The AI Race Just Changed Direction

[00:00] China's universities have cut more than

[00:02] 12,000 degree programs and they are

[00:04] redesigning their entire school system

[00:07] for artificial intelligence focused

[00:09] majors. We're going to look at what

[00:11] people call the holy grail of artificial

[00:13] intelligence breakthroughs, telling you

[00:15] where you lost your keys. Electronic

[00:17] Arts, the game company, actually

[00:20] released a new way for for you to

[00:23] advertise directly into video games. I

[00:26] was like, this is like another Google,

[00:29] right? I mean, if they build the

[00:30] plumbing for an entire ad marketplace

[00:32] that lives in three-dimensional space, I

[00:35] was like, I'm looking into it. We'll

[00:36] look at some thoughts from Rob Mason,

[00:38] take the 10,000 foot view, step back

[00:40] from the world of technology today, and

[00:43] figure out where are we, where are we

[00:45] going, and where have we been. I'll

[00:47] break down the viral economist post

[00:49] about how humanity is not ready for the

[00:51] upcoming intelligence explosion. I'll

[00:53] kind of share some of my thoughts on

[00:54] what they wrote. Author Tim Ferris, who

[00:56] I like I remember reading his books like

[00:59] decades ago, and I kind of forgot. I

[01:01] know he's been a podcaster for a while,

[01:03] but he just wrote this thing about like

[01:05] what AI is doing to the book publishing

[01:07] world. Then we'll get a little more

[01:08] philosophical about who should be

[01:10] benefiting from all of this change.

[01:12] David Kurpatre wrote the sharing

[01:14] algorithm. If AI is built on humanity,

[01:17] shouldn't humanity benefit? I'll share

[01:19] kind of the flip side of some of all of

[01:21] these AI tests that I'm always sharing

[01:23] because there was a really stringent

[01:25] math test and humans actually won it. AI

[01:27] did not take the crown even though we

[01:29] see all of these breakthroughs all the

[01:30] time. I mean, I don't know if they will

[01:32] be forever or how much longer, but today

[01:34] humans win. Some internet psychology

[01:37] that took me by surprise. Turns out

[01:39] harsh critics of things are usually the

[01:42] most loyal fans, especially of video

[01:44] game brands. But it actually kind of

[01:45] gives me a different perspective if

[01:46] they're actually the most loyal people.

[01:48] So over 100 cyber security experts are

[01:51] saying that the Fable 5 ban could

[01:53] backfire on the US. But experts are

[01:56] growing around one consensus and that is

[01:58] China is starting to crush the US in

[02:01] terms of AI growth. I mean there's a lot

[02:03] of ways to look at it. Certainly America

[02:05] has the lead in some very serious ways,

[02:07] but it's kind of wild how many edge

[02:09] cases we're starting to lose, especially

[02:11] when we have so much money and we had

[02:13] the head start. You know, good old chat

[02:15] GPT is now below 50% of the market

[02:18] share. And actually, after we talk about

[02:19] that, I'll dive a little bit into the

[02:21] big business of what's going on with

[02:24] this SpaceX IPO. It's it's a wild time

[02:28] to watch billionaires jockey for

[02:30] position. And there's it's pretty

[02:32] interesting positioning. And hey, when

[02:34] you're a trillionaire, it's like not

[02:35] even a tenth of your net worth. And you

[02:37] don't even have to pay for it. You just

[02:38] use your companies that have trillions

[02:40] of dollars. So, I get it. Just buy it. I

[02:42] mean, unless of course you care about

[02:43] citizens having any kind of say or

[02:45] control in our democracy, which nobody

[02:47] does.

[02:50] But look, Anthropic figured out the

[02:52] playbook about why coding is the single

[02:55] focus that you should be having and

[02:57] everyone else got a little distracted

[02:59] with side quests. So, we'll see if now

[03:01] that everybody sees what Anthropic sees

[03:03] where it all plays out. But first, if

[03:05] you don't mind heading over to Dylan

[03:07] Curious, actually, you're probably

[03:08] already there if you're watching this on

[03:09] YouTube, and hitting the share button,

[03:12] just like boom, share, copy, and send it

[03:14] to anyone. That would be super helpful.

[03:16] I know you guys did that on the last

[03:18] video, so thanks. It definitely helped.

[03:19] Look, you can see it did above average.

[03:21] Like by day four, we almost hit 10,000

[03:24] views. I made $77 on that video. That's

[03:26] crazy. I'm going to get some food.

[03:29] That's like four or five Chipotles. All

[03:31] right, so what are your thoughts? China

[03:32] is now cutting thousands of university

[03:35] degrees. Didn't even know there were

[03:36] that many. But they're thinking about an

[03:38] AI era and it kind of makes sense. I

[03:41] mean, it's a lot of government control

[03:42] over school, but also like we kind of

[03:45] need to get a little kick in the butt

[03:46] here to get ready for these huge changes

[03:48] that are coming. China's universities

[03:50] are making major changes to what

[03:52] students can study. The country is

[03:54] trying to line higher education up with

[03:56] a national development goal and that is

[03:59] to dominate and win against the United

[04:02] States in the future of artificial

[04:04] intelligence and technology and many

[04:06] other STEM programs. Arts and language

[04:09] programs are being dropped while tech

[04:11] focused fields are being favored. And

[04:13] it's what universities should do, right?

