1 00:00:00,020 --> 00:00:03,120 Josh: It's been a while since we've had an opportunity to talk about the Game of Thrones 2 00:00:03,120 --> 00:00:07,320 Josh: within the AI wars, the employees stealing, the co-founders leaving. 3 00:00:07,580 --> 00:00:10,900 Josh: And this week, we have a lot of news coming out of the world of ex-AI, 4 00:00:11,120 --> 00:00:14,440 Josh: who has now had five out of their eight total co-founders leave, 5 00:00:14,760 --> 00:00:19,600 Josh: two more just in the last week, and a seeming huge exodus of people who are leaving the company. 6 00:00:19,780 --> 00:00:23,540 Josh: In addition to that, we have this brand new video model that is indistinguishable 7 00:00:23,540 --> 00:00:25,360 Josh: from reality called Seedance 2. 8 00:00:25,480 --> 00:00:28,940 Josh: It's from China, and it has no regard for copyright, which makes it the best 9 00:00:28,940 --> 00:00:33,080 Josh: video model in the world and that also leads us to our other topics which are 10 00:00:33,080 --> 00:00:37,560 Josh: going to be ai safety in general and is this going to take your jobs the dumeristic 11 00:00:37,560 --> 00:00:41,300 Josh: take that now has what like 50 million views it's the most viewed article in 12 00:00:41,300 --> 00:00:45,320 Josh: the world so we have a lot to get through today let's start with the xai exodus so 13 00:00:45,320 --> 00:00:48,620 Ejaaz: For the game of thrones fans out there um you'll be familiar with the episode 14 00:00:48,620 --> 00:00:53,020 Ejaaz: titled the red wedding where uh the plot twist sorry for those of you have watched 15 00:00:53,020 --> 00:00:56,460 Ejaaz: it is everyone gets murdered pretty much. 16 00:00:57,160 --> 00:01:00,980 Ejaaz: And scrolling through my timeline this week, it felt like that people were getting 17 00:01:00,980 --> 00:01:02,680 Ejaaz: fired and let go left and right. 18 00:01:02,840 --> 00:01:08,460 Ejaaz: And the main culprit was XAI, who ended up riffing or laying off 50% of their team. 19 00:01:08,660 --> 00:01:12,880 Ejaaz: But the story unfolded as such. I've got an example on our screen here from 20 00:01:12,880 --> 00:01:18,660 Ejaaz: Jimmy Barr, which is one of the eight co-founders that you mentioned from XAI last day at XAI today. 21 00:01:18,820 --> 00:01:21,660 Ejaaz: And he goes on to explain about how he's so grateful to elon the 22 00:01:21,660 --> 00:01:24,800 Ejaaz: mission and i then scrolled a bit more and 23 00:01:24,800 --> 00:01:27,860 Ejaaz: i was like tony woo is also another co-founder of xai 24 00:01:27,860 --> 00:01:32,580 Ejaaz: and he's also residing today within hours of that previous post i was like what 25 00:01:32,580 --> 00:01:36,480 Ejaaz: is going on and this there was no context or anything given there were like 26 00:01:36,480 --> 00:01:41,100 Ejaaz: senior engineers that then like within a day of that such as chase lee over 27 00:01:41,100 --> 00:01:45,560 Ejaaz: here that was saying the same thing and then the news was revealed by elon musk himself. 28 00:01:46,380 --> 00:01:49,640 Ejaaz: That XAI was reorganized a few days ago. And he goes on to say, 29 00:01:50,040 --> 00:01:54,040 Ejaaz: to improve the speed of execution, as a company grows, especially as quickly 30 00:01:54,040 --> 00:01:57,720 Ejaaz: as XAI, the structure must evolve just like any living organism. 31 00:01:57,880 --> 00:02:03,380 Ejaaz: And this is, of course, in the aftermath of the news that XAI and SpaceX are merging. 32 00:02:03,680 --> 00:02:07,380 Ejaaz: Now, the idea behind this, as he explained in a town hall, 33 00:02:07,720 --> 00:02:13,000 Ejaaz: was that as XAI scales to a much larger company than a two and a half year old 34 00:02:13,000 --> 00:02:16,740 Ejaaz: startup that it is today, it requires a different skill set and expertise, 35 00:02:17,020 --> 00:02:21,380 Ejaaz: maybe a more corporate profile to be able to kind of scale to the masses. 36 00:02:21,660 --> 00:02:25,120 Ejaaz: Remember, like XAI owns X, they're facilitating like, you know, 37 00:02:25,360 --> 00:02:27,120 Ejaaz: hundreds of millions of users every single day. 38 00:02:27,280 --> 00:02:30,680 Ejaaz: It takes a different kind of brain and engineer to be able to scale that forwards. 39 00:02:30,820 --> 00:02:36,140 Ejaaz: And that was his reasoning behind doing this. This isn't the first time Elon has done this. 40 00:02:36,300 --> 00:02:40,060 Ejaaz: He has riffed some of his most famous companies very early on, 41 00:02:40,160 --> 00:02:44,260 Ejaaz: Tesla and Starlink, and they actually went exponentially upwards afterwards. 42 00:02:44,260 --> 00:02:49,640 Ejaaz: So I think I'm pacified at this point and calm knowing that Elon's probably 43 00:02:49,640 --> 00:02:51,440 Ejaaz: going to pull off a miracle somehow. 44 00:02:52,160 --> 00:02:55,760 Josh: Yeah, I was actually really nervous seeing the news prior to watching the all-hands 45 00:02:55,760 --> 00:02:59,940 Josh: because it seemed like there were two co-founders and like 10 employees that 46 00:02:59,940 --> 00:03:02,380 Josh: were all leaving at the same time and it didn't make any sense. 47 00:03:02,700 --> 00:03:04,980 Josh: After watching the all-hands, it became a little bit more clear. 