Latest / In the Loop AI / This Week in AI: NVIDIA Earnings, Anthropic vs Pentagon, Perplexity Computer

Transcript

This Week in AI: NVIDIA Earnings, Anthropic vs Pentagon, Perplexity Computer · In the Loop AI

0:00
Josh: NVIDIA earnings are in, the stock is moving, and pretty much everyone has an
0:03
Josh: opinion on what this means.
0:05
Josh: But most of the commentary, I think, is missing the real story,
0:08
Josh: which isn't actually about this quarter at all.
0:10
Josh: It's about whether AI is in a super cycle that has likes for years or whether
0:14
Josh: we're about one overbuilds away from a dot com style wipeout.
0:17
Josh: Today, we're getting into the thesis that has Wall Street buzzing,
0:21
Josh: which is, I mean, kind of answering the question, are we in a bubble or not?
0:24
Josh: And if we're not, how much longer do we have? We have a lot of new data points
0:27
Josh: now with this NVIDIA earnings report that I think are worth getting excited about.
0:31
Josh: This is also the AI Weekly Roundup where we talk about the top news of the week.
0:34
Josh: So in addition to this, we have Perplexity. They launched a new personal computer from the ground up.
0:39
Josh: Cloud Cowork has become OpenClaw now. There's many OpenClaw alternatives that
0:44
Josh: come from everywhere. And Anthropic
0:45
Josh: wiped out, what was it, 15% of IBM's market cap in one single post.
0:50
Josh: So lots of things are happening quickly, but maybe we'll start with the NVIDIA earnings report.
0:54
Ejaaz: Yeah, so NVIDIA earnings pretty much broke the entire stock market. And I'll tell you why.
1:00
Ejaaz: For every single target, they beat it.
1:04
Ejaaz: They beat earnings per share. They brought in $68.1 billion in Q4,
1:10
Ejaaz: which is up about 70% from the year before and 20% from the previous quarter.
1:15
Ejaaz: And they're also estimating $78 billion for Q1 of this year,
1:20
Ejaaz: whereas the Wall Street estimates were about $72 billion.
1:23
Ejaaz: So things are looking pretty positive, right? And actually, in post-closed market,
1:28
Ejaaz: the stock pumped 4%. But this morning, it's down 4%. It completely retraced the move entirely.
1:36
Ejaaz: So it added $150 billion to their market cap, and then it completely removed it.
1:40
Ejaaz: And I think that there's just a clear misunderstanding of what NVIDIA is doing
1:45
Ejaaz: here and why it's actually bullish.
1:47
Ejaaz: I think a lot of people look at this and say, this seems like an AI CapEx bubble.
1:51
Ejaaz: I don't know how much more sustainable it is.
1:53
Ejaaz: And to be honest, I'm going to spend a bit of time arguing as to why it actually
1:56
Ejaaz: is pretty amazing, because we're talking about a 75% year upon year growth.
2:01
Ejaaz: And I think that what the market misunderstands is there is a lot of investment in GPUs.
2:07
Ejaaz: Buying GPUs is very expensive. And at some point, the music has to stop, right?
2:11
Ejaaz: Except that the companies buying the GPUs are, number one, fully utilizing all
2:17
Ejaaz: their GPUs, so they're maxed out and they need to actually buy more.
2:20
Ejaaz: But number two, there's a story about a little thing known as electricity or
2:24
Ejaaz: energy, which needs to power these GPUs.
2:27
Ejaaz: And we just don't have enough of it.
2:30
Josh: Yeah, any way you look at this report, the numbers are pretty outrageous.
2:33
Josh: I mean, revenue of this quarter coming in at $80 billion estimate versus 72.
2:38
Josh: I mean, the amount of money they're going to be earning and the profit margin
2:43
Josh: that they have on those earnings is just absolutely gigantic.
2:47
Josh: And I think one of the best takes that answered the question,
2:51
Josh: are we in a bubble or not? Where are we in the bubble?
2:53
Josh: What does this bubble look like? Is from this guy named Gavin Baker.
2:57
Josh: I think everyone here, we love Gavin Baker. He's great.
3:01
Josh: He's an analyst at Atreides Capital, I think.
3:03
Josh: And right before the earnings report, he published this
3:06
Josh: post not knowing what the results were going to be not knowing
3:09
Josh: kind of how impressive it was going to be and he
3:11
Josh: kind of broke it down into a few sections with
3:15
Josh: the core question is like nvidia's earnings durability is it actually going
3:20
Josh: to be able to hold through not only this report but through future reports and
3:24
Josh: he he outlines in this paper that every major tech revolution like railroads
3:29
Josh: radio the internet it follows the same playbook generally where the market correctly
3:33
Josh: identifies the potential of
3:35
Ejaaz: That creates a bubble.