[04:15] They want to be preparing students for

[04:17] industries that are most important to

[04:19] the next decade. So that requires a

[04:21] national reshuffleling of education amid

[04:23] the backdrop of a much more techfocused

[04:26] future. All right. So if you've ever

[04:28] misplaced anything, which I think is

[04:29] pretty much all of us, I don't know how

[04:30] many do actually we should put in the

[04:32] comments like how many times a week do

[04:33] you lose something like your phone?

[04:35] Anything like phone or like scissors if

[04:37] I'm trying to like open some boxes or

[04:38] something. I probably misplace things

[04:41] like a dozen times a week. Yeah. Like a

[04:43] couple times a day or something. And it

[04:44] kind of sucks. I'm like or especially my

[04:46] AirPods. Actually, probably maybe more

[04:48] than that. I'm always using the find my

[04:50] feature where I'm like beep beep beep

[04:51] and I'm like oh it's on that side of the

[04:52] house. Yeah, that's right. But listen,

[04:54] people lose stuff all the time,

[04:56] especially groups of people working on

[04:58] construction sites or big companies like

[05:00] it can just get confusing. You're

[05:02] focused on your work, you set something

[05:03] down, someone else uses it, whatever. If

[05:06] you think about it, AI with camera

[05:08] vision watching everyone should be

[05:11] easily able to just be queried. Hey,

[05:13] where did I leave my phone? If that

[05:15] information is in context, it will

[05:17] remember when and where it was lost. And

[05:18] even if it's just visually available, it

[05:20] should be able to just, you know, hunt

[05:22] it down. So researchers are building

[05:24] robots that may soon remember places

[05:26] more like humans do, including where an

[05:29] object was last seen. And you can jump

[05:31] into the eyes of the AI to see the same

[05:34] thing that you should have remembered.

[05:36] So a worker can remember where she or he

[05:38] left a part in a factory last night. And

[05:40] a robot can do the same. A robot usually

[05:43] cannot do that well at this job because

[05:45] it struggles with memory that connects

[05:47] places and times together. But MIT

[05:49] researchers have built a new system to

[05:51] help robots learn that kind of memory.

[05:53] It is called DAM. But it's got extra A's

[05:56] in there. So it's not a swear word. It's

[05:59] D A A M. It stands for describe

[06:02] anything, anywhere, anytime, any moment.

[06:05] Oh my gosh. Actually, the acronym is

[06:07] missing an A. It's D A A M. Wait, what?

[06:11] So 1 A is anything. Second A is

[06:13] anywhere. Third one is any time. And

[06:16] then at any moment. You could have you

[06:18] could have thrown a fourth one in there.

[06:19] They missed an opportunity there. But

[06:20] anyways, as the robot moves through a

[06:22] space, it builds a 3D map. But it also

[06:25] adds rich labels to what it sees as it's

[06:27] going along. For example, it could

[06:29] remember a bike rack outside of a campus

[06:30] building. Maybe it has five bikes on it,

[06:33] one red, one yellow, one with a flat

[06:35] tire, etc. The system then groups

[06:37] objects by location so the robot can

[06:40] answer in plain language where something

[06:42] is later. And that is a new AI system

[06:45] where you can ask a question like where

[06:46] is the part that we started assembling

[06:49] last night. You don't even have to have

[06:50] it be something that it could tag like a

[06:53] part that doesn't even have a specific

[06:55] name because it can figure out what you

[06:57] mean based on the context of it. Knowing

[06:59] it was last night, knowing it was

[07:01] something you were assembling and it can

[07:03] just eliminate all of those other

[07:04] choices. And it has a special system for

[07:06] choosing clear images, describing

[07:08] several objects at once and avoiding

[07:11] labels on the same object again and

[07:13] again. And that actually was something

[07:15] that was really hindering some of the

[07:16] systems before this. So, what do you

[07:18] think? Do you think that's too much

[07:19] privacy invasion or do you just like the

[07:21] idea of knowing an AI knows everything

[07:22] about you and where you lost everything?

[07:24] All right, let's talk about this new EA

[07:26] advertising thing. Like maybe if you're

[07:28] already in the game world, this isn't

[07:29] like such a surprise to you. Maybe

[07:31] there's been like ads kind of shoved

[07:33] down your throats for a long time, but I

[07:35] don't really play much games, so I was

[07:38] not ready for this. I always imagined EA

[07:40] was just going to make money based on

[07:42] the games that it sold. I kind of knew

[07:44] it was moving towards a world where

[07:45] there's like recurring fees or maybe

[07:46] they like, you know, charge you for skin

[07:48] upgrades or whatever, but EA is going to

[07:50] get into the ad business. Like now

[07:52] they're going to fight for your

[07:52] attention inside of a game. Like what?