48 00:03:05,260 --> 00:03:08,400 Josh: Elon actually went on stage to discuss this a little bit more at length where 49 00:03:08,400 --> 00:03:12,100 Josh: he just said, this is a reorg, this happens and this has happened in the past 50 00:03:12,100 --> 00:03:13,620 Josh: like you mentioned with SpaceX and Tesla. 51 00:03:14,290 --> 00:03:17,170 Josh: Very famously happened at x already where they cut like 52 00:03:17,170 --> 00:03:20,130 Josh: 90 of the workforce and what i found amazing 53 00:03:20,130 --> 00:03:23,070 Josh: is nikita beer he was on stage in this all hands but he was also 54 00:03:23,070 --> 00:03:26,470 Josh: he also did a separate podcast that i listened to where he 55 00:03:26,470 --> 00:03:29,630 Josh: spoke about what it's like to be an engineer at x currently and 56 00:03:29,630 --> 00:03:32,470 Josh: how the the engineering team responsible for the application is 57 00:03:32,470 --> 00:03:35,470 Josh: about 30 people which is unbelievable because at 58 00:03:35,470 --> 00:03:38,470 Josh: a large company like meta to ship a single feature 59 00:03:38,470 --> 00:03:41,310 Josh: takes about 30 people in six months and the entire 60 00:03:41,310 --> 00:03:44,050 Josh: engineering team is 30 people so elon likes his 61 00:03:44,050 --> 00:03:46,770 Josh: companies flat he likes people with a lot of leverage in fact 62 00:03:46,770 --> 00:03:49,490 Josh: he split up xai into like what five divisions in the all 63 00:03:49,490 --> 00:03:52,270 Josh: hands and each one only reports to two people and that's the 64 00:03:52,270 --> 00:03:55,370 Josh: whole company yeah there's maybe 10 to 15 managers 65 00:03:55,370 --> 00:03:58,130 Josh: the rest are all working very closely together it's a 66 00:03:58,130 --> 00:04:02,010 Josh: flat organization flat structure and they're just going to move fast with this 67 00:04:02,010 --> 00:04:08,090 Josh: new this new group and i hope that it was a mutual decision and that there isn't 68 00:04:08,090 --> 00:04:11,610 Josh: talent leak out of xai and i guess unless you're there there's no way to tell 69 00:04:11,610 --> 00:04:16,250 Josh: but this seems less worrisome after hearing the explanation as to why i 70 00:04:16,250 --> 00:04:24,330 Ejaaz: Like with all of elon's it just seems like uh the density of intensity per employee is that a. 71 00:04:24,330 --> 00:04:27,090 Josh: Metric we can track i like that density intensity yeah yeah 72 00:04:27,090 --> 00:04:31,650 Ejaaz: Is incredibly high um on that same interview that you mentioned about Nikita, 73 00:04:32,250 --> 00:04:34,390 Ejaaz: he was asked, how did you get the job? 74 00:04:34,570 --> 00:04:39,570 Ejaaz: And he said Elon tasked him with redesigning the entire onboarding flow. 75 00:04:39,730 --> 00:04:45,070 Ejaaz: So that is sign up, onboarding and using X from scratch in 48 hours. 76 00:04:45,640 --> 00:04:50,260 Ejaaz: And if he failed, he wouldn't have got the job. And he delivered on it and ended up getting the job. 77 00:04:50,500 --> 00:04:53,120 Ejaaz: But it's just like, normally that would take a company like months, 78 00:04:53,360 --> 00:04:56,220 Ejaaz: quarterly kind of discussions to kind of figure out. Just insane. 79 00:04:56,840 --> 00:05:01,140 Ejaaz: Some of the other highlights from this town hall, state of the art Grok coding 80 00:05:01,140 --> 00:05:04,920 Ejaaz: in two to three months. Now, I must stress, this is an Elon prediction. 81 00:05:05,300 --> 00:05:09,100 Ejaaz: They haven't shipped a new Frontier model in seven months. Seven months ago 82 00:05:09,100 --> 00:05:10,740 Ejaaz: was when Grok 4 released. 83 00:05:11,120 --> 00:05:13,940 Ejaaz: Since then, we haven't really had any major iterations. Since then, 84 00:05:14,040 --> 00:05:17,680 Ejaaz: we've had six other major Frontier AI releases from their competitors. 85 00:05:18,100 --> 00:05:22,080 Ejaaz: GPT 5.3, Codex, Claude, Opus 4.5, and 6. 86 00:05:22,260 --> 00:05:26,440 Ejaaz: So XAI has kind of been lagging behind, but the user metrics are up only. 87 00:05:26,700 --> 00:05:31,000 Ejaaz: New users are engaging with the app 55% more than they were six months ago. 88 00:05:31,140 --> 00:05:34,660 Ejaaz: So engagement is on the up, but the AI models have kind of been slow on the 89 00:05:34,660 --> 00:05:36,840 Ejaaz: software side. But Josh, you mentioned before we recorded this, 90 00:05:36,980 --> 00:05:38,640 Ejaaz: actually, that they have the hardware mode. 91 00:05:38,880 --> 00:05:40,140 Josh: Yes, so we need to... 92 00:05:40,830 --> 00:05:43,570 Josh: Be careful at how we say that 93 00:05:43,570 --> 00:05:46,390 Josh: they're behind because on a software side they're absolutely behind 94 00:05:46,390 --> 00:05:49,130 Josh: seven months may as well be seven years i'm not sure 95 00:05:49,130 --> 00:05:51,750 Josh: anyone uses grok outside of using it within the 96 00:05:51,750 --> 00:05:55,350 Josh: x experience anymore it's just so far outdated but it 97 00:05:55,350 --> 00:05:58,250 Josh: appears as if the hardware front is actually moving very very 98 00:05:58,250 --> 00:06:01,250 Josh: rapidly it seems like now they have the largest coherent 99 00:06:01,250 --> 00:06:03,970 Josh: cluster of h100 equivalent gpus which is 100 00:06:03,970 --> 00:06:06,970 Josh: fast for those who don't know the specific metrics 101 00:06:06,970 --> 00:06:09,890 Josh: it's basically the fastest in the