3:36
Josh: And then the excitement creates that bubble and the demand temporary collapses.
3:40
Josh: And then there's a crash.
3:41
Josh: And he has this line that I love that says, believing this time is different
3:44
Josh: is dangerous, but I think this time might be different.
3:48
Josh: Maybe you have some insight as to why he really believes this time is different,
3:51
Josh: because it seems like this is backed up by like real information that separates
3:55
Josh: this bubble from those previous bubbles, like railroads, like the radio,
3:58
Josh: like the internet, even this time actually is different for a few key reasons.
4:03
Ejaaz: Yeah, so he makes the very valid point that in every previous bubble,
4:07
Ejaaz: we have overstepped what is actually real.
4:11
Ejaaz: So in this case, the people that are calling for the AI CapEx bubble,
4:14
Ejaaz: their argument would be that all the GPUs that are being bought aren't actually
4:18
Ejaaz: being utilized and they're just kind of sitting and collecting dust.
4:21
Ejaaz: The truth is all these GPUs, all these new ones and five to 10 year old GPUs are maxed out.
4:29
Ejaaz: In fact, if you wanted to go and rent a five-year-old GPU from NVIDIA or from
4:34
Ejaaz: a secondary market right now, you'd be paying roughly one and a half times the
4:39
Ejaaz: original cost of the GPU five years ago.
4:42
Ejaaz: The demand is off the scale and it's not stopping anytime soon.
4:46
Ejaaz: But the second more prescient point that he makes is the hyperscalers,
4:51
Ejaaz: even if they wanted to blow up a bubble that would lead to an ultimate collapse
4:55
Ejaaz: that all the doomers are talking about,
4:57
Ejaaz: they can't because they're limited by the amount of electricity and power that
5:02
Ejaaz: our grid can currently support.
5:05
Ejaaz: So he's saying that we can buy all these GPUs and they could overbuy it,
5:08
Ejaaz: but they will have not enough power to power them up in the first place.
5:12
Ejaaz: So in a weird sort of way, the hyperscalers are restricted from creating the
5:17
Ejaaz: bubble that everyone is fearing and, you know, worrying about.
5:21
Josh: Yeah, in the dot-com bubble, there's this huge race to deploy fiber as fast
5:24
Josh: as possible across the oceans, across the continents to get everyone connected online.
5:28
Josh: And the reality is, is that they were able to deploy the fiber so much faster
5:32
Josh: than, I mean, we could come up with use cases for it.
5:34
Josh: And in this case, the inverse is true where everyone wants to move quickly.
5:37
Josh: But he has this really elegant line that says, we're constrained by watts and
5:41
Josh: wafers, meaning we don't have enough electricity, we don't have enough chips.
5:44
Josh: And as a result, that becomes the limiting factor, enabling the growth.
5:49
Josh: And it will stop us from actually accelerating as fast as we could and smoothing
5:53
Josh: out what could eventually become a bubble, but still has a lot of narrative runway left.
5:58
Josh: The summary of this is that so long as nvidia is printing cash and doing well
6:04
Josh: there is basically infinite demand because they are just not able to create
6:09
Josh: as much as the industry desires and as a result of watts and wafers being the limiting factor
6:16
Josh: we are just not able to reach bubble levels required to make things,
6:20
Josh: to put things into the danger zone.
6:22
Josh: So kind of his sentiment, and it seems like the sentiment after this NVIDIA
6:25
Josh: report is that there's a lot of time left before you have to worry about things collapsing.
6:29
Josh: And there will be ruts along the way, but we are like still green lights full speed ahead.
6:35
Ejaaz: A lot of people are worried about AI basically replacing human labor entirely.
6:40
Ejaaz: And he makes the point that because we're restrained by energy and the electrical
6:44
Ejaaz: grid, and we will be for the foreseeable future for the next at least five years
6:48
Ejaaz: until we build this thing, this stuff takes a lot of time.
6:50
Ejaaz: We'll reach a point where we buy enough GPUs, but the cost of paying an AI to
6:55
Ejaaz: do a human job actually becomes more expensive than just employing a human.
6:59
Ejaaz: As long as we have this limiting factor of electricity and power,
7:03
Ejaaz: humans are good for the foreseeable future.
7:05
Ejaaz: And of course, take that with a pinch of salt. It's interesting,
7:07
Ejaaz: this article was a direct rebuttal to a really.
