz: A shoe company pivoted to become a GPU provider, which sent their stock up 850%.
0:44
Ejaaz: And the world's greatest trader, he's only 24 years old, Leopold Ashenbrenner,
0:48
Ejaaz: is back, and he's made about $2.5 billion this week. Not bad.
0:52
Josh: Before we get into all of that, Let's start with the biggest news of the day, which is Opus 4.7.
0:57
Josh: I mean, this is the release that everyone's been waiting for for a long time,
0:59
Josh: and it is better, good, not great.
1:02
Josh: You'll notice the language they used. It is their best Opus model,
1:06
Josh: not their most capable model, because if you notice on these benchmarks,
1:09
Josh: there is a model on the far right that they're comparing it to, which is Mythos.
1:13
Josh: That is the private, secret, super-powered model that is not going to be available
1:18
Josh: for the public in the foreseeable future.
1:20
Josh: So what we have today is Opus 4.7. it's better than opus
1:23
Josh: 4.6 as expected it accelerates in a few key things
1:26
Josh: five to be specific the first one is coding which is
1:29
Josh: obviously better gains on the hardest long
1:32
Josh: horizon software tasks so basically if you have a task that
1:35
Josh: runs for a very long period of time it will be better at that
1:37
Josh: than opus 4.6 vision is an interesting one it can see three times better than
1:42
Josh: opus 4.6 so if you ever upload screenshots or documents or anything of that
1:46
Josh: nature it's going to be much better at understanding text or understanding details
1:52
Josh: inside of these images it's better at following instructions
1:56
Josh: Interprets instructions literally instead of figuratively, which we like.
2:00
Josh: So if you tell it what to do, it will actually do the thing specifically how you ask.
2:03
Josh: A lot of times people will ask, I don't know, something very specific and it
2:06
Josh: doesn't quite understand that. It will do that now.
2:09
Josh: Memory is much better. This is the one that I think is most interesting to me
2:12
Josh: is it's better at file system management and memory, which means
2:17
Josh: If you close out of a chat and come back, it will have some sort of persistent memory.
2:21
Josh: This is closer to what we realize with like agentic tasks or claws that have full memory systems.
2:26
Josh: And finally, it's really good at real world knowledge. It has state of the art,
2:30
Josh: well, state of the art, financial information that can access all the time.
2:34
Josh: So if you are asking it about specific current up-to-date information,
2:38
Josh: it should be much better at that.
2:39
Josh: And that's mostly the large changes that are coming from it.
2:42
Josh: I assume that a lot of the more subtle nuance changes will be things that you
2:47
Josh: just feel as you engage with the model.
2:48
Josh: I mean, Ejaz, we were both playing around with it before recording this episode
2:52
Josh: and noticed there was a few things that are slightly different about this one.
2:55
Ejaaz: Logically, not what is different, but there's a lack of difference.
2:59
Ejaaz: So I asked it basically, hey, you're 4.7.
3:02
Ejaaz: What can you tell me is new and cool with what you can do versus 4.6?
3:06
Ejaaz: And its response immediately was, hey, I'm 4.7. I'm not exactly sure.
3:11
Ejaaz: There is no live announcement, which details my comparison with 4.6.
3:15
Ejaaz: So I can't actually tell you how I'm better than my predecessor.
3:18
Ejaaz: So I was like, okay, okay, well, let's just assume you're good at long-term
3:23
Ejaaz: tasks, which is what you just explained, Josh, and is what Anthropic is claiming.
3:26
Ejaaz: I said, can you give me like a couple of prompts that I can run with you just
3:30
Ejaaz: to test that capability?
3:31
Ejaaz: And he said, sure. And he gave me this amazing prompt where it said,
3:34
Ejaaz: ask me to plan a two-week itinerary to Japan and set all these like complicated criteria.
3:40
Ejaaz: So I did, I copied, I pasted it and it gave me like a very, dare I say,
3:45
Ejaaz: ChatGPT 4.0-esque response with a bunch of emojis.
3:49
Ejaaz: It didn't sound anything meaningfully different from 4.6. So I think I need
3:56
Ejaaz: to use it a little bit more.
3:57
Ejaaz: I need to try out some different tasks. I do wanna check out the visual reasoning
4:01
Ejaaz: because I have had issues with Claude specifically, not ChatGPT,
4:05
Ejaaz: where I would like put in like images or screenshots of like a paper or something
4:09
Ejaaz: that I don't really understand.
4:10
Ejaaz: Sometimes struggles to read the actual thing so i'm glad to
4:13
Ejaaz: see that this is a stepping stone to the actual thing that
4:16
Ejaaz: people do want access to which is
4:19
Ejaaz: mythos uh they're all powerful model which you
4:21
Ejaaz: know um is sending the world into a spiral right now i feel
4:24
Ejaaz: like this is just like a version upgrade um and it we're kind of like in a holding
4:28
Ejaaz: pattern until we get to the real thing um if we look back at the benchmarks
4:32
Ejaaz: here uh this almost seems like chart crime josh like why are we putting like
4:36
Ejaaz: the best model right on the right hand side so that like maybe i don't like
4:40
Ejaaz: see how much better Mythos is compared to this.