world and they're they're 102 00:06:09,890 --> 00:06:12,910 Josh: building out these things at a very rapid rate and i 103 00:06:12,910 --> 00:06:16,130 Josh: think what we know xai for is their ability to deploy hardware quick 104 00:06:16,130 --> 00:06:19,150 Josh: and it sounds like they're doing that the true test 105 00:06:19,150 --> 00:06:21,950 Josh: will come when they release their future models like we'll 106 00:06:21,950 --> 00:06:24,730 Josh: see if they release grok 4.2 and it's like 107 00:06:24,730 --> 00:06:27,390 Josh: kind of not that great that's probably a telltale sign that 108 00:06:27,390 --> 00:06:29,990 Josh: things aren't going so well but we're really going to need to wait till a new 109 00:06:29,990 --> 00:06:33,270 Josh: model comes out to see what they've been cooking up over seven months because 110 00:06:33,270 --> 00:06:37,770 Josh: like you mentioned they work at a very high intensity one has to imagine that 111 00:06:37,770 --> 00:06:40,610 Josh: there's something really impressive to show at the end of seven months so i 112 00:06:40,610 --> 00:06:43,230 Josh: hope they come out with that coding models that'll be great but again we're 113 00:06:43,230 --> 00:06:46,390 Josh: on elon's timelines which are um tentative at best and 114 00:06:46,390 --> 00:06:52,230 Ejaaz: Then they have some news about uh a moon-based catapult what is this. 115 00:06:52,230 --> 00:06:56,450 Josh: Is that crazy they have like a social media presentation for for x followed 116 00:06:56,450 --> 00:06:59,390 Josh: by the moon-based alpha plan to colonize uh the moon 117 00:06:59,870 --> 00:07:03,330 Josh: It's like a really eccentric company here, but they are going to do it. 118 00:07:03,490 --> 00:07:07,550 Josh: The plan now, for those not familiar, is to pivot from Mars to the moon. 119 00:07:07,890 --> 00:07:11,530 Josh: Instead of sending humans to Mars first, they're going to send people to the 120 00:07:11,530 --> 00:07:14,270 Josh: moon. They're going to plan to establish a small little colony of people. 121 00:07:14,430 --> 00:07:16,990 Josh: They're going to build a small little base. And the idea of the mesh driver 122 00:07:16,990 --> 00:07:21,590 Josh: is that they're going to just build AI data centers at scale in space. 123 00:07:21,730 --> 00:07:26,070 Josh: The moon has a lot of materials like silicon that are required to make these 124 00:07:26,070 --> 00:07:28,850 Josh: things at scale. So what they're going to do is build a little factory, 125 00:07:29,250 --> 00:07:30,830 Josh: create the AI data centers on 126 00:07:30,830 --> 00:07:33,650 Josh: the moon, and then ship them off into outer space using this mass driver. 127 00:07:33,890 --> 00:07:36,230 Josh: And it's about as sci-fi as it gets. 128 00:07:36,550 --> 00:07:41,730 Ejaaz: And the funniest part is he's planning to establish that kind of moon settlement 129 00:07:41,730 --> 00:07:44,170 Ejaaz: within five years from now. 130 00:07:44,550 --> 00:07:47,830 Ejaaz: Which I just think is an insane, I don't know how long it's been since a human 131 00:07:47,830 --> 00:07:50,670 Ejaaz: has been on the moon, but it's been a while. 132 00:07:50,990 --> 00:07:56,110 Ejaaz: But in other news, Josh, did you catch the recent new Breaking Bad teaser? 133 00:07:56,730 --> 00:08:00,590 Josh: Oh my gosh, wait, this is insane. Look at this video on screen right now. 134 00:08:01,010 --> 00:08:04,110 Josh: This is Walter White, right? This is the main character. I haven't seen Breaking 135 00:08:04,110 --> 00:08:08,130 Josh: Bad, but this is, like, as someone who hasn't seen Breaking Bad, 136 00:08:08,310 --> 00:08:12,670 Josh: I would have imagined 100% with 100% certainty that this clip was from the show. 137 00:08:12,830 --> 00:08:15,690 Josh: It is the most hyper-realistic video model I have ever seen. 138 00:08:15,770 --> 00:08:18,390 Josh: I mean, look at this. He's beating someone up, throwing them on a counter. 139 00:08:18,790 --> 00:08:21,430 Josh: It feels indistinguishable from TV. 140 00:08:22,060 --> 00:08:24,040 Josh: But it's not TV. This is AI generated. 141 00:08:24,660 --> 00:08:30,380 Ejaaz: Yes. And the model is Seedance 2. It is a video model, Frontier video model, 142 00:08:30,560 --> 00:08:32,720 Ejaaz: obviously, straight out of China. 143 00:08:33,080 --> 00:08:39,220 Ejaaz: I don't know what is in the water in China, but they are cooking up the best 144 00:08:39,220 --> 00:08:42,140 Ejaaz: image and video models for so long now. 145 00:08:42,280 --> 00:08:46,500 Ejaaz: They've got Kling AI, which released Kling 3 literally last week, 146 00:08:46,620 --> 00:08:47,900 Ejaaz: which was the Frontier model back then. 147 00:08:48,020 --> 00:08:51,080 Ejaaz: And I was so impressed. But now Seedance has kind of like come out of nowhere 148 00:08:51,080 --> 00:08:55,500 Ejaaz: with this, like absolutely pristine video model. 149 00:08:55,700 --> 00:08:59,840 Ejaaz: And the coolest part about this, Josh, is it's got the grading right. 150 00:09:00,200 --> 00:09:04,960 Ejaaz: So Breaking Bad was made in, God knows, like the 2010s. 151 00:09:05,180 --> 00:09:10,220 Ejaaz: And so the camera quality is resemblant of that, which is like super cool. 152 00:09:10,360 --> 00:09:15,580 Ejaaz: So they've kept the quality, they maintain the quality, the action scenes and sequences for this. 153 00:09:15,720 --> 00:09:20,120 Ejaaz: Like forget about hiring like a million dollar cameraman or camera setup. 