[07:55] So Electronic Arts has launched EA

[07:57] advertising. This is a new system that

[07:59] lets brands appear directly inside its

[08:01] games. That means ads can show up on

[08:03] stadium signs, digital billboards,

[08:05] scoreboards, broadcasting graphics,

[08:07] custom in-game items, and branded

[08:09] challenges. Is this going to ruin games

[08:11] or is this okay or is it make games

[08:14] better be free at least? The idea is to

[08:15] make ads feel more like part of the game

[08:18] setting in the same way brands already

[08:19] appear in real sports broadcasts. He

[08:22] says it reaches more than 120 million

[08:24] players each month. Oh my god, that's

[08:26] like the entire Super Bowl audience. It

[08:28] also says players are spending huge

[08:30] amounts of time in games like Madden,

[08:32] the NFL, EA Sports, where matches,

[08:35] seasons, and live sports style moments

[08:37] happen consistently. Like, you'd think

[08:39] if the NFL licenses the entire NFL to EA

[08:43] to make a game out of, wouldn't the NFL

[08:45] still own the brands or can EA just do

[08:47] this? And what does this mean for the

[08:48] future of VR and AR where everything's

[08:51] just being automatically generated? But

[08:53] yeah, this the kind of architecture

[08:55] behind it already includes reward-based

[08:57] goals, vanity items, branded content,

[08:59] live events, creator tools, social blade

[09:01] features, and community programs. What

[09:03] EA was the company that was just bought?

[09:05] It was bought, right, for like $60

[09:06] billion or something? Was EA recently

[09:08] bought? And if so, who was the buyer and

[09:10] price? What was the logic? Yeah. Okay,

[09:13] so $55 billion leveraged buyout, Saudi

[09:16] Arabia's sovereign wealth fund. That's

[09:18] right. I remember talking about that. I

[09:20] wonder if this was their plan. like they

[09:21] were just thinking, gosh, it's going to

[09:22] be so valuable if we can take some of

[09:24] Google's profits from ads and get them

[09:27] into games. Honestly, we're like later

[09:29] in this video, we'll talk about SpaceX

[09:30] buying Cursor for 60 billion. Th it

[09:33] would not shock me at all if SpaceX

[09:36] purchased EA for this much money because

[09:38] I mean, he's talked about creating a

[09:39] video game company. And if you already

[09:41] have Cursor and you already have like a

[09:43] bunch of inference systems that you can

[09:45] take back from Anthropic anytime you

[09:47] want, you know what? That might honestly

[09:49] happen. Maybe EA implements this ad

[09:52] thing, generates some insane revenues,

[09:54] and then sells it for what do you think?

[09:56] $180 billion to SpaceX in two years.

[10:00] There you go. That's my prediction. Let

[10:02] me know in the comments if you if you

[10:03] want to be on board with it. Over,

[10:05] under, happen, not happen, true, false,

[10:07] however you want to bet. All right,

[10:08] let's look at mid June 2026. Where are

[10:11] we at right now in the landscape of AI?

[10:14] Because you I mean, you know, if you

[10:15] watch this video how fast it feels like

[10:18] it's been going recently. Um, AI is

[10:20] moving fast, right? And I think I you

[10:22] see cracks all over governments and like

[10:24] banks. I think young voters are now

[10:27] really thinking about it like it comes

[10:29] up sometimes with politicians in a way

[10:31] that it never used to. And Rob Mason

[10:32] kind of broke this all down for us. So

[10:34] in one major a AI lab, Anthropic to be

[10:38] precise, they published a plan that

[10:40] threatened mass job losses as a real

[10:42] possibility. We talked about this in the

[10:44] last video too. Universal basic income,

[10:46] public wealth funds, some of those

[10:47] things. Um, timing was obviously messy

[10:50] because they're also trying to go IPO

[10:51] and they want to look like super

[10:52] profitable and super capitalistic at

[10:54] that time. So that, you know, cha-ching.

[10:56] So critics were asking whether this was

[10:58] real policy support or just smart IPO

[11:00] positioning, basically being like, we're

[11:02] going to gobble up all that value.

[11:03] Invest now. And then few days later,

[11:05] government says, no more Fable 5. Take

[11:08] away your like frontier model. Here's

[11:10] where we're at. The government sometimes

[11:12] is like totally handsoff. We just want,

[11:14] you know, business to like go. It feels

[11:16] like kind of a corrupt government.

[11:17] nobody's going to stop anything and like

[11:19] just big companies and K-shaped

[11:21] economies are going to happen. So, the

[11:22] government has sounded hands off on AI,

[11:24] but then it's used national security

[11:26] power to control access without clear

[11:28] legislation with all of this mythos

[11:30] stuff. Europe and the UK are reacting by

[11:33] taking it more seriously, but they're

[11:35] talking about sovereign AI, meaning they

[11:37] want to control their own advanced AI

[11:39] systems and they're trying to keep money

[11:40] inside their own borders. There's

[11:42] definitely a lot of fences going up all

[11:44] over the world. And at the same time, US

[11:46] senators, major banks, young voters,

[11:48] they're all moving into the debate in

[11:50] different ways. And some companies are

[11:52] just openly talking about AI replacing

[11:54] roles through automation. And it's just

[11:57] it, you know, it is their fiduciary

[11:59] duty, I suppose, but it also just

[12:01] doesn't resonate well with a society

[12:03] that's coming into the workforce. Young

[12:05] people are becoming less excited and

[12:07] more angry about AI. So, we'll have to

[12:09] see what happens. So just on top of like

[12:12] where politics and technology are

[12:14] intersecting, there's also just a

[12:15] natural trend, right? That this is like

[12:17] Ray Curtzwhile has been talking about it

[12:18] forever and we're getting into the

[12:20] hockey stick part of it. So this article

[12:22] is called humanity isn't ready for the

[12:24] coming intelligence explosion and it's a

[12:27] classic talking point on this channel.