4:42
Ejaaz: In its defense, I do notice that there are three things here that it is completely in its own realm.
4:48
Ejaaz: So if you look at scale tool use, Mythos doesn't really have that option or ability.
4:53
Ejaaz: If you look at financial analysts, so things like financial engineering,
4:56
Ejaaz: modeling, stuff like that, it is completely in a world of its own.
5:00
Ejaaz: It can beat Mythos at that.
5:02
Ejaaz: And then multilingual stuff as well. Now, that shouldn't come as a surprise
5:05
Ejaaz: because Mythos, as we've spoken about on previous episodes, is oriented around
5:09
Ejaaz: coding in particular, which has always been Anthropics mode.
5:13
Ejaaz: And it's nice to see that we now have a generalized model. I'm going to be using
5:16
Ejaaz: 4.7 all the time now. It doesn't make sense for me to use 4.6.
5:20
Ejaaz: What does concern me, the final thing, is I don't know if Anthropics should
5:24
Ejaaz: be spending their time serving us up a slightly better model when they've been struggling on compute.
5:29
Ejaaz: One thing I've noticed when I booted up Opus 4.7 is it's kind of automatically
5:34
Ejaaz: toggled towards something called adaptive thinking, which a few other people
5:39
Ejaaz: have spotted online as well.
5:40
Ejaaz: What that means is it'll like think hard when it thinks it should think hard.
5:46
Ejaaz: Now, typically, if you're paying 20 bucks, 100 bucks, 200 bucks for a particular plan.
5:51
Ejaaz: Kind of want the model to always be performing at its best right and like you
5:54
Ejaaz: can kind of figure it out but the fact that they're kind of like nerfing or
5:57
Ejaaz: throttling the model isn't really a good sign and i wish they would just throw
6:01
Ejaaz: that compute at mythos and give us a better model i.
6:03
Josh: Like the iterative nature i like that my model that i use every day is better it's nice
6:06
Josh: that the pricing is the same 4.6 is priced the same as 4.7
6:09
Josh: in terms of api use and in terms of just the general usage
6:13
Josh: for the plans it is interesting that their default is
6:16
Josh: that adaptive token generation for answering with 4.7 it
6:19
Josh: seems as if they're just slightly nudging like hey if you use a little
6:22
Josh: less tokens it's probably fine we'll let you know when you should use
6:25
Josh: more um but i think one of the things i'm most excited about this is
6:27
Josh: that financial modeling and planning i think that's where there is a new
6:30
Josh: spike that pops out of this model that hasn't really been done in previous models
6:34
Josh: where it has the financial analysis ability you can talk to it about investments
6:38
Josh: you can get strategies you can get up-to-date information that's gonna be a
6:41
Josh: fun thing to play around with i noticed the token generation was slower uh it
6:45
Josh: could just be because the model's new they're having a tough time serving up all the usage,
6:48
Josh: but it is an iterative improvement on the model that we use every single day.
6:51
Josh: So it's a nice upgrade. We'll take it.
6:53
Josh: It's not any more expensive. I'm looking forward to using Mythos or a Mythos-like
6:58
Josh: model so we really get that step function improvement.
7:00
Josh: But for now, 4.7 is a great incremental improvement.
7:04
Josh: I think we're probably going to continue to see a lot of these incremental improvements
7:06
Josh: that just kind of increase certain
7:08
Josh: things here and there. But play around with it. See what you think.
7:10
Josh: And yeah, that's mostly Opus 4.7, the update.
7:13
Ejaaz: The craziest part about that sentence, Josh, is this was the fourth product
7:17
Ejaaz: release that Anthropic released this week.
7:20
Ejaaz: They had three completely other different products and features which we're going to get into now.
7:24
Ejaaz: One of them, it's a new feature called Routines in Claude Code.
7:28
Ejaaz: Now, typically, when you're building an app or when you're coding something
7:31
Ejaaz: up, there are a bunch of other tasks that are separate from the main code development itself.
7:37
Ejaaz: And some of these tasks are kind of, you wish you could automate these things.
7:42
Ejaaz: They're kind of tasks that you need to run kind of like every now and then to check the damn thing.
7:45
Ejaaz: And it would be cool if you could just automate those things.
7:48
Ejaaz: There is something that exists in the world today called Cron Jobs.