7:11
Ejaaz: Tantalizingly written post from citrini which is a
7:14
Ejaaz: research firm which basically outlined the doomer case where
7:17
Ejaaz: ai takes over everything and you know humans basically need to kind of like
7:21
Ejaaz: fight for themselves and like subsidize themselves through i don't know some
7:24
Ejaaz: kind of just voluntary payment uh structure and so he's outlining the positive
7:28
Ejaaz: side of this here josh i don't know if you saw but there was also a piece that
7:31
Ejaaz: came out from citadel securities which by the way,
7:35
Ejaaz: is the most funniest thing ever because Citadel is like known for making most
7:39
Ejaaz: of their money from shorting stuff.
7:41
Ejaaz: So the fact that they're now like super bullish, this new technology is just
7:44
Ejaaz: kind of hilarious. They must have quite a few long positions open.
7:47
Ejaaz: But yeah, I think overall, the point that's being made here is like NVIDIA is beating its earnings.
7:51
Ejaaz: It's consistently done this for about eight quarters now. Let me repeat that.
7:56
Ejaaz: Around eight quarters. They're up 75% in gross revenue from last year.
8:03
Ejaaz: They've been repeating this for about two years now.
8:07
Ejaaz: This train isn't stopping. And I don't know how much longer the doomers can
8:10
Ejaaz: keep saying, hey, I think this is the top.
8:13
Ejaaz: This is the top. This is the end. I just don't think we're there yet.
8:15
Josh: What we see is the CapEx spend that we talked about a few weeks ago.
8:20
Josh: Like clearly this is all going to places like NVIDIA. These companies have pledged
8:23
Josh: $650 billion of CapEx spend, which is almost exclusively going to AI buildouts.
8:30
Josh: NVIDIA is going to see a ton of that. These earners are not going to slow down.
8:32
Josh: The build-outs are not going to stop.
8:34
Josh: I mean, it's just a very clear path to moving forward and to continuing this
8:37
Josh: growth. And it's really exciting to be a part of this ride.
8:40
Josh: Another ride that I went on yesterday
8:42
Josh: was with Perplexity saying that they rebuilt the personal computer.
8:46
Ejaaz: Josh, how much did that trigger you, that single sentence?
8:49
Josh: I'm just a little confused because, you know, Perplexity, they've been quiet.
8:54
Josh: We had Arvin, the CEO, as a guest on the show. Feel free to go back and find that interview.
8:57
Josh: It was really interesting because
8:59
Josh: they pivoted the entire company around building an agentic browser.
9:02
Josh: And now it appears like they have expanded from the browser to the entire computer.
9:07
Josh: So, Ijaz, I haven't looked into this news entirely. I'm not exactly sure what it is they built.
9:11
Josh: But based on the announcement video, it seems like computer use is actually
9:15
Josh: getting pretty good on perplexity. Is that true?
9:17
Ejaaz: Let's start with the good news. The vision with perplexity computer is number
9:20
Ejaaz: one, they didn't rebuild an actual physical computer.
9:25
Ejaaz: What they rebuilt was the operating system for a computer.
9:29
Ejaaz: So right now, if you have Windows, you booted up Windows. If you have Mac OS, you have Mac OS.
9:34
Ejaaz: What they've tried to do here is build a layer on top of that that unites all
9:38
Ejaaz: the things that you might potentially do on a computer, whether that's coding.
9:43
Ejaaz: Whether that's doing your nine-to-five job, whether it's opening up a browser,
9:46
Ejaaz: whether it's like playing around with a few apps.
9:48
Ejaaz: They've unified it on a single layer called perplexity computer.
9:52
Ejaaz: Why did they do this? Well, their argument is very simple. They're saying,
9:56
Ejaaz: hey, you're using a different model for your browser.
9:59
Ejaaz: You're using a different model to
10:00
Ejaaz: code, using a different model to spin up agents to do your work for you.
10:04
Ejaaz: Let me just bring all of this under one roof and make a single unified experience.
10:09
Ejaaz: So I actually love the vision. And I think that something like this is actually
10:13
Ejaaz: desperately needed in a world where we have open claw.
10:16
Ejaaz: And it's funny, we're just talking about like 10 different open claw variants.
10:19
Ejaaz: We have clawed co-work. I have a million subscriptions.
10:22
Ejaaz: My mind is going crazy. I just want one singular thing that I can like operate on.
10:26
Ejaaz: And this is what that thing supposedly does. It can research,
10:31
Ejaaz: design, code, deploy, and manage any project end to end.