[12:29] Uh AI leaders and everyone who's

[12:32] enabling them are moving humanity

[12:34] towards a race that is making smarter

[12:36] and smarter machines. And it's doing

[12:39] that without a real serious good

[12:41] trustworthy safety plan around it. Tons

[12:44] of emergent properties, tons of things

[12:45] that can't be predicted when systems are

[12:47] actually live and in people's hands,

[12:49] tons of things that are completely un

[12:52] unpredictable when different agents are

[12:54] acting with different intentions and

[12:56] different goals in some kind of like

[12:58] bigger system. So could AI be advancing,

[13:01] becoming more dangerous faster than

[13:03] governments are ready for? Absolutely.

[13:06] That's probably likely, not just a

[13:08] possibility. The article is comparing AI

[13:09] risk to nuclear risk. Nuclear plant is

[13:12] expected to have a tiny chance of a

[13:14] disaster. You know, a lot of people

[13:15] disagree with it and there are smart

[13:17] people that disagree with it. So, I

[13:18] wouldn't, you know, bank on it. That's

[13:20] just what feels right to me. And I'm

[13:22] sure a lot of new technology felt a lot

[13:24] scarier than it was when people first

[13:26] heard about it, before it was

[13:27] implemented, before structure came. So,

[13:29] it might move down. But, you know, the

[13:32] concern is not just coming from

[13:33] outsiders. It's coming from industry

[13:36] insiders too. Some of the people inside

[13:37] the biggest labs especially like

[13:39] recursive self-improvement. I mean you

[13:41] can see Anthropic and Musk moving

[13:43] towards it at the same time worried

[13:45] about whatever like happens when

[13:47] self-improvement is AI writing its own

[13:49] next model then generating it and then

[13:51] it writing its own code. I mean this is

[13:53] why a $60 billion investment in cursor

[13:55] might be nothing to SpaceX in the long

[13:57] run when it's a you know hundred

[13:59] trillion dollar broke the economy kind

[14:01] of company. And once there's an

[14:02] intelligence explosion, if the

[14:04] guardrails aren't good, there might be

[14:06] no way back. How are you going to, you

[14:08] know, there was that there's this

[14:09] metaphor once where somebody said, "Go

[14:11] up on the top of a hill and like cut

[14:13] open a pillow of feathers and like let

[14:16] the wind carry those feathers away and

[14:18] like now go put them all back in in the

[14:20] pillow." Like you can open it up pretty

[14:22] easily. You can just take a knife to it,

[14:24] cut it open, let all those feathers out.

[14:25] But to actually like put them back

[14:27] together, it's a much harder problem.

[14:30] and evolution and and growth like this

[14:32] now like actual intelligence in the way

[14:34] we see it in the biological world it

[14:35] once it gets going you don't just go

[14:38] back so the economist is now arguing

[14:40] that America and China in particular

[14:42] because these are the two countries

[14:44] controlling the vast majority of this

[14:46] progress and need to create serious AI

[14:48] safety agreements right now not based on

[14:52] trust based on verification based on

[14:54] inspections based on clear red lines if

[14:56] you go a little too far and slow down

[14:58] progress so be it like we'll have plenty

[15:00] of time when these things are

[15:02] exponentially growing to reap the

[15:03] benefits, but really right now

[15:05] biological weapons, cyber attack, fraud,

[15:08] child abuse, like there's child abuse

[15:10] material like being generated like all

[15:12] this stuff is coming really fast. So the

[15:14] article ends with a warning that

[15:16] civilization may fail when powerful

[15:19] tools grow faster than their ability to

[15:21] govern them. What do you think? Let me

[15:23] know in the comments. Yeah, I mean it's

[15:25] hard for me to say pause, right? Like

[15:27] I've never quite been a pause AI guy.

[15:29] Although, if I could magically just stop

[15:32] everything right now, I would. I just

[15:34] don't really think of any realistic way

[15:36] to just tell everybody like pause and

[15:38] have important people pause and have

[15:39] that be a a good outcome. But I kind of

[15:42] do agree that maybe just big governments

[15:45] right now getting serious about like red

[15:49] lines might and inspection like all

[15:51] getting all that system in place for

[15:53] inspections might be a really smart

[15:55] move. Now, I know bureaucracy is

[15:57] bureaucracy, but like let a bunch of

[15:59] people see what's going on behind closed

[16:01] doors and like force their hands as much

[16:03] as you can. Like one good thing about

[16:05] China and America, if we both let each

[16:07] other kind of look and see, there would

[16:09] be stealing and there would be copying,

[16:11] but on the other hand, we'd be seeing,

[16:13] you know what I mean? Like those

[16:14] conversations would get out and when

[16:16] super dangerous things happen, everybody

[16:17] might be aware of it at least. I go talk

[16:19] to Tim Ferrris, 4-hour work week. Like I

[16:21] remember he was he was like a in my

[16:23] world like a little cultural phenomenon

[16:25] in my a couple decades ago. I haven't

[16:27] heard much from him but he actually

[16:29] mentions the 4-hour work week which was

[16:30] like his self-help book from a long time

[16:32] ago and how it's been selling and how

[16:35] the industry is getting crushed. So

[16:37] having him talk about AI I found sort of

[16:39] surprising. So his core argument is that

[16:41] AI may be collapsing the market for how

[16:44] to non-fiction because people

[16:46] increasingly want the answer not the

[16:48] path there. And that's interesting.