7:51
Ejaaz: You can kind of like design specific tasks or routines that kind of like run
7:55
Ejaaz: the thing over and over again and you could achieve something similar like this
7:59
Ejaaz: in clawed code but with this new feature routines you can now
8:03
Ejaaz: create a routine or create a design of a task that you want to loop over and
8:06
Ejaaz: over again. But the best part is you can run it completely offline.
8:11
Ejaaz: So with your laptop or computer shut down, it runs completely on Anthropics
8:16
Ejaaz: version of the cloud. So you could be asleep at 2am, you don't need to wake
8:20
Ejaaz: up and these jobs and tasks now autonomously work for you.
8:23
Ejaaz: The benefit of this is you could go to bed with a code base that is being built
8:28
Ejaaz: by code code, your computer shuts down,
8:30
Ejaaz: you're still asleep, and it's already running security evaluation,
8:33
Ejaaz: security tests, and you wake up to a fresh new repo with a bunch of proposed
8:37
Ejaaz: fixes, and you just click approve or deny.
8:40
Ejaaz: And so it just makes you incredibly productive. You're like working like a full
8:43
Ejaaz: nine to five job whilst you're asleep.
8:45
Josh: Yeah, and this removes the need to buy something like a Mac mini or an always
8:48
Josh: on machine, which I really love.
8:49
Josh: I think there's been this craze to getting machines and purchasing all these
8:53
Josh: things that can stay online so that your claw features can go off and use them anytime.
8:58
Josh: This you can just set it and forget it let it go run
9:00
Josh: in the background it'll run on their servers on their time and then
9:03
Josh: you wake up and you have whatever you set the task to
9:06
Josh: do throughout the course of the night or throughout the course of the day and
9:08
Josh: it's that persistent computing that can always be doing tasks for you that i
9:11
Josh: think is really valuable and the market seems to like this along with all of
9:15
Josh: the other features that they've been releasing because the valuation is skyrocketing
9:19
Josh: i mean it's what up to 850 billion dollars evaluation if i'm not mistaken that's
9:24
Josh: puts them neck and neck with open ai
9:26
Ejaaz: Yeah, I think it's important to point out that OpenAI is at $900 billion.
9:32
Ejaaz: And the reason why that's significant is Anthropic has been kind of behind on
9:36
Ejaaz: the valuations for a while now.
9:38
Ejaaz: And a while within the realm of like the AI sense was three months ago.
9:43
Ejaaz: Three months ago, they were valued at around $300 billion.
9:45
Ejaaz: And now they've already jumped up to $850. Now, OpenAI's IPO is rumored to be
9:50
Ejaaz: around $1.1 to $1.2 trillion.
9:53
Ejaaz: Anthropic is not too far behind. I think a lot of this is spurred from their
9:57
Ejaaz: Mythos preview launch, which kind of like ruffled a bunch of feathers and pushed their valuation up.
10:02
Ejaaz: I was actually chatting to Josh before we started recording and I was out with
10:06
Ejaaz: dinner and a friend of a friend of a friend who runs a fund had just purchased
10:11
Ejaaz: $100 million at a $750 billion valuation from secondary shares from employees.
10:16
Ejaaz: So the market is incredibly hot.
10:19
Ejaaz: This news isn't just rumors or fake. We're seeing this through secondary share
10:23
Ejaaz: sales through employees and also through like open markets that people have
10:27
Ejaaz: created like tokenized versions of Anthropic shares.
10:29
Ejaaz: It's just an extremely hot market right now. We have three big IPOs coming up.
10:34
Ejaaz: We've got the SpaceX IPO, we've got the OpenAI IPO, and then we have the Anthropic
10:36
Ejaaz: IPO and they have caught up massively.
10:38
Josh: Yeah, there's a huge amount of money that is going public over the next 12 to 18 months.
10:43
Josh: And I don't know what type of impact it's going to have on the market,
10:45
Josh: but man, things are up only and quick.
10:48
Josh: Now for the remaining changes this week that came out of Claude. We have a couple.
10:52
Josh: One is the redesign of Claude code and just the desktop app in general.
10:56
Josh: I think everyone is fighting for this all-in-one desktop application.
10:59
Josh: They're trying to put all of the functionality under one roof.
11:02
Josh: Currently, OpenAI has like Codex and they have Sora, then they have the ChatGPT
11:06
Josh: app, and there's a whole bunch of different things.
11:08
Josh: Claude has pressed them all into one and they redesigned it this week.
11:11
Josh: So now there's a few different tabs.
11:13
Josh: The Claude code feature in particular works much better. That was a nice just
11:17
Josh: kind of user interface improvement.
11:18
Josh: I think the next natural state would just be to remove co-work and code and
11:22
Josh: just make it all one thing.