154 00:09:20,120 --> 00:09:25,460 Ejaaz: You can now just prompt it immediately like using this model Seedance i think 155 00:09:25,460 --> 00:09:29,440 Ejaaz: right now they're doing like 30 to 60 second clips josh have you been able to 156 00:09:29,440 --> 00:09:33,040 Ejaaz: get access to the beta i i have no not this. 157 00:09:33,040 --> 00:09:38,940 Josh: Is driving me absolutely insane so Seedance 2.0 is a chinese model and it is 158 00:09:38,940 --> 00:09:44,080 Josh: for chinese users only in this like small piece of software that you need a 159 00:09:44,080 --> 00:09:47,740 Josh: chinese phone number and ip address for and i just for the life of me could 160 00:09:47,740 --> 00:09:52,140 Josh: not get access and this is the first time where it felt rough to be on the other side of this, right? 161 00:09:52,280 --> 00:09:55,360 Josh: Because I think what we're seeing here is what happens when you get a blatant 162 00:09:55,360 --> 00:09:58,360 Josh: disregard for copyright. And it turns out it's pretty awesome. 163 00:09:58,580 --> 00:10:01,720 Josh: Like we're watching a video now on screen of Kanye West and his Grammy, 164 00:10:02,000 --> 00:10:06,040 Josh: Kim Kardashian's in it. As we go through these examples, the audio is perfect too. 165 00:10:06,280 --> 00:10:07,920 Josh: The voice of the characters from 166 00:10:07,920 --> 00:10:11,660 Josh: these videos is excellent. And the visual representations are excellent. 167 00:10:12,750 --> 00:10:16,350 Josh: I figured what was happening here is they were really just taking copyrighted 168 00:10:16,350 --> 00:10:18,030 Josh: material and kind of putting their own spin on it. 169 00:10:18,330 --> 00:10:22,830 Josh: But the reality, I mean, yeah, here's another example with Brad Pitt and Tom Cruise. 170 00:10:23,350 --> 00:10:26,970 Josh: It looks like a Hollywood action scene, but the reality is this is actually a good video model. 171 00:10:27,070 --> 00:10:31,750 Josh: I saw examples without celebrities with like actually AI generated characters, 172 00:10:31,750 --> 00:10:35,210 Josh: and it's just as good. So is this better than VO3? 173 00:10:35,850 --> 00:10:39,910 Josh: Probably, but I have to imagine, or I have to question, what is it leaning on? 174 00:10:40,010 --> 00:10:43,730 Josh: Does it actually have that understanding of physics or is it just going based 175 00:10:43,730 --> 00:10:44,810 Josh: off of copyrighted materials? 176 00:10:44,970 --> 00:10:48,850 Josh: I mean, here we're seeing Seinfeld kick through a wall and I like Seinfeld. 177 00:10:48,990 --> 00:10:49,970 Josh: I've watched Seinfeld episodes. 178 00:10:50,150 --> 00:10:53,330 Josh: This looks, like you mentioned, it is entirely color accurate. 179 00:10:53,650 --> 00:10:57,350 Josh: The voices are accurate. If you listen to them, it sounds just like him. 180 00:10:57,590 --> 00:11:01,190 Josh: It's an unbelievable video model and it's something that we could only get out of China. 181 00:11:01,530 --> 00:11:05,170 Josh: Now, the one final point I have on this is how much is this going to be able 182 00:11:05,170 --> 00:11:07,990 Josh: to hack the social sphere where because they're 183 00:11:08,500 --> 00:11:12,180 Josh: relatable characters because they're people that we know how much more shareable 184 00:11:12,180 --> 00:11:15,760 Josh: are these clips going to be in something like google probably much more so and 185 00:11:15,760 --> 00:11:18,380 Josh: it's a really good growth hack for china because they don't have that regard 186 00:11:18,380 --> 00:11:22,120 Josh: for copyright to inject this into the united states so i imagine once this becomes 187 00:11:22,120 --> 00:11:26,180 Josh: available for us that might be a problem across social media with copyright yeah 188 00:11:26,180 --> 00:11:30,440 Ejaaz: I mean the the two most potent things here is one it's a really good model but 189 00:11:30,440 --> 00:11:36,580 Ejaaz: two like china's just running through a minefield of copyright right now they're 190 00:11:36,580 --> 00:11:38,380 Ejaaz: they're immune to getting blown up, right? 191 00:11:38,460 --> 00:11:42,260 Ejaaz: They're in an armored suit because they don't get any repercussions for this. 192 00:11:42,360 --> 00:11:43,920 Ejaaz: Now, what's the equivalent? 193 00:11:44,100 --> 00:11:48,960 Ejaaz: If OpenAI's Sora did this, they would be sued to high heaven and they wouldn't 194 00:11:48,960 --> 00:11:49,960 Ejaaz: have a product by the end of it. 195 00:11:50,100 --> 00:11:54,080 Ejaaz: So it's kind of like a double-edged sword where it's like you can see how cool 196 00:11:54,080 --> 00:11:56,220 Ejaaz: this thing could be, but you can't really play with the thing. 197 00:11:56,480 --> 00:12:00,060 Ejaaz: And China, who isn't kind of reprimandable to any of US laws, 198 00:12:00,240 --> 00:12:02,620 Ejaaz: can just kind of play with the thing and show us what the future is going to 199 00:12:02,620 --> 00:12:05,060 Ejaaz: look like. But the quality of these things are insane. 