[16:50] Like, is that another byproduct of using

[16:52] these LLMs all the time? So, if a book's

[16:55] main value is to tell me all the steps

[16:56] that I need to take and a chatbot can

[16:58] now feel faster, cheaper, more

[17:00] personalized, and more convenient,

[17:02] what's the point of self-help books? But

[17:04] Tim Ferris is also arguing that the real

[17:06] separation is between information and

[17:08] transformation. So, information is

[17:10] something like, give me five steps to

[17:12] lose fat, but transformation is walk me

[17:14] through a carefully designed journey

[17:16] that makes me actually change. And that

[17:18] difference matters. Chat bots can spit

[17:21] out fat loss protocols in 15 seconds.

[17:24] But Tim's example is when he took the

[17:27] 4-hour work week, which is this whole

[17:29] body transformation thing, and distilled

[17:31] it into some bullets and then sent it to

[17:33] friends, none of them acted on it. But

[17:35] thousands of readers who followed the

[17:37] full book's journey felt what they

[17:39] needed to like incorporate the

[17:41] information deep enough to actually

[17:42] change their lives. So, when I read this

[17:44] and then I started using LMS like I

[17:47] always do to um I do like moving some

[17:49] stuff around and trying to figure out

[17:50] how to like set up everything. I was at

[17:54] the end thinking did I actually learn

[17:56] how to do this for someone else? And I

[17:58] thought to myself like kind of not right

[18:00] like I if if someone else asked can you

[18:02] help me with this? I would say like yeah

[18:04] bring up bring up chat GPT and then ask

[18:07] this question and then it'll walk you

[18:09] through the steps. I think we went here

[18:10] first and I would follow it again. and I

[18:12] wouldn't actually be able to tell them.

[18:14] So, I didn't process it and it kind of

[18:16] freaked me out. And he's saying that

[18:18] prescriptive non-fiction is getting hit

[18:20] really hard. So, he says that his own

[18:22] print catalog may be down about 80%

[18:25] compared to the 2022 year. So, there's

[18:27] clearly less people just buying books.

[18:29] But self-help, he thinks, is especially

[18:32] vulnerable. Eventually, the same threat

[18:34] is going to be extended to YouTube, like

[18:36] the videos I make. like you might might

[18:38] not get as much out of this whole thing

[18:39] as you could just like go down to Gemini

[18:41] and distill this into like five news

[18:43] bullet points or interesting thoughts

[18:45] and then move along, you know, like I

[18:47] understand maybe some of you would want

[18:48] to do that. Podcast, courses,

[18:50] newsletters, advice blogs, like I guess

[18:53] if you just want to extract actionable

[18:55] information instantly, that's a better

[18:57] route to go. Next up, let's talk about

[18:59] the sharing algorithm. If AI is built on

[19:02] humanity, shouldn't humanity all benefit

[19:04] David Kurpatre? So he's arguing that AI

[19:07] didn't really come out of nowhere,

[19:08] right? Like artificial intelligence was

[19:10] built from human language, books, music,

[19:12] code art research news questions

[19:14] jokes, years of public investment, lots

[19:17] of people posting things on social

[19:18] media. But then companies built powerful

[19:20] AI systems and all of a sudden that work

[19:22] mattered for training. So if AI creates

[19:25] huge wealth, should only the companies

[19:27] that invented the thing that sucked it

[19:29] up and learned from it benefit or should

[19:31] all of us? I mean, the idea that a

[19:33] shared resource creates wealth that the

[19:35] public can own seems like one of the

[19:37] kind of futures that we almost

[19:38] inevitably will fall into. Like humans,

[19:41] the interesting things humans do, the

[19:43] data that we make, like that could be

[19:45] pretty valuable. I do think synthetic

[19:47] data is actually going to make

[19:48] everything people do kind of, you know,

[19:50] not as important as we'd hope. I I don't

[19:53] know how long human data will matter,

[19:55] but it probably will always matter.

[19:56] maybe at least enough that companies

[19:58] could somewhat pay taxes or the

[20:01] government would be rich enough or

[20:02] something that there could be some kind

[20:03] of universal basic income. And I will

[20:05] say for a long time I actually thought

[20:07] what's going to happen to AI is they're

[20:10] going to digest the internet like OpenAI

[20:11] was the first to kind of grab it and

[20:14] then I thought maybe right after them

[20:15] Google and a couple others and then

[20:17] they're going to produce so much garbage

[20:18] on the internet from chat bots that

[20:20] there won't be anything to put into the

[20:22] system. But over time I have come to

[20:24] just think that oh synthetic data

[20:26] actually I think it actually works.

[20:27] There's different novel ways to pull

[20:29] data off the internet. There's different

[20:30] ways to generate language in a way where

[20:34] it seems like the systems do learn. Um

[20:36] in some cases maybe they do kind of move

[20:39] to the middle and don't like have that

[20:40] massive creativity. But also I don't

[20:43] know. I got a feeling I got a feeling

[20:45] that's not going to be as as obvious as

[20:47] I once thought.