11:23
Josh: Um i shouldn't have to sift through different modes i just wanted to know what
11:27
Josh: i want to do the things i suspect that's what's coming next they also released
11:31
Josh: a new clod in word plan which is um not so great for microsoft i could imagine
11:38
Ejaaz: Yeah um i think the plan that anthropic is executing right now the strategy
11:43
Ejaaz: is get clod into as many people's hands as they can and it's not primarily going
11:48
Ejaaz: to be through an anthropical clod subscription so one way that they do this
11:51
Ejaaz: is they've partnered pretty strongly with Microsoft,
11:55
Ejaaz: which is weird because Microsoft usually has a very strong alignment with OpenAI and ChatGPT.
11:59
Ejaaz: They own about 27% of OpenAI.
12:02
Ejaaz: But now three of Microsoft's major AI products released in the last three weeks are powered by Claude.
12:09
Ejaaz: They have Microsoft Cowork, so they didn't even change the name.
12:12
Ejaaz: It's powered by Claude Cowork.
12:14
Ejaaz: And now they have this new integration where the entire Microsoft suite of programs,
12:17
Ejaaz: so PowerPoint, Excel, Word, now have Claude natively baked into it,
12:23
Ejaaz: which begs the question,
12:25
Ejaaz: How has Microsoft's AI product Copilot, who is designed to do the exact same
12:30
Ejaaz: thing, how would they be into the punch by Claude themselves?
12:33
Ejaaz: And it's a better product. If you look at this demo, if you actually test out
12:35
Ejaaz: the thing, if you speak to Copilot users, their response is,
12:39
Ejaaz: wow, Claude is a better product.
12:40
Ejaaz: The reason why is it maintains the context of your entire conversation,
12:44
Ejaaz: which Copilot hasn't been able to achieve off.
12:46
Ejaaz: So for a while now, Satya and Microsoft have been pushing away from ChatGPT.
12:50
Ejaaz: And maybe it's just because they have a better product. I don't really understand
12:54
Ejaaz: why they're taking a more tool agnostic approach.
12:56
Josh: Yeah, I was going to ask, isn't this the Microsoft that owns half of ChatGPT
12:59
Josh: and OpenAI? Yep. That has access to all the code base.
13:02
Josh: Yep. Yeah, and has, what, billions of dollars to build their own AI,
13:05
Josh: and yet, for some reason, are using Claude and Anthropix models. So, interesting.
13:10
Josh: If you use Word, this is a huge win for you. So, we just mentioned how Anthropix
13:14
Josh: has reached an $850 billion valuation.
13:16
Josh: This new company, or this old company, I should say, has had a similar trajectory,
13:21
Josh: although a little bit more unhinged.
13:23
Josh: If you are chronically online, you probably saw the news that Allbirds,
13:26
Josh: the shoe company, is pivoting to AI.
13:28
Josh: And the market loved it. The stock went up 850% in a single day.
13:35
Ejaaz: It was unbelievable.
13:36
Josh: And the timeline is really funny, right? It's like they IPO'd in 2021 at a $4
13:40
Josh: billion valuation. And this was the biggest company in Silicon Valley.
13:43
Josh: Everyone was wearing the world's most comfortable shoes.
13:46
Josh: I owned a pair. They were made of wool. they were really great turns out
13:49
Josh: people don't really care that much about wool shoes the stock lost
13:52
Josh: 99 and a half percent of its value over the next
13:55
Josh: four years they closed every retail store which is a
13:58
Josh: shame because they had a nice one over here in manhattan and i actually went
14:00
Josh: there they sold the entire brand for 39 million dollars and then rebranded to
14:05
Josh: new bird ai so all birds is becoming new birds they're using get this 50 million
14:11
Josh: dollars that's million with an m not a b or a T that is million to buy GPUs and compete with AWS.
14:17
Josh: And the market loved it. It's like a meme stock. It went from like $3 to $23.
14:21
Josh: It had the largest single day gain in company history and wiped out
14:26
Josh: I don't know, maybe like 5% of its 99.5% loss, but it's just a testament to
14:31
Josh: how frothy the market really is because a company can announce a pivot and just go absolutely nuclear.
14:36
Josh: It's just an interesting story on the frothiness of the market currently.
14:40
Ejaaz: There's a famous saying that goes, you may not know you're in a bubble, but there will be signs.
14:47
Ejaaz: This might be a pretty strong sign. I don't know what to say about this. I never liked the shoes.
14:53
Ejaaz: Maybe that's a hot take and I don't like this angle that they're taking right now.
14:57
Ejaaz: $50 million is the salary that Zuckerberg is paying one single AI researcher
15:04
Ejaaz: to work at Meta and on his entire lab.
15:07
Ejaaz: So it's not going to make much of a splash in the ocean, but I appreciate the
15:12
Ejaaz: sentiment and an effort.
15:13
Ejaaz: It seems like it's a bit of a stretch. I don't know how many GPUs they're going
15:17
Ejaaz: to be able to buy with that money, maybe 10.
15:19
Ejaaz: And I don't know if they even have the expertise to be able to spin those up.