10:34
Ejaaz: So that sounds good. But the question is, can it actually do it?
10:38
Ejaaz: So I want to go through like a few demos.
10:40
Ejaaz: One, which is showing, I think this might be the actual video where it takes
10:44
Ejaaz: a long form video and creates a short form type of content. Do I have that right, Josh?
10:48
Josh: Oh yeah, this is, it's about halfway into this video. It's, it's pretty well
10:51
Josh: done because I've been trying to get this to work with a service like Claude
10:54
Josh: And it doesn't have the ability to edit videos within it.
10:59
Josh: So I think that is a plus one to perplexity where the example that it uses,
11:02
Josh: it's like go to an interview that Dario Mode, the CEO of Anthropocad,
11:06
Josh: and then pull out a specific part and turn it into a clip.
11:09
Josh: And then it actually went and it rendered the clip and it added subtitles and
11:13
Josh: descriptions and it did everything that a producer on a show like ours would do.
11:17
Josh: So that was really interesting. And I think it's exciting to see these companies
11:21
Josh: leaning more into this new AI first operating system because as these ai models
11:26
Josh: are getting better at computer use and anthropic just acquired another company
11:29
Josh: that is exceptional at computer use they're going to continue to kind of rebuild
11:34
Josh: what the computer does but through the ai first lens
11:37
Josh: And this seems pretty interesting. Like as someone who hasn't used Perplexity
11:40
Josh: in months, I'm reconsidering maybe I should sign up for a subscription and try this out.
11:45
Ejaaz: Yeah, the way I would describe this is it is open core, but for normies,
11:50
Ejaaz: for the everyday person, you don't have to worry about massive security issues.
11:54
Ejaaz: It's a really simple and easy looking interface in terms of like connecting your apps.
11:59
Ejaaz: I think a portion of this data is held privately.
12:02
Ejaaz: The agents are very intuitive to use. You just kind of like prompt them.
12:05
Ejaaz: If you look at one of these demos right here, he says, I'm thinking about buying
12:09
Ejaaz: Walter White's house and turning it into a rental. Please build me a financial model.
12:13
Ejaaz: And it just kind of goes off and does its own thing. You don't need to tell it what tool to use.
12:17
Ejaaz: If it requires access to Slack or something, it'll say, hey,
12:20
Ejaaz: can I get access to Slack? You just say, yes, approved.
12:22
Ejaaz: And it just kind of like gets on with it. So I think that's pretty cool.
12:25
Ejaaz: There are also other few demos that I quite enjoyed.
12:29
Ejaaz: So if you're not in the finance world, or maybe you are, but you would have
12:32
Ejaaz: heard of something called Bloomberg Terminal. It is like the mecca.
12:35
Ejaaz: It costs $30,000 a year to access, and you get all the latest,
12:40
Ejaaz: greatest financial news before the public gets it.
12:42
Ejaaz: If important financial news breaks, it breaks on Bloomberg Terminal first. It is the mecca.
12:47
Ejaaz: Someone rebuilt it from scratch using perplexity computer, and it took them
12:52
Ejaaz: less than an hour to do so.
12:54
Ejaaz: So you've already automated away $30,000 a
12:58
Ejaaz: year worth of value for the cost of a single prompt and if i look at this interface
13:02
Ejaaz: right here from this demo it actually looks super legit now i don't know what
13:07
Ejaaz: ip infringements happen here and like in defense of the non-perplexity computer supporters.
13:15
Ejaaz: We actually kind of did something similar, Josh. I think we,
13:18
Ejaaz: what did we use to build this?
13:19
Ejaaz: We used, oh yeah, we did a test between Cloud Code, Opus 4.6 and Codex 5.3.
13:25
Ejaaz: And we said, hey, can you build up an investment tracker analysis app for us?
13:30
Ejaaz: And I did it in a few minutes, but it wasn't to the extent that this thing is.
13:33
Ejaaz: So I think it's just that added layer of an operating system helps you kind
13:37
Ejaaz: of piece together a bunch of different tools and builds a better app or a better
13:42
Ejaaz: tool or a better web experience for versus like prior.
13:45
Josh: Models it'll get you close like what what
13:48
Josh: is this missing well it's missing the real-time news it's missing the
13:51
Josh: private chat room so it loses a lot of the value that comes from that thirty
13:55
Josh: thousand dollars or however much it costs but it created an interface that without
13:59
Josh: those added values it actually looks pretty much like a bloomberg terminal and
14:03
Josh: it all the data works all the news tracks all the charts are accurate and it
14:08
Josh: creates this fun dashboard with a single prompt.