[20:50] Especially

[20:53] with like physics simulations and stuff

[20:54] like that and diffusion models of just

[20:56] like oh okay you can you can actually

[20:58] kind of come up with some pretty unique

[20:59] random stuff to learn from over even

[21:02] though you generated it. It's weird but

[21:03] it seems to work. You just kind of

[21:05] iterate towards better and better. But

[21:07] in some cases humans still matter. How

[21:09] about this one? Humans outperform AI on

[21:12] this highly rigorous mathematical test.

[21:14] This is good. I thought we were going to

[21:15] get crushed after like reading all these

[21:17] like math olympiad was destroyed by open

[21:20] AI and by deep mind. But a project

[21:23] called first proof gave AI models a very

[21:26] hard math exam. This was not the kind of

[21:28] school math. This was 10 research level

[21:31] problems from mathematicians own work.

[21:34] The problems were completely unpublished

[21:36] at the time of testing. So the models

[21:38] could not have just seen the answer and

[21:39] repeated them. 30 mathematicians checked

[21:42] their answers and that made this test

[21:44] very different from any of the earlier

[21:46] AI math demos that have ever been done

[21:48] before. This was new work. It was done

[21:50] without human help and it was judged by

[21:51] real experts. Four AI systems took this

[21:54] test. OpenAI entered Chat GPT 5.5 Pro

[21:59] and then the university teams built

[22:01] automated systems called harnesses on

[22:03] top of chat bots like Gemini and Claude

[22:06] for testing. Also, the harnesses asked

[22:08] the models questions. check the answers

[22:10] and pushed to improve. Now, the best

[22:13] system was chat GPT. It was GBT 55. It

[22:17] solved six of the 10 problems. It did

[22:19] not match human top mathematicians. The

[22:23] results showed AI can help with serious

[22:25] math, but not yet a full replacement for

[22:27] human expert problem solving. Wuzzing.

[22:30] What do you guys think about that one?

[22:33] Humans still in the lead. All right.

[22:35] also had a new epiphany that some of the

[22:36] harshest critics potentially even not

[22:38] just in games but like in my comment

[22:40] section might be the most loyal fans. I

[22:42] thought I thought the harshest people

[22:44] were just I was like gez guys like just

[22:46] chill. But no, harsh critics are the

[22:47] most loyal fans of video game brands.

[22:49] Study shows a new study looked at the

[22:51] way video game fans stay loyal to brands

[22:53] like Call of Duty and Battlefield. And

[22:56] instead of bringing people into a lab,

[22:57] the researchers used AI to study more

[23:00] than 23,000 Reddit comments. They wanted

[23:02] to see how loyalty shows up in real fan

[23:05] communities. And the results were not

[23:07] what people expected. There wasn't one

[23:09] clear pattern either. So, Call of Duty

[23:11] fans showed showed loyalty through

[23:13] emotion, stories, and shared memories.

[23:16] But Battlefield fans showed it through

[23:18] skill, technical details, and respect

[23:20] for how the game works. I was like,

[23:22] "Wow, these are really deep communities,

[23:24] like different languages, same deep

[23:27] loyalty." And the most interesting

[23:28] finding is that harsh criticism is

[23:31] usually a sign of commitment. Fans do

[23:33] not always complain because they hate

[23:34] the brand. They complain because they're

[23:36] invested and they want the game to

[23:38] improve. The same person who tears apart

[23:40] a flaw may defend the broad movement in

[23:43] an outsider attack. So the researchers

[23:46] are telling brands like you should stop

[23:48] trying to silence useful criticism. Like

[23:50] they're always trying to cut. Remember

[23:51] the uh was that Cyberpunk that release?

[23:54] Everyone's like just tearing on it. like

[23:55] those people probably really wanted the

[23:57] game to be good. They also learned that

[23:58] like fake corporate language like really

[24:00] pisses off fans. So maybe try to be a

[24:02] little more authentic when you talk to

[24:04] them. But either way, food for thought.

[24:06] So let me know in the comments. Do you

[24:07] think that the um mythos uh pullback was

[24:11] actually good for the United States in

[24:15] their position in the world? Do you

[24:17] think it's good for companies to not

[24:18] have cyber threats or do you feel like

[24:21] um it actually made us weaker? because

[24:23] there's some debate could go either way.

[24:26] But there's this open letter signed by

[24:28] hundreds of cyber security experts that

[24:30] asks US officials to lift export

[24:33] controls on anthropics fable and mythos

[24:36] AI models. The writers say that AI is

[24:38] already changing cyber security and it

[24:40] can help find software flaws and write

[24:42] exploits which are the tools that attack

[24:44] those flaws. But they argue that

[24:46] anthropics models are not uniquely

[24:48] dangerous. The major and open source

[24:50] models can already do similar work.

[24:52] their concern is that taking strong AI

[24:55] tools away from the defenders gives an

[24:57] advantage to the attackers or do you

[25:00] think that the attackers not having it

[25:02] make them weaker? You know what I mean?

[25:04] It's so tricky. They're arguing that

[25:05] security teams need these models to find

[25:07] and fix bugs in new code and old systems

[25:10] before the adversaries do. The letter

[25:12] also says that Fable already had strict

[25:14] protections against offensive cyber use.