15:23
Ejaaz: So I don't know who's managing this but it's a
15:26
Ejaaz: smart marketing move and good for whoever bought
15:29
Ejaaz: the stock like it's interesting when we were chatting about this
15:32
Ejaaz: pre-recording uh yesterday's episode josh and
15:34
Ejaaz: i were like wow it's up 200 that's crazy
15:38
Ejaaz: uh and then by the time we finished recording the episode what
15:41
Ejaaz: was it like 25 minutes later it was at 800 so maybe
15:44
Ejaaz: i'm the idiot here but just a wild wild story but
15:47
Ejaaz: there was someone that predicted this well well
15:51
Ejaaz: before its time um brian brian
15:54
Ejaaz: michael f uh goes if all birds just pivots to being an lm it could get its four
15:59
Ejaaz: billion dollar valuation back this was back in march 30th so he is like the
16:04
Ejaaz: listen he knew everything that was going on and yeah the stock is down around
16:08
Ejaaz: 30 today we're using perplexity computer to to show this here It's a nice little feed,
16:14
Ejaaz: which still gives it a huge marginal increase.
16:17
Ejaaz: I think the market cap is at like almost 100 million dollars.
16:20
Ejaaz: So that's still like a four and a half X. Pretty insane.
16:23
Josh: Yeah. This is like a joke. This is not even a serious. It's unbelievable that this happened.
16:28
Josh: And I think it's a testament to the maybe lack of seriousness right now.
16:32
Josh: It's like everything is going up. We don't care. And I do feel that frothy nature.
16:36
Josh: I find this whole story really interesting, but we do have a more interesting story.
16:40
Josh: This is from our good friend, Leopold Ashenbrenner, who last time we spoke about him, I think he was 24.
16:45
Josh: He's 25 now. Happy birthday, Leopold. since then he's
16:48
Josh: also turned uh another couple billion dollars in profit
16:51
Josh: that we haven't discussed so we need to have a brief update because
16:54
Josh: this dude is on a roll uh leopold for
16:56
Josh: those who aren't familiar famously created a situational awareness
16:59
Josh: fund which is designed exclusively to invest in kind of frontier ai technology
17:04
Josh: companies and my god it is doing so well to the point that not only is doing
17:09
Josh: like hostile takeovers of companies but now is leading funding rounds in them
17:13
Josh: as well and they are up so much money this this guy's crushing it Okay,
17:18
Ejaaz: So some quick context. At 23 years old, he starts a fund and raises a humble,
17:24
Ejaaz: very modest $250 million.
17:27
Ejaaz: In less than a year, he turns that into $1.5 billion.
17:32
Ejaaz: And at the end of last year, so we're talking December 31st,
17:36
Ejaaz: he's holding around $2.2 billion in the fund.
17:40
Ejaaz: So that's a huge markup. Any investor would be super happy.
17:43
Ejaaz: But he does something that no one expects.
17:45
Ejaaz: He takes major positions in a few core stocks.
17:49
Ejaaz: And I have it listed over here. He takes a massive, almost $900 million position
17:54
Ejaaz: in this company called Bloom Energy, which not many people outside of like the
17:58
Ejaaz: ASFA knew about back then, which is a fuel cell company.
18:01
Ejaaz: Basically, you can take portable fuel cells and power AI data centers.
18:06
Ejaaz: AI data centers are very hot right now, but they need energy.
18:08
Ejaaz: Bloom Energy solves that. And then he took a massive additional $300 million position on CoreWeave.
18:14
Ejaaz: Now, fast forward to today, Bloom Energy had a crazy week. Their stock is up
18:20
Ejaaz: 48% as I am speaking right now, which means that his original position is up $1.2 billion.
18:29
Ejaaz: That's a $1.2 billion return on his original investment in three and a half months.
18:36
Ejaaz: Total $2 billion position from $885 million.
18:39
Ejaaz: Absolutely insane. And the reason why is Bloom Energy is signing deals left, right, and center.
18:45
Ejaaz: But it's not only that. I looked at his SanDisk position, Josh.
18:49
Ejaaz: Why did I look at his Sandisk position? Sandisk is already up massively.
18:52
Ejaaz: He decided to take a $250 million position at the end of last year.
18:56
Ejaaz: That is now worth almost a billion dollars, but I wish that was the end.
19:00
Ejaaz: The final stock that he was invested in was this little stock over here called
19:04
Ejaaz: CoreWeave, which is up, I believe, around 25% as we're talking right now.
19:10
Ejaaz: Which means that he's up another couple hundreds of million dollars.
19:13
Ejaaz: The point is, Leopold is absolutely killing it. And the worst part is,
19:17
Ejaaz: if you wanted the blueprint as to what he's investing in, he wrote us a 64-page
19:20
Ejaaz: essay titled Situational Awareness, the name of his fund that was released about
19:25
Ejaaz: two years ago now, and it's held true until now. It's pretty amazing.