14:10
Josh: So perplexity seems like they're onto something.
14:12
Josh: I hope that they are aware that the rest of the industry is also working very
14:17
Josh: hard to doing this. In fact, they might even be there already.
14:20
Josh: And when I'm saying that, I'm referencing Claude Cowork because essentially
14:23
Josh: this week they have turned Claude Cowork into OpenClaw one step at a time.
14:27
Josh: And I think this started with news that was discussing scheduled tasks.
14:32
Josh: So a big thing about OpenClaw in particular is that it has a heartbeat and every
14:37
Josh: 30 minutes it kind of wakes up and it goes through its tasks and what Claude
14:41
Josh: does now is you can schedule tasks to happen throughout the course of the day.
14:45
Josh: You could ask it for a morning brief, you could ask for weekly updates,
14:48
Josh: for Friday presentations, whatever it may be, it adds these things called
14:52
Josh: cron jobs into the Claude co-work operating system and i think it's it's another step of
15:00
Josh: Claude and Anthropic and a lot of these companies kind of working towards this
15:04
Josh: first operating system in a way that is is very compelling like now i have it
15:08
Josh: do a little morning roundup where in the morning it'll go through all the news
15:12
Josh: that i missed over the night and just give me a couple bullet points to see
15:15
Josh: if there's anything i find interesting it's pretty cool yeah i
15:17
Ejaaz: Mean that's one very essential part of what makes open claw an amazing product
15:21
Ejaaz: and the fact that now you have like this kind of sandbox version where you don't
15:24
Ejaaz: need to kind of check for any security risks or anything like that is awesome.
15:28
Ejaaz: That's not the only tool that they launched this week. Like Anthropik's been
15:31
Ejaaz: on a generational run this week alone.
15:34
Josh: It seems like every day they're announcing a new feature.
15:36
Ejaaz: I need them to give me a break because it takes a while to process these things,
15:41
Ejaaz: but they're just shipping every single day at this point. I'm tired, dude.
15:44
Ejaaz: So this new feature is called Remote Control, which is exactly what it sounds
15:49
Ejaaz: like. you can now code using Claude from your phone whilst you're going on a
15:54
Ejaaz: walk, whilst you're walking on the dog.
15:55
Ejaaz: So if you need to get away from a screen,
15:59
Ejaaz: Now, I'm sorry, you're chained, you're ball and chained even when you leave
16:02
Ejaaz: your house at this point.
16:03
Ejaaz: It's really cool because one of the great parts about OpenClaw was the fact
16:07
Ejaaz: that you could lay in bed or communicate with your coding agent or your claw
16:11
Ejaaz: bot from anywhere in the world. And you didn't need to be at your desktop terminal at that specific time.
16:17
Ejaaz: Clawed code and Anthropic prior to this announcement required you to do so.
16:21
Ejaaz: Now you don't need to be able to do that. So now when you combine this with
16:24
Ejaaz: the autonomous cron jobs that you just mentioned, Josh, you start to see what
16:28
Ejaaz: the vision is going to be.
16:29
Ejaaz: And it looks something very similar to OpenClaw.
16:32
Ejaaz: But to go back to the original form of Anthropic announcements,
16:36
Ejaaz: they typically announce, I don't know, what is it?
16:38
Ejaaz: One to three new plugins for Clawed in general, Cowork specifically, maybe every week.
16:44
Ejaaz: And it results in a, I saw a stat here, it results in a 10 to 35% stock decline
16:50
Ejaaz: of any company that currently has the winning leading software,
16:55
Ejaaz: because now Clawed has automated that entirely. and we have a new one this week. That's crazy.
16:59
Josh: We need to pause there for a second to just to acknowledge how powerful Anthropic
17:03
Josh: has become, where a single tweet and a single feature unlock has the ability
17:09
Josh: to move markets billions, maybe tens of billions of dollars.
17:12
Josh: And what I find funny and ironic before we get into what happened this week
17:15
Josh: with the news is that it seems as if this is a power only Anthropic holds.
17:20
Josh: If ChatGPT announced something similar to this, I'm not sure it has the impact of anthropic and quad.
17:27
Josh: And I think that's worth acknowledging is because they haven't.
17:30
Josh: They haven't. Yeah, we haven't seen this. And I think it's worth acknowledging
17:33
Josh: because Anthropic very clearly is seen as not only a powerhouse,
17:37
Josh: but a real threat to a lot of institutions, a lot of companies that have been
17:41
Josh: kind of grounded in their ways.