[25:17] So banning access in total in their view

[25:19] just creates the market fear and weakens

[25:21] American AI without leading to any clear

[25:24] risk reduction. So consensus is growing

[25:27] that China is crushing the US at AI.

[25:30] There is a few things that AI in the US

[25:33] is still clearly in charge of and front

[25:36] and leading the pack. In terms of raw

[25:39] intelligence, the frontier models and

[25:41] the amount of money spent on

[25:42] infrastructure, we're definitely

[25:44] winning. But global opinion is shifting

[25:46] the AI race away from the United States

[25:49] and towards China. So a poll by Public

[25:52] First asked people in 15 other countries

[25:55] which nation they think is leading in AI

[25:57] capability and innovation. And in 11 of

[25:59] those countries, respondents said that

[26:01] China has passed the United States. And

[26:03] what were the countries that still

[26:04] thought the United States was leading?

[26:06] Well, one was US, the United States. So

[26:08] we think we're doing great. Also

[26:10] Vietnam, India both also agree with us

[26:13] and same with Japan. But even inside the

[26:16] US, confidence was not strong. Just 51%

[26:19] of Americans said their own country is

[26:21] leading. So three, you know, just 51% is

[26:24] like barely over half of us. Canada,

[26:26] Mexico, France, United Kingdom, they're

[26:28] all thinking that China is beating us.

[26:30] It also reflected a certain frustration

[26:32] with American AI models, which the

[26:33] article describes as corporate,

[26:35] extractive, and hostile towards workers.

[26:38] These countries saw China as a country

[26:40] that presented itself as having stronger

[26:43] competition and more people centered

[26:45] regulation. The main claim is not that

[26:47] like China is ahead. It's that the

[26:49] world's trust is moving away from the

[26:51] United States and towards China and that

[26:53] creates a legitimacy problem for Silicon

[26:56] Valley. What do you think? Are you out

[26:58] of the 51% do you lean towards thinking

[27:00] like American AI is I don't know more

[27:03] people centric or it's going to be

[27:04] better less extractive? I mean,

[27:06] obviously it's not going to be like less

[27:08] corporate or less extractive. We're the

[27:09] United States, but is it still still

[27:11] winning or in the long run, do they have

[27:13] a better sort of pipeline built? I don't

[27:15] know. I'll make this one quick, but just

[27:17] so you guys know, Chad GBT is now under

[27:19] 50% for the first time. So, what's going

[27:22] on here? I want to dive into it a little

[27:23] bit. I'm also going to touch on the

[27:25] cursor and the codeex and like what

[27:28] companies are doing as far as their

[27:29] positioning each other and uh what

[27:32] exactly was $60 billion like just that

[27:35] is that that's so overvalued in a nonAI

[27:38] world that is so almost undervalued in

[27:41] the long-term thinking of an AI frontier

[27:45] model play. So, it's just the whole you

[27:48] if you buy the story that like AI is the

[27:51] invention of intelligence and whoever

[27:53] gets there first gets like so much

[27:55] reward then you know what that that

[27:58] makes sense and honestly like probably

[28:00] no matter what you buy at this point if

[28:02] you truly believe that's the story you

[28:04] could justify it. I mean look

[28:05] realistically you have to even kind of

[28:07] wonder if SpaceX what's it trading at

[28:09] like two trillion market cap and then uh

[28:12] Tesla's another 1.3 1.5 trillion. So if

[28:16] you take all of that Elon's in chart and

[28:18] plus like Neuralink and all these other

[28:19] things, he's got to be coming up on $4

[28:22] trillion worth of buying power and a

[28:24] trillion dollars personal wealth,

[28:27] you know, and if OpenAI just goes

[28:29] public, if Anthropic goes up public at a

[28:32] trillion dollars, like maybe he could

[28:34] buy both of those companies or one of

[28:36] them and just merge them in or

[28:38] something. So, you know, I don't know

[28:40] who's big enough really. I feel like

[28:42] Google and maybe Microsoft are too big

[28:44] to be acquired. maybe meta, but there

[28:47] might be we might be coming down to like

[28:49] in the long run and might not even be

[28:51] that long. Like I'm thinking just two,

[28:53] three years, we might see some crazy

[28:55] consolidations. Like there could be a

[28:56] true merger between like Meta and

[28:59] Microsoft or something because there's

[29:01] something happening with like Google and

[29:03] SpaceX. I think Google owned a bunch of

[29:05] SpaceX stocks, so there's they're like

[29:06] tied into that a little closer. But uh

[29:08] let's look at this article. Michael

[29:10] Spencer, Jeff uh Mohouse wrote, "Codeex,

[29:13] Cursor, and the great beyond. Open AI

[29:15] will be put under pressure by Meta AI,

[29:17] SpaceX, and China, not just Anthropic's

[29:20] Enterprise AI momentum." So, there were

[29:23] less than 60 people that owned the stock

[29:26] for Cursor. So, $60 billion made them

[29:29] all billionaires, which is like

[29:30] mind-blowing to me because it's a fork

[29:33] of VS Code. It was like an open- source

[29:35] software, and it was awesome. and it was

[29:38] a way to just plug large language models

[29:39] in and code. But, you know, they built

[29:41] an amazing tool for enterprise and it

[29:43] just has so many people using it and it

[29:45] has such an interesting information loop

[29:47] and it's so valuable to an AI company

[29:50] like SpaceX that's behind anthropic in

[29:54] terms of coding. And if you believe

[29:55] coding is the flywheel effect where if

[29:58] you get code right, then you can catch

[30:00] up on everything else. Like there's no

[30:01] need to have a video model right now or

[30:03] an image model. just like just code and

[30:05] then we'll code everything that we need

[30:07] in the future and then it'll code some

[30:09] super advanced model that does even

[30:11] better video and audio or whatever all

[30:13] these side quests that you want to

[30:15] tackle. Like it will just code them all

[30:16] up for you and you will win them all.