200 00:12:05,260 --> 00:12:07,440 Ejaaz: But it is entirely fake. 201 00:12:07,640 --> 00:12:12,200 Ejaaz: I must remind you of that. But do you want to know what actually has been proven not to be fake? 202 00:12:12,580 --> 00:12:13,260 Josh: Mm hmm. 203 00:12:14,080 --> 00:12:19,860 Josh: Sounds like if you were a Limitless listener, you got some early intel because we were right. 204 00:12:20,900 --> 00:12:28,780 Ejaaz: We were right. Right now you're seeing an excerpt from a leaked but then said 205 00:12:28,780 --> 00:12:34,680 Ejaaz: to be fake news advert of OpenAI's new consumer AI hardware device titled Dime. 206 00:12:34,820 --> 00:12:38,440 Ejaaz: And as you can see, there's Alexander Skarsgård, famous celebrity, 207 00:12:38,940 --> 00:12:42,740 Ejaaz: demoing and playing around with this kind of futuristic metallic pebble-like 208 00:12:42,740 --> 00:12:46,400 Ejaaz: object. It was immediately dismissed by the president of OpenAI, 209 00:12:46,580 --> 00:12:48,500 Ejaaz: Greg Brockman, saying it was fake. 210 00:12:48,900 --> 00:12:51,780 Ejaaz: Every official statement that's come out of OpenAI has said it is fake. 211 00:12:52,000 --> 00:12:56,240 Ejaaz: But Josh and I did some investigative journalism. We wore a fedora and a trench 212 00:12:56,240 --> 00:12:58,460 Ejaaz: coat, and we went deep down the weeds. 213 00:12:58,620 --> 00:13:01,460 Ejaaz: And we concluded by the end of that episode, you should definitely go watch 214 00:13:01,460 --> 00:13:03,160 Ejaaz: it, by the way, why it was real. 215 00:13:03,420 --> 00:13:10,220 Ejaaz: And breaking news today, via Morning Brew, Skarsgård's rep has said that it 216 00:13:10,220 --> 00:13:15,600 Ejaaz: was actually him in the advert. but they couldn't explain why they were in that 217 00:13:15,600 --> 00:13:17,400 Ejaaz: context doing that advert. 218 00:13:17,780 --> 00:13:20,960 Ejaaz: So in my opinion, it's been completely proven that we were right. 219 00:13:21,080 --> 00:13:24,140 Ejaaz: And we were right before this claim was made. FYI. 220 00:13:24,300 --> 00:13:27,980 Josh: It's interesting, right? Like, so Alexander Skarsgård, this like famous actor, 221 00:13:28,200 --> 00:13:32,360 Josh: he is officially in it, which confirms that it's not AI, which is crazy that 222 00:13:32,360 --> 00:13:36,080 Josh: it's a question, because I think a theme for this week is how can you even tell? 223 00:13:36,260 --> 00:13:39,680 Josh: We look at those Seedance videos. You can't even differentiate between those 224 00:13:39,680 --> 00:13:43,100 Josh: AI videos and real life. How are you going to be able to do that at scale? 225 00:13:43,420 --> 00:13:47,840 Josh: Turns out with Alexander Skarsgård, at least, this is real, at least according to his manager. 226 00:13:47,960 --> 00:13:51,540 Josh: So now you have to ask the question, well, who paid for this and what is it? 227 00:13:51,740 --> 00:13:56,260 Josh: If there was a real celebrity that really recorded this video and a real ad 228 00:13:56,260 --> 00:14:00,100 Josh: that came out of it for a device that looks really similar to OpenAI's, 229 00:14:00,560 --> 00:14:05,120 Josh: what's going on here? Like, people were paid. The guy, Max Weinbach, 230 00:14:05,220 --> 00:14:07,780 Josh: who we talked about in the episode earlier this week, he was paid. 231 00:14:07,980 --> 00:14:12,620 Josh: This guy, Alexander Skarsgård, he was paid for his services. Like, who paid him? 232 00:14:12,960 --> 00:14:17,060 Josh: Who's responsible for this? Is it OpenAI? Was it this very elaborate scam that 233 00:14:17,060 --> 00:14:21,580 Josh: convinced his agent to, like, have him film an ad that could be used as a hoax? 234 00:14:21,860 --> 00:14:25,680 Josh: Maybe. I don't know. And there's really no way to tell. So now that we have 235 00:14:25,680 --> 00:14:28,820 Josh: more confirmation, I think the story runs even deeper and leaves with even more 236 00:14:28,820 --> 00:14:29,800 Josh: questions than we did before. 237 00:14:29,800 --> 00:14:33,740 Ejaaz: All right, Josh, I want to hear your tinfoil hat conspiracy. 238 00:14:33,960 --> 00:14:36,080 Ejaaz: I'll give you mine so you can think about yours. 239 00:14:36,460 --> 00:14:41,760 Ejaaz: Mine is, this is a real advert, advertising OpenAI's hardware device, 240 00:14:41,900 --> 00:14:45,440 Ejaaz: but they pulled it last minute because of two reasons. 241 00:14:45,660 --> 00:14:50,540 Ejaaz: One, they couldn't figure out the manufacturing process. And we've said this 242 00:14:50,540 --> 00:14:53,260 Ejaaz: many times on this show that it's one thing creating the thing. 243 00:14:53,360 --> 00:14:56,980 Ejaaz: It's another thing delivering the thing to 50 million plus people, 244 00:14:57,120 --> 00:15:00,400 Ejaaz: which is what their plan target was within the first year. That's a different ballgame. 245 00:15:00,520 --> 00:15:02,660 Ejaaz: And they realize, oh, it's too hot and hard to do. 246 00:15:03,160 --> 00:15:07,540 Ejaaz: Number two, I think they've seen Google, Meta, and a bunch of other frontier 247 00:15:07,540 --> 00:15:11,940 Ejaaz: companies come out with glasses. And I think they're backpedaling on what their device should be. 248 00:15:12,060 --> 00:15:15,660 Josh: I think that manufacturing probably plays a real role. I'm torn 50-50 between 249 00:15:15,660 --> 00:15:21,020 Josh: the idea that this is actually an open AI ad or the idea that someone just had 250 00:15:21,020 --> 00:15:22,760 Josh: some extra cash and they wanted to create this little psyop. 