19:29
Josh: And we also published an update on the portfolio, what, a month or two ago?
19:32
Josh: And if you did follow along with that, chances are you're probably up a good
19:35
Josh: amount right now. I mean, Sandisk, as an example, is the largest single gainer
19:40
Josh: of the year for 2025 and 2026.
19:43
Josh: There has been no single better stock. So, Leopold's on it. Congratulations
19:47
Josh: on the success. If you have been copy trading him, congratulations on your success. And if not,
19:51
Josh: I might have to suggest that maybe you try it. Not financial advice as always,
19:56
Josh: but interesting nonetheless.
19:58
Josh: Now we have some more information this week coming from the person who is at
20:01
Josh: the top of the ladder, which is Jensen Wong, the guy who creates all these GPUs.
20:06
Josh: He was on Dwarkesh recently and had some interesting choice words to say about
20:11
Josh: China and many other things, but China in particular, this was the clip that
20:14
Josh: went viral about him getting a little defensive.
20:16
Josh: And I actually, the episode just came out. I haven't had a chance to watch it
20:19
Josh: yet. So what is he going on about here?
20:21
Ejaaz: Yeah, so it's an amazing episode. Let's start with this particular segment before
20:26
Ejaaz: we get into the other takeaways.
20:27
Ejaaz: So basically, I'm going to play the clip in a second.
20:30
Ejaaz: Duwakesh asked him a very specific question, which is, NVIDIA is selling a lot
20:34
Ejaaz: of GPUs to our adversary, China.
20:37
Ejaaz: In doing so, you're going to empower China to build super intelligent weapons,
20:42
Ejaaz: which may be used against the US.
20:44
Ejaaz: An example might be Claude Mythos, which can be used as a cybersecurity security attack tool.
20:49
Ejaaz: Jensen's response isn't what you might think. His response is not only we should
20:54
Ejaaz: align with China because they have a bunch of researchers and it's good to foster
20:58
Ejaaz: an AI community. He doesn't recognize them as an adversary in that sense.
21:02
Ejaaz: But two, he got incredibly defensive saying, I don't align with that loser energy.
21:08
Ejaaz: Let me play this episode.
21:09
Josh: You're not talking to somebody who woke up a loser.
21:11
Ejaaz: It's getting defensive.
21:13
Josh: That loser attitude, that loser premise makes no sense to me.
21:17
Josh: We are not, we're not a car. We are not a car.
21:22
Josh: The fact that I can buy a car, this car brand one day and use another car brand another day, easy.
21:30
Josh: Computing is not like that. There's a reason why the x86 still exists.
21:34
Josh: There's a reason why ARM is so sticky.
21:36
Josh: These ecosystems are hard to replace.
21:40
Josh: It costs an enormous amount of time and energy and most people don't want to
21:43
Josh: do it. And so it's our job to continue to nurture that ecosystem,
21:48
Josh: to keep advancing the technology so that we could compete in the marketplace.
21:53
Josh: Conceding a marketplace based on the premise you described, I simply can't acknowledge that.
21:58
Josh: It makes no sense because I don't think the United States is a loser.
22:04
Josh: Our industry is not a loser. And that losing proposition, that losing mindset makes no sense to me.
22:11
Ejaaz: You heard it from himself. we are not a
22:14
Ejaaz: car the point he goes on to make is you can't
22:16
Ejaaz: compare gpus to switching out a car to a
22:19
Ejaaz: different model like it's just leisurely right with gpus you need a very specific
22:23
Ejaaz: type of machine and nvidia has the machines to provide it he's basically talking
22:27
Ejaaz: his book and the point that he's kind of skirting is he wants to sell more gpus
22:31
Ejaaz: to china because he's going to make billions of dollars from it he makes a point
22:34
Ejaaz: later on in the episode that china has the most compute and energy in the world.
22:39
Ejaaz: So what is the substrate that uses that compute?
22:42
Ejaaz: GPUs, obviously, and he wants to be their biggest seller. He kind of explains
22:45
Ejaaz: that they have 50% of AI researchers and they have a good attitude towards building
22:50
Ejaaz: good AI models, but really, he just wants to make a buck.
22:53
Josh: Yeah, I liked the idea that we would sell them GPUs, but they were the like
22:57
Josh: previous version or they're the nerf down versions of them.
22:59
Josh: So there was some angle there. I could see the arguments for both.
23:02
Josh: It's an interesting conversation. I will be watching that conversation.
23:06
Josh: But also the United States seems to be panicking.
23:09
Josh: And perhaps not about this, but they are certainly panicking about the speed
23:12
Josh: in which AI is improving.
23:14
Josh: I'm not even sure they know that China's getting these chips yet.