17:42
Josh: And this example this week, I love, it's so funny.
17:45
Josh: So please share with us what the hell happened with this new post that knocked
17:50
Josh: a certain company off the grid.
17:53
Ejaaz: Okay. So there are many different types of programming languages and one very
17:58
Ejaaz: important and I guess essential programming language for the business world
18:02
Ejaaz: for data processing specifically is known as COBOL, C-O-B-O-L.
18:07
Ejaaz: It stands for Common Business Oriented Language.
18:10
Ejaaz: And one particular company called IBM has held the monopoly on this.
18:14
Ejaaz: And the reason why they've held the monopoly on this is that it's quite a complex
18:18
Ejaaz: programming language to understand.
18:19
Ejaaz: You need to kind of like learn how the different tools work.
18:22
Ejaaz: Then you need to adopt how the language actually fits to your business. it's a whole process.
18:27
Ejaaz: IBM walks you through hand in hand and figures it out for you,
18:30
Ejaaz: and they charge you a premium price to be able to do that.
18:34
Ejaaz: Until two days ago, when Anthropic announced that they had launched a single
18:38
Ejaaz: plugin that was coded from scratch using Claude Code, and it does all of that
18:43
Ejaaz: for you, for the cost of a single prompt.
18:46
Ejaaz: The news wasn't too great. IBM stock, it says here it dumped 10%.
18:50
Ejaaz: It actually dumped 15% within an hour of that tweet going live.
18:56
Ejaaz: Just insane to see. And this isn't the first time Anthropic has done this.
18:59
Ejaaz: They announced a legal plugin about a week and a half ago, and all law stocks
19:03
Ejaaz: dumped about 30 billion dollars just insane.
19:05
Josh: That's a huge amount of money that's blown and i
19:08
Josh: think it's it's a testament to kind of where things are going right it's
19:11
Josh: ibm makes so much money because cobalt is such a old outdated
19:15
Josh: protocol that no one really supports no
19:18
Josh: one really understands how it works they're maintaining dinosaur code and now
19:21
Josh: anthropic comes in and says we're not like humans we don't get tired we don't
19:25
Josh: get bored we will actually just go and work with your cobalt code and fix things
19:28
Josh: and that knocked 15 that's a huge amount of money and that's still not the extent
19:33
Josh: of all the philanthropic news this week because there is another security implementation.
19:38
Josh: So if I remember correctly, when they announced Opus, maybe it's 4.6,
19:43
Josh: they announced it with the fact that
19:45
Josh: it found 500 zero-day vulnerabilities in code.
19:49
Josh: Correct. Meaning that there is existing code bases out there that have exploits.
19:53
Josh: It uncovered 500 of them just via the AI model itself.
19:57
Josh: And now what they're doing, it seems, is they're kind of reversing that and
20:00
Josh: allowing you to have the tools to play defense against Claw's own hacking abilities. Is that right?
20:05
Ejaaz: Yeah, I actually think that it might have been Codex 5.3 from OpenAI,
20:09
Ejaaz: but the fact is like both models can do the same thing.
20:12
Ejaaz: And these coding models are getting so good finding the vulnerabilities that
20:16
Ejaaz: humans are very prone to implementing in their own code.
20:20
Ejaaz: And these vulnerabilities, by the way, can be exploited for billions of dollars.
20:23
Ejaaz: I'm not like underselling this. Like if you are a multi-billion dollar company,
20:27
Ejaaz: a hundred billion dollars plus, right?
20:29
Ejaaz: And you have security floor that gives people access to personal data or back
20:33
Ejaaz: account information, that could be a really large financial and legal risk for
20:37
Ejaaz: you. So it's very important that you patch these things up.
20:39
Ejaaz: But the issue is humans, well, number one, need to be an expert in that specific
20:45
Ejaaz: code base in order to find those specific flaws.
20:49
Ejaaz: So there's only usually about three to five people in any of these big companies that can do that.
20:53
Ejaaz: So there's a heavy risk point there, where if those people leave the company,
20:58
Ejaaz: or they decide to like go focus on something else, you now have an open risk.
21:03
Ejaaz: Claude announced this new tool called Claude Code Security, which now does this for you.
21:09
Ejaaz: It scans your entire code repository in about 10 to 30 minutes.
21:14
Ejaaz: It understands the nuance and the way that you wrote this code,
21:20
Ejaaz: the context with your business, et cetera, et cetera.
21:22
Ejaaz: And then it finds all the security flaws for you.
21:25
Ejaaz: Typically, this would have taken a team of five to 10 people months,
21:29
Ejaaz: or it's like a recurring thing.