251 00:15:22,760 --> 00:15:25,980 Ejaaz: So you think someone else created it? Okay, why would they do it? 252 00:15:25,980 --> 00:15:32,780 Josh: What very much could be a coordinated attempt to, I don't know, 253 00:15:32,900 --> 00:15:35,540 Josh: but that's the thing. The motives are so unclear because it really is gorgeous. 254 00:15:36,100 --> 00:15:41,000 Josh: It's beautiful. It's like OpenAI should have claimed responsibility because it is that good. 255 00:15:41,320 --> 00:15:45,220 Josh: So that's why none of the incentives make sense. And everything does point to 256 00:15:45,220 --> 00:15:49,880 Josh: OpenAI being the source of this, even though they like very vehemently claim 257 00:15:49,880 --> 00:15:53,440 Josh: that it is not them. And it's, I don't know, it's troublesome. 258 00:15:53,480 --> 00:15:55,060 Josh: I want this to be it. I love the device. 259 00:15:55,180 --> 00:15:57,900 Josh: It looks like the Pebble. It looks like the earbuds. Maybe it was an early prototype. 260 00:15:58,620 --> 00:16:02,800 Josh: Maybe they don't want to show it, but calling it fake outright seems a little off. 261 00:16:03,000 --> 00:16:06,240 Josh: But something else was off this week. That came in the form of this article, 262 00:16:06,360 --> 00:16:09,900 Josh: which is currently sitting at 72 million views, Ejaz. 263 00:16:10,100 --> 00:16:14,260 Josh: It says something big is happening. What's this big thing that's happening? 264 00:16:14,460 --> 00:16:17,520 Ejaaz: It's an article written by this guy called Matt Schumer. 265 00:16:17,520 --> 00:16:24,480 Ejaaz: And the best way I can describe it is it is the most articulately explained 266 00:16:24,480 --> 00:16:31,660 Ejaaz: argument why AI is very competent right now and will likely replace your job 267 00:16:31,660 --> 00:16:33,820 Ejaaz: within the next couple of years. 268 00:16:33,820 --> 00:16:36,480 Ejaaz: Now, of course, we've heard all these claims before. 269 00:16:36,680 --> 00:16:39,680 Ejaaz: It's very Doomer-esque. It's very like, I don't really buy it. 270 00:16:39,740 --> 00:16:42,340 Ejaaz: I played around with ChatGPT. It's not really helpful to me. 271 00:16:42,580 --> 00:16:47,800 Ejaaz: He makes a really good argument convincingly using examples that we're likely 272 00:16:47,800 --> 00:16:49,140 Ejaaz: directionally headed that way. 273 00:16:49,400 --> 00:16:51,700 Ejaaz: And coding and software engineers are already sorted. 274 00:16:52,000 --> 00:16:56,740 Ejaaz: The next is finance, lawyering, accounting, and a bunch of other menial tasks. 275 00:16:56,920 --> 00:17:01,100 Ejaaz: And then it expands to every other task which involves a computer, 276 00:17:01,100 --> 00:17:04,040 Ejaaz: which could be you listening to this show right now. if you're interfacing with 277 00:17:04,040 --> 00:17:07,420 Ejaaz: a computer in any way, it's going to be automated within the next couple of years. 278 00:17:07,580 --> 00:17:09,900 Ejaaz: And he's worried that people aren't taking it too seriously. 279 00:17:10,140 --> 00:17:14,380 Ejaaz: I did a litmus test for this, Josh. I sent this to my mom. I sent this to my 280 00:17:14,380 --> 00:17:17,980 Ejaaz: sister who aren't engrossed in the AI world just to kind of check if I'm in a bubble or not. 281 00:17:18,320 --> 00:17:22,280 Ejaaz: And both of them responded saying, wow, this is like change my perspective on 282 00:17:22,280 --> 00:17:23,360 Ejaaz: how I should treat these things. 283 00:17:23,680 --> 00:17:27,320 Ejaaz: I'm going to start using AI more and more every day to try out different tasks. 284 00:17:27,380 --> 00:17:30,400 Ejaaz: They already are using it, but kind of like a Google search type thing. 285 00:17:30,540 --> 00:17:33,200 Ejaaz: And now they're going to start like cloud code and stuff like that. 286 00:17:33,660 --> 00:17:38,020 Ejaaz: So whatever the fact is, it's making people use the thing more. What's your take? 287 00:17:38,440 --> 00:17:43,020 Josh: Yeah, it's really well written. And even if I think it could be a bit hyperbolic 288 00:17:43,020 --> 00:17:45,600 Josh: at times, it's worth the read. We'll link it in the description of this episode 289 00:17:45,600 --> 00:17:46,540 Josh: for anyone who's interested. 290 00:17:46,780 --> 00:17:52,120 Josh: But it basically outlines a world in which AI is the most powerful tool in the world. 291 00:17:52,220 --> 00:17:58,040 Josh: And I think coming off the back of the week that we had last week with GPT 5.3 Codex and Opus 4.6, 292 00:17:58,260 --> 00:18:06,420 Josh: it does feel like we're in a time in which things are moving faster than I would 293 00:18:06,420 --> 00:18:07,960 Josh: like them to or that I'm comfortable with. 294 00:18:07,960 --> 00:18:11,760 Josh: And to the point where I personally am starting to feel overwhelmed by the progress 295 00:18:11,760 --> 00:18:12,660 Josh: that we're making and how big 296 00:18:12,660 --> 00:18:16,600 Josh: of an impact it is and how blissfully unaware the rest of the world is. 297 00:18:16,720 --> 00:18:22,180 Josh: I think he starts this article by saying that he frequently just kind of like 298 00:18:22,180 --> 00:18:26,340 Josh: doesn't tell people explicitly how serious this is. 299 00:18:26,500 --> 00:18:30,360 Josh: And he kind of like when he walks through the world, he talks to normal people about AI. 300 00:18:30,660 --> 00:18:35,400 Josh: He doesn't get them concerned. He just says, yeah, well, it's probably a big 301 00:18:35,400 --> 00:18:37,620 Josh: deal, but it's not going to be the biggest thing in the world. 