[30:18] But it puts OpenAI, which now is under

[30:21] 50% market share, in a strange position.

[30:24] Like they seemed untouchable a couple

[30:26] years ago, but they're under pressure

[30:27] from every side. And now SOAR has been

[30:30] cancelled and um Stargate, it doesn't

[30:32] feel like they're going to get as much

[30:34] money as they thought and you know

[30:35] there's all this other competition going

[30:37] on. So Ched GBT, no doubt about it,

[30:39] still has a huge reach. But I mean I

[30:41] don't find myself in it as much as I

[30:43] ever used to. I mean first off that

[30:44] whole ad thing was bothering me. But

[30:47] also, when you're in Chrome and there's

[30:50] Gemini right there and you're kind of in

[30:52] a Google workspace or you're in a

[30:54] Microsoft 360 workspace, they have their

[30:56] own models where you're going to go to.

[30:58] When you have questions inside meta,

[31:00] they have their own unique data source.

[31:01] And chat GBT still has its spot, but it

[31:04] just I don't know, it does feel kind of

[31:06] like I I got this sense that Google's in

[31:09] the best position in the long run. that

[31:10] that's my main bet. But I'm also I

[31:13] wouldn't be surprised especially with

[31:15] SpaceX having like essentially infinite

[31:17] money and having having so much extra

[31:22] compute. I mean they Elon built so many

[31:24] servers so fast and they really are

[31:26] valuable. Like they're I think he's paid

[31:28] off all of them just with the money he's

[31:29] getting from Anthropic. So that was like

[31:32] an insanely profitable decision to just

[31:34] push those things out. then China as a

[31:37] whole can mount a pretty good, you know,

[31:40] a little bit of that communism there can

[31:41] like really just force companies to work

[31:43] together and try to take on individual

[31:46] companies in the US which might force

[31:48] individual companies to go to these like

[31:50] crazy mergers. But anyways, ChadB still

[31:52] has huge reach but the article says that

[31:54] its market share is failing because

[31:56] Gemini, Claude, Meta AI, Space XAI and

[32:00] Chinese models are putting that pressure

[32:02] on them from every direction. Really,

[32:04] Meta should have made that. If

[32:05] Zuckerberg was serious, he's putting a

[32:07] lot of money on talent. Like, they

[32:08] probably should have bought Cursor.

[32:11] Maybe Microsoft should have bought

[32:12] Cursor. I mean, they already have

[32:13] GitHub, which is pretty powerful. So,

[32:15] maybe I understand why Microsoft didn't.

[32:17] But Curser already had fast revenue

[32:19] growth, strong use among software

[32:21] engineers, and a product people trusted.

[32:23] And that makes it a direct threat to

[32:25] OpenAI's Codeex plans. Codeex used to be

[32:28] simple. It was OpenAI's coding

[32:30] assistant. You gave it a software task

[32:32] and it helped turn that idea into code.

[32:34] But now Codeex is expanding beyond

[32:36] programming. AI wants it to handle

[32:38] broader work across tools like Gmail,

[32:41] Calendar, PowerPoint, and computer use.

[32:43] And don't get me wrong, this is probably

[32:45] the right move for them. I just don't

[32:47] think they're going to pull it off. I

[32:48] just think that's hard. But if I was if

[32:51] I was Sam Alman, I also actually would

[32:52] be doing this cuz you don't want to

[32:54] think small at this moment. Maybe maybe

[32:56] they'll have something that kind of

[32:58] carries them through the day. But the

[33:00] risk is execution. If OpenAI can merge

[33:03] chat GPT and codeex really seamlessly,

[33:06] if they can somehow make the average

[33:08] person just sit down with the tool and

[33:10] start kind of accidentally dreaming up

[33:13] things that are coded into reality and

[33:16] hosted and profitable. And it really is

[33:19] like the Apple moment for them. They

[33:22] might they might be able to continue

[33:24] this route. If not, I think I think the

[33:28] walls are closing in on them.

[33:31] I mean, don't get me wrong, everyone

[33:32] there is going to end up being a

[33:33] billionaire and like Chad GPD won't just

[33:35] be gone, but it's not Sam Alman's moment

[33:39] to potentially outdo Elon will be gone.

[33:42] All right, so if you enjoyed this video,

[33:44] please share it with a friend. You're on

[33:46] YouTube, there's a little share button

[33:48] right there. Please send it to somebody.

[33:50] But if not, leaving a comment, hitting

[33:52] that hype button if you guys remember

[33:54] that, thumbs up, all that stuff. Like

[33:56] any interaction with the the video

[33:58] helps. And just the fact that you're

[33:59] still here watching it means a lot to

[34:01] me. So, thanks and I will see you in the

[34:03] next video.