21:31
Ejaaz: And now you can just run a single prompt like every 30 minutes or every week,
21:35
Ejaaz: and it'll discover all the flaws for you, but it gets even better.
21:38
Ejaaz: It proposes the solution that you need to fix in order to resolve that security issue.
21:46
Ejaaz: It doesn't go to the extent where it writes up the fix for you,
21:49
Ejaaz: although you can definitely say like, hey, okay, that sounds good,
21:52
Ejaaz: build the fix for me, because they wanted to maintain the human in the loop.
21:55
Ejaaz: But if you can imagine, if the AI is already good enough to find the flaws and
21:59
Ejaaz: create the fix for the flaws, you might just let this thing run autonomously for you.
22:03
Ejaaz: Now, let me ask you this, how much is that product worth for your business?
22:07
Ejaaz: It's probably like tens of billions of dollars, if not more.
22:09
Josh: They're just continuing to snipe and cherry pick the value from every existing
22:13
Josh: company until it's all in one. It's amazing the amount that they're absorbing in the market.
22:18
Josh: We have a few things left to cover this week that I kind of want to go through quickly.
22:22
Josh: The first one being the XAI Pentagon deal.
22:25
Josh: XAI and Grok is now powering Pentagon's AI technology. And we had a crazy sentence.
22:30
Josh: This is like a strange narrative timeline that we're running.
22:32
Josh: And we had an episode earlier this week that I would recommend checking out
22:35
Josh: about Anthropic and their kind of difficulties they've been having in dealing
22:39
Josh: with the government because they seemingly don't want to give access to all
22:43
Josh: of the ways in which the Pentagon wants to use it.
22:47
Josh: But it appears as if Grok and XAI did not have this problem.
22:51
Josh: And the plan is that they're going to fully integrate Grok starting later this
22:55
Josh: month into military operations. What those are, I don't know.
22:58
Josh: I'm not sure we'll ever find out. But I just find it interesting...
23:01
Josh: That there is a new kind of role in government here for AI Labs and Grok is
23:06
Josh: now going to be one of the front runners here.
23:08
Josh: This comes on the back of, I mean, more drama with Anthropic, right?
23:12
Josh: It was after we recorded the episode, there was still even more.
23:15
Josh: I mean, apparently Anthropic has an ultimatum where by Friday they have to either
23:21
Josh: decide if they're going to comply with the Pentagon's request or not.
23:25
Josh: And if they don't, what's at stake here for Anthropic?
23:29
Ejaaz: So there's this thing I'm highlighting here called the Defense Production Act.
23:33
Ejaaz: The Pentagon basically has two options if Anthropik refuses.
23:36
Ejaaz: On Friday, they can force their hand, initiating the Defense Production Act to basically say,
23:41
Ejaaz: sorry, you legally now need to give me an uncensored version,
23:45
Ejaaz: even though you declined, because it is now the official law and it's a case
23:48
Ejaaz: of national state security, which, by the way, is an insane thing.
23:52
Ejaaz: Or they can take option two, which is blacklist Anthropik and all of their models
23:58
Ejaaz: as a security risk to the nation,
24:01
Ejaaz: which means that Anthropic probably ceases to receive any kind of government
24:06
Ejaaz: funding or regulatory support in the future.
24:09
Ejaaz: And it also pushes up all the other models that have signed partnerships with
24:13
Ejaaz: the Pentagon, such as Grok.
24:15
Ejaaz: And I hear OpenAI and Google is doing the same as well.
24:18
Ejaaz: The sentiment has been split on this, Josh. I've seen both sides.
24:21
Ejaaz: One saying Anthropic just needs to wisen up. And I think that was a lot of the
24:25
Ejaaz: argument that we gave on our episode earlier this week. Again,
24:27
Ejaaz: please go check it out. Let us know if you think we're wrong.
24:29
Ejaaz: But then the other side, like I saw Jeff Dean tweet this morning.
24:32
Ejaaz: Jeff Dean is like the guy from Google. He built so many of Google's key products.
24:38
Ejaaz: And he said, I like that Anthropics taking a stand and it's important that they don't violate,
24:43
Ejaaz: I think is what the first amendment or whatever it is where you can't surveil
24:46
Ejaaz: all Americans. And I respect the fourth, the fourth.
24:49
Josh: Yeah. It's seems heavy handed, right? Like I can understand not working with
24:54
Josh: them, but putting them on a blacklist seems a bit extortive,
24:58
Josh: especially when there's other options. Yeah.
25:00
Ejaaz: Well, here's the thing, like the government's favorite model.