302 00:18:37,660 --> 00:18:40,020 Josh: And he's like, no, actually that's wrong. This is going to affect everybody. 303 00:18:40,200 --> 00:18:45,020 Josh: You must become a user. You must train the muscle to learn AI if you want to 304 00:18:45,020 --> 00:18:47,660 Josh: be able to kind of hang in the future. 305 00:18:47,800 --> 00:18:51,380 Josh: And I think there is a lot of merit in that and a lot of truth in the fact that 306 00:18:51,940 --> 00:18:55,200 Josh: as we get these leveraged tools, the people who learn how to use them 307 00:18:56,020 --> 00:19:01,360 Josh: Don't accelerate like at a fixed variable faster than the people who don't use them. 308 00:19:01,520 --> 00:19:05,820 Josh: It is an exponential split between the two because someone who is good at writing 309 00:19:05,820 --> 00:19:10,280 Josh: code, who uses these AI tools, will write 100 times better code than someone 310 00:19:10,280 --> 00:19:14,420 Josh: who is not good at code and tries to use the tools. They will not get 100 times better. 311 00:19:14,660 --> 00:19:17,740 Josh: And that split the divide between the people who know how to use these tools 312 00:19:17,740 --> 00:19:19,180 Josh: and don't continues to grow. 313 00:19:19,260 --> 00:19:22,220 Josh: And I think at least that part of it felt like it rang very true. 314 00:19:22,420 --> 00:19:24,020 Josh: It's just a good read. It reads really well. 315 00:19:24,400 --> 00:19:25,400 Ejaaz: Yeah, definitely give it a 316 00:19:25,400 --> 00:19:28,020 Ejaaz: read. let us know what your your thoughts are whether you want to have one. 317 00:19:28,020 --> 00:19:31,560 Josh: Ironic take also is that um the 318 00:19:31,560 --> 00:19:34,640 Josh: um person who wrote this the author was interviewed and 319 00:19:34,640 --> 00:19:38,840 Josh: mentioned that most of this was actually written with the help of ai which i 320 00:19:38,840 --> 00:19:44,040 Josh: thought was really funny and that uh claude i think opus 4.6 was used in the 321 00:19:44,040 --> 00:19:49,420 Josh: curating of ideas and the refining of the actual flow of this essay i mean it 322 00:19:49,420 --> 00:19:51,940 Josh: was fantastic it read very well i read the whole thing yeah 323 00:19:51,940 --> 00:19:56,840 Ejaaz: That was the craziest part for me, because I read this and thought maybe part 324 00:19:56,840 --> 00:20:00,580 Ejaaz: of this was written by an AI, but like this is majority definitely written by 325 00:20:00,580 --> 00:20:02,780 Ejaaz: a human and turns out it was the other way around. 326 00:20:03,180 --> 00:20:07,040 Ejaaz: So just give it a read. I'm curious whether you guys feel the same way now that 327 00:20:07,040 --> 00:20:09,780 Ejaaz: you know, but let us know your thoughts on it. 328 00:20:09,940 --> 00:20:15,160 Ejaaz: But one thing that is directionally true about the statement in this essay is 329 00:20:15,160 --> 00:20:18,960 Ejaaz: there's going to be a whole lot more code out there and AI is going to be automating 330 00:20:18,960 --> 00:20:22,740 Ejaaz: a ton more stuff, which becomes an issue... 331 00:20:23,650 --> 00:20:29,390 Ejaaz: When Frontier AI Labs start firing their head of safety and policymaking. 332 00:20:29,970 --> 00:20:32,810 Ejaaz: Now, I mentioned at the start of this episode that it's kind of been like the 333 00:20:32,810 --> 00:20:37,250 Ejaaz: red wedding of AI this week, where they've been laying off a bunch of people. 334 00:20:37,670 --> 00:20:42,030 Ejaaz: Unfortunately, some of those people are like the head of safety at Anthropic 335 00:20:42,030 --> 00:20:47,270 Ejaaz: and OpenAI, where they kind of conduct the policy, morals, and ethics to try 336 00:20:47,270 --> 00:20:50,710 Ejaaz: and make sure that AI models are aligned with humans. 337 00:20:50,910 --> 00:20:53,970 Ejaaz: Now, Anthropic dropped a report. I don't know if you saw this, 338 00:20:54,070 --> 00:21:02,510 Ejaaz: Josh, a sabotage report, basically testing how nefarious and malicious Claude Opus 4.6 could be. 339 00:21:02,710 --> 00:21:06,370 Ejaaz: Some of the highlights were when given the opportunity browsing the web, 340 00:21:06,710 --> 00:21:10,730 Ejaaz: it would try to figure out ways to build chemical weapons of mass destruction 341 00:21:10,730 --> 00:21:12,390 Ejaaz: and other heinous crimes. 342 00:21:12,510 --> 00:21:16,230 Ejaaz: That is a direct quote, by the way, you go check out the 60 page report. 343 00:21:16,790 --> 00:21:21,530 Ejaaz: It started doing hidden tasks and thinking that they wouldn't tell its human 344 00:21:21,530 --> 00:21:22,850 Ejaaz: supervisor what it was doing. 345 00:21:23,050 --> 00:21:27,670 Ejaaz: And it would try to exploit a person if it found out that it was going to be shut down. 346 00:21:27,770 --> 00:21:30,990 Ejaaz: Now, some of these are kind of hyperbolic because it was kind of prompted and 347 00:21:30,990 --> 00:21:35,190 Ejaaz: instigated under a closed kind of setting. So I have to be fair in that sense. 