23:16
Josh: But they apparently have been summoned to discuss Mythos, the anthropic model
23:22
Josh: that is so powerful that it is not being released to the public, that one.
23:26
Josh: It seems as if there's been some pretty serious closed-door discussions about
23:29
Josh: how to deal with this technology in the government just in the last week.
23:33
Ejaaz: Yeah, so some important context here. Anthropic went from being the kind of like...
23:38
Ejaaz: Weird kid to being the super popular kid when Claude Code kind of like rose
23:42
Ejaaz: to fame to then becoming the most hated person or hated company by the government
23:47
Ejaaz: because they refused to hand over their model and reins to be used in war.
23:52
Ejaaz: And so they got blacklisted.
23:54
Ejaaz: Now the government seems to be kind of backstepping here after they realize
23:58
Ejaaz: that the next model that Claude or rather Anthropic is about to release is super
24:02
Ejaaz: powerful and probably something that they want to have under their control.
24:05
Ejaaz: So two things have happened. One, Secretary Besant and Jerome Powell, Fed Chair,
24:12
Ejaaz: have gathered all banking heads, so from Goldman Sachs, JP Morgan,
24:17
Ejaaz: etc., to discuss the potential of this new model and to prepare defensive tactics
24:22
Ejaaz: against people that might use this model against them.
24:25
Ejaaz: So they need to secure data, accounts, so that money is installed and all that
24:29
Ejaaz: kind of stuff. So it's a real kind of like nation-level threat or global-level threat.
24:32
Ejaaz: Number two, they've re-engaged conversations with Anthropic themselves.
24:36
Ejaaz: So the company that they've blacklisted, they've now started conversations again
24:39
Ejaaz: with to try and figure out how they can work together.
24:42
Ejaaz: So my bet is, and I tweeted about this when the Pentagon stuff was going down,
24:46
Ejaaz: I was like, this is going to be a nothing burger in about three months from
24:49
Ejaaz: now when they realize that they do need to be aligned with Anthropic.
24:52
Ejaaz: Now, this could go two ways. One, politically correct.
24:55
Ejaaz: So Anthropic signs a deal and they just kind of shake hands.
24:57
Ejaaz: Or two, in a more extreme way, the government might even like take the steps
25:01
Ejaaz: towards nationalizing it if Anthropic refuses to help them.
25:04
Ejaaz: So I don't know which way this is going to go, but it proves to us that Mythos
25:08
Ejaaz: is very real and it is the best marketing campaign for Anthropic that they could have ever asked for.
25:13
Josh: Yeah, these things are getting powerful and they're only going to be getting
25:15
Josh: more powerful. And as that power dynamic shifts from the government to this
25:18
Josh: private industry, I think we're
25:19
Josh: going to see a lot of conflicting opinions and issues that will arise.
25:24
Josh: There is a random topic that I want to cover briefly just because I thought it was interesting.
25:27
Josh: This is like the fun part of the show we'll get into. if you remember there
25:30
Josh: is a mafia known as the paypal mafia where everyone
25:34
Josh: who was the founder of paypal has gone on to create all these amazing
25:36
Josh: companies a similar phenomenon is happening at spacex where
25:40
Josh: as spacex is going public a lot of the engineers who worked under
25:42
Josh: elon who did really hard and difficult engineering challenges they're rich they're
25:47
Josh: leaving they're doing passion projects and we got one of the first passion projects
25:50
Josh: this week which is called vital life and i just found it really interesting
25:54
Josh: i want to share briefly because it is a 25 pound cooler looking machine that
25:58
Josh: is able to process and clean water from anywhere in the world including oceans.
26:02
Josh: It does so using this really cool technology in a box that is
26:05
Josh: 50 times smaller than the average one by using
26:09
Josh: a lot of really interesting technology around reverse osmosis
26:12
Josh: and creating these different sensor and pressure feeds
26:15
Josh: that it allows it to clean and purify water anywhere in the world so
26:18
Josh: it's really amazing you have a hose you hook it into any water source
26:20
Josh: and it cleans it for 850 bucks i just thought
26:23
Josh: if you are like a camper if you are a fisherman if you are
26:26
Josh: ever in need of water they have kind of commoditized it and turned into a tiny
26:30
Josh: little cooler and it was just an interesting version one of what i'm hopeful
26:34
Josh: the spacex mafia is going to begin releasing which is just passion projects
26:38
Josh: man spacex is going at a two trillion dollar valuation they're collecting a
26:42
Josh: lot of that they get to make cool things with it this one i just thought was
26:44
Josh: pretty interesting awesome
26:46
Ejaaz: Um and yeah i have a fun story that i want to share to
26:48
Ejaaz: round this episode up uh this one is a
26:52
Ejaaz: little dystopian um so a group of researchers in korea has built a on and off
26:58
Ejaaz: switch for your genes which they can operate from outside of your body using
27:05
Ejaaz: the electromagnetic pulse that comes out of your electrical socket that you're
27:09
Ejaaz: charging your phone or yeah.