25:03
Ejaaz: And Pete said this himself,
25:05
Ejaaz: the head of the Pentagon, he said that Claude is the best model and they've
25:08
Ejaaz: been using it for over a year now so you can imagine if claude has been trained
25:12
Ejaaz: with all your employees it's been entrenched on your data you kind of don't
25:16
Ejaaz: want to rip that out and then like train a new model from scratch so,
25:20
Ejaaz: Antarmic has everything to win and lose here.
25:23
Josh: Yeah, it's interesting because I'm sure we're missing a lot of context too,
25:26
Josh: due to just the general private nature of how these deals work.
25:28
Josh: So it's going to be tough to really give an honest opinion because I'm sure
25:33
Josh: we're missing a lot of data, but interesting nonetheless.
25:35
Josh: And congrats to XAI and the team for getting a new contract.
25:39
Josh: Last winner of the week, AMD. Oh my God, 15% jump in one day on a Meta deal.
25:46
Josh: Meta is now working with AMD. I thought Meta was working with NVIDIA. That's what I thought.
25:52
Ejaaz: But no, they signed a new $6 billion compute partnership.
25:56
Ejaaz: And remember, it's not a deal now. It's just, we're specifically calling it partnerships.
26:01
Ejaaz: These are crazy. Yeah, it's insane. The circular economy is back.
26:05
Ejaaz: The nature of the deal is basically Meta doesn't want to rely solely on NVIDIA
26:10
Ejaaz: chips, although NVIDIA is the big boy here.
26:13
Ejaaz: So they want to diversify their compute arsenal so that if anything were to
26:17
Ejaaz: happen to NVIDIA or any of the other chip makers, they can still stay alive
26:20
Ejaaz: and feed their product to you.
26:23
Ejaaz: Now, Meta has been out of the kind of news for a while because their models
26:28
Ejaaz: weren't able to keep up and they spent about, what was it?
26:31
Ejaaz: Like a hundred to maybe $200 billion to hire like something ridiculous,
26:36
Ejaaz: like 50 people in their new AI team and they have had nothing to show for it.
26:40
Ejaaz: Rumor says they're meant to be releasing a new model in a few weeks time.
26:43
Ejaaz: Avocado is the code name. and so if that is all it's hyped up to be then they're
26:48
Ejaaz: going to need all the data center compute that they can and they seem to be
26:52
Ejaaz: sure that amd will be the ones to be one of the main providers for them so they've
26:56
Ejaaz: locked in that deal they're building out data centers and yeah amd stock jumped
27:00
Ejaaz: 15 absolutely insane well.
27:02
Josh: Happy for all parties involved i think one thing throughout the episode there's
27:06
Josh: a central theme in which there is no amount that can satiate the demand And
27:11
Josh: for AI compute and energy and electricity.
27:14
Josh: And like, I'm sure Meta would prefer to work with NVIDIA.
27:18
Josh: This is the option that they have because they need GPUs they
27:21
Josh: need compute they need energy and they have no choice so
27:24
Josh: we are going to continue this run this
27:27
Josh: generational run that we've had really in the world of AI
27:30
Josh: it doesn't seem like the top is in it doesn't seem like the top is anywhere near
27:34
Josh: and that leaves me optimistic and wanting to continue to deploy
27:37
Josh: capital into this ever-changing absolutely outrageous
27:40
Josh: market but that is another week of the AI
27:43
Josh: roundup I hope you guys enjoyed this one I hope you enjoyed all the
27:46
Josh: episodes this week if you missed any they're pretty short normally 20 to
27:49
Josh: 25 minutes we talked about uh google had some pretty
27:51
Josh: amazing tools where now we can create songs and jingles for
27:54
Josh: the podcast we talked about anthropic at length and
27:57
Josh: now we have this kind of roundup that is covering just about anything
28:00
Josh: that you would need if you don't feel satiated we
28:03
Josh: have a second form of content which is the newsletter every wednesday
28:07
Josh: release the think piece this week was about a new company that i am
28:09
Josh: very excited about a startup that i guarantee 95 plus maybe 99 of the people
28:15
Josh: listening have not heard of so if you're interested in that or if you're interested
28:18
Josh: in the second roundup which is released every friday uh ejes is tackling that
28:22
Josh: one and i would highly advise subscribing the links are all in the description
28:25
Josh: down below ejes any final notes before we we part for the weekend nope
28:29
Ejaaz: Wherever you listen or watch us please like subscribe give us a rating it helps
28:32
Ejaaz: us out so much and we will see you on the next one.