27:10
Josh: That's crazy now this
27:12
Ejaaz: Is pretty insane so the analogy is as follows if
27:15
Ejaaz: you imagine a garage door right you can use a remote to open it up and to close
27:21
Ejaaz: it down right you have similar channels in your cells where you can kind of
27:25
Ejaaz: send this calcium ion right calcium ions kind of open the gate or close the
27:30
Ejaaz: gate and the calcium ions can influence your genes.
27:32
Ejaaz: It could turn the genes on and it could turn the genes off. What they realized
27:36
Ejaaz: is if they created an engineered protein, I think it's called like ACBYB or something like that,
27:42
Ejaaz: and implant it near the gene that they want to switch on and off,
27:46
Ejaaz: they could just put you in an electromagnetic field,
27:49
Ejaaz: 60 megawatts to be specific, which is exactly the thing that comes out of your
27:52
Ejaaz: socket, right? So it's surrounding you right now.
27:54
Ejaaz: They could trigger a calcium channel to open up
27:57
Ejaaz: and switch on that gene, and then close the channel to close the gene.
28:01
Ejaaz: Now, you might think, oh my god, I can get remotely attacked,
28:05
Ejaaz: and I could like change my personality, or they could like affect my aging process.
28:10
Ejaaz: That's exactly what they did with the mice, at least in the study.
28:13
Ejaaz: So they were able to reverse aging or halt aging in these mice using three of the Nakamoto genes.
28:19
Ejaaz: Nakamoto genes are like one of the famous set of genes that can like trigger
28:22
Ejaaz: stem cell reversal. They were able to do that completely remotely.
28:26
Ejaaz: So I don't know what to think of this, except I am scared and concerned.
28:31
Josh: Well, it's good to know that like all of the what is it all the conspiracy theorists
28:35
Josh: that say that they're implanting chips in us plausibly true now.
28:38
Josh: Maybe they just haven't turned it on. They just got to activate the right electricity
28:41
Josh: field and change us up however you want.
28:43
Josh: It is fascinating, though. I think it's a testament to how weird the world of
28:46
Josh: biology is going to get as we kind of evolve and progress through this AI intelligence phase.
28:52
Josh: Is I think one of the more interesting industries is going to be biology because
28:55
Josh: we really understand so little about the human body and the nature of how different
28:59
Josh: things interact with us and unlocking those things and being able to actually
29:02
Josh: engage with our beings in a way that we've never been able to before is something
29:07
Josh: that's really exciting.
29:08
Josh: This is awesome research and development. We keep seeing more and more of it
29:11
Josh: coming at an increasing really rapid rate.
29:13
Josh: So I'm excited to continue to iterate, see where that comes.
29:17
Josh: Maybe just get, like, if wherever we want to change ourselves,
29:20
Josh: you just kind of push a button on your screen and it'll just adjust the magnetic
29:23
Josh: field around your body and switch over those genes.
29:25
Josh: Who knows? It's going to be weird and wild. But that is all of the updates for this week.
29:29
Josh: I think we've covered everything. If you've made it through all four episodes,
29:32
Josh: you are fully in the know of everything you need to.
29:34
Josh: On this week in the world of frontier technology, AI, all of it,
29:39
Josh: I hope you enjoyed the episode. Before we send you off for the weekend, we have a couple asks.
29:42
Josh: One is to subscribe to the newsletter. We post on Subsec twice a week.
29:46
Josh: Share it with your friends.
29:48
Josh: Like and comment on the video if you're watching it on YouTube.
29:51
Josh: Don't forget to subscribe if you are listening.
29:53
Josh: Leaving a five-star review helps a lot. Any final thoughts for the weekend?
29:57
Josh: What are we sending everyone off? What are you thinking about this weekend?
30:00
Ejaaz: The top of mind one is why didn't I copy trade Leopold when I reported on it two months ago?
30:07
Ejaaz: And then the second thing is I should have bought and owned all bird stock simply
30:13
Ejaaz: because the market is crazier than I could have ever have thought rational.
30:18
Ejaaz: So it's a lesson I've learned over and over again through my training experience.
30:21
Ejaaz: And I keep being forced to learn the same lesson whilst not making money, Josh.
30:26
Ejaaz: So I'm thinking why am I poor? I'm poor.
30:29
Josh: Okay. Well, yeah, I think I could resonate with that one too.
30:32
Josh: I also did not participate.
30:33
Josh: I hope someone listening did. Take some of our advice here and go follow these amazing traders.
30:38
Josh: So I hope I'm wishing everyone a profitable weekend.
30:42
Ejaaz: And we'll see you guys.
30:43
Josh: On the other side with an episode next week.
30:46
Ejaaz: See you guys.