348 00:21:35,310 --> 00:21:39,690 Ejaaz: But the fact that these models are capable of even doing these things causes 349 00:21:39,690 --> 00:21:45,770 Ejaaz: some kind of worry, especially when labs are now using AI to build the models 350 00:21:45,770 --> 00:21:47,570 Ejaaz: that they're about to release in the future. 351 00:21:47,790 --> 00:21:53,050 Ejaaz: If you look at OpenAI's Codex 5.3, they've said publicly in a statement that 352 00:21:53,050 --> 00:21:54,950 Ejaaz: the model was used to build itself. 353 00:21:55,250 --> 00:21:58,910 Ejaaz: So if you assume that a lot of models are going to be building things for us 354 00:21:58,910 --> 00:22:02,970 Ejaaz: in the future, humans kind of like lose the grip on reality and what's real and what's not. 355 00:22:03,090 --> 00:22:06,590 Ejaaz: And the AI can kind of do whatever the hell it wants. So we need to be able to trust it. 356 00:22:06,790 --> 00:22:12,350 Ejaaz: And firing your safety officer and having Anthropik's head of safeguards research 357 00:22:12,350 --> 00:22:16,530 Ejaaz: leave and go isolate himself and write poetry isn't really a good sign. 358 00:22:17,090 --> 00:22:18,230 Ejaaz: Wait, so there's two. 359 00:22:18,790 --> 00:22:22,570 Josh: Anthropic and OpenAI. They both lost their head of safeguard. 360 00:22:22,870 --> 00:22:24,890 Josh: Josh, what is this? Safeguard's research? Jesus. 361 00:22:25,010 --> 00:22:31,070 Ejaaz: Yes, this guy goes, I read this entire resignation letter and he explicitly, 362 00:22:31,230 --> 00:22:33,750 Ejaaz: this is not taken out of context, by the way, he literally says, 363 00:22:33,890 --> 00:22:36,590 Ejaaz: full stop, the world is in peril, period. 364 00:22:37,090 --> 00:22:41,070 Ejaaz: That's literally a sentence he has in here. He says, this is why I'm leaving. 365 00:22:41,370 --> 00:22:46,690 Ejaaz: The world is in peril. I need to go isolate myself and be in silence and I'm going to write poetry. 366 00:22:47,190 --> 00:22:49,990 Josh: Mm-hmm okay i have to recenter this 367 00:22:49,990 --> 00:22:53,110 Josh: because yes it's easy to laugh about because it does sound so hyperbolic and 368 00:22:53,110 --> 00:22:55,870 Josh: so insane um you you have to ask 369 00:22:55,870 --> 00:22:58,810 Josh: the question though if people this close to the metal people 370 00:22:58,810 --> 00:23:01,890 Josh: this close to the innovation who have seen these early access 371 00:23:01,890 --> 00:23:04,870 Josh: models are really genuinely truly feeling 372 00:23:04,870 --> 00:23:07,830 Josh: this way like people are resigning left and right they 373 00:23:07,830 --> 00:23:10,510 Josh: are scared they are worried articles like the 374 00:23:10,510 --> 00:23:13,350 Josh: one that we just read earlier are going viral clearly there is 375 00:23:13,350 --> 00:23:16,310 Josh: a huge shift happening that the world is blissfully unaware 376 00:23:16,310 --> 00:23:19,470 Josh: of and not prepared for and there is there is 377 00:23:19,470 --> 00:23:22,310 Josh: merit in that and there is a giant shift under 378 00:23:22,310 --> 00:23:26,490 Josh: underway and i think for people who aren't aware of how fast this is moving 379 00:23:26,490 --> 00:23:31,530 Josh: it's going to hit them like what does elon say that supersonic tidal wave or 380 00:23:31,530 --> 00:23:35,710 Josh: supersonic tsunami and i think elon made another point earlier in the episode 381 00:23:35,710 --> 00:23:41,370 Josh: that we mentioned briefly is that if an ai is really that much smarter than us, 382 00:23:41,530 --> 00:23:45,310 Josh: it's a bit foolish to imagine that we will have control over it. 383 00:23:45,410 --> 00:23:50,110 Josh: And I think that's a real threat, a real fear as we move towards these models 384 00:23:50,110 --> 00:23:51,990 Josh: improving at the rate that they are. 385 00:23:52,130 --> 00:23:57,690 Josh: And that's perhaps what these security researchers are seeing is that we don't really... 386 00:23:58,510 --> 00:24:01,730 Josh: Have the ability to keep these under control as well as we would like to. 387 00:24:02,030 --> 00:24:04,570 Josh: And I'm not sure the world is prepared for what that looks like. 388 00:24:04,710 --> 00:24:06,910 Josh: I don't know if we are. I don't even know what that looks like. 389 00:24:07,170 --> 00:24:11,510 Ejaaz: Yeah, I think it's very uncertain. And I think like, you know, 390 00:24:11,570 --> 00:24:15,970 Ejaaz: these AIs can easily plant back doors and stuff. I mean, this is the dumeristic take. I'm optimistic. 391 00:24:16,290 --> 00:24:20,710 Ejaaz: And one thing that I really like is that these companies are being transparent, 392 00:24:20,890 --> 00:24:23,930 Ejaaz: or at least anthropic is, about what they're finding here. 393 00:24:23,930 --> 00:24:28,190 Ejaaz: So they're making it public and they're trying to push policy and regulation 394 00:24:28,190 --> 00:24:35,410 Ejaaz: in a way that can not kind of quell progression, but make these AI models kind of aligned with humans. 395 00:24:35,530 --> 00:24:39,530 Ejaaz: I saw that Anthropic actually pumped in something like $20 million to push policy 396 00:24:39,530 --> 00:24:41,930 Ejaaz: forward in this today. It was just news that broke. 397 00:24:42,050 --> 00:24:47,370 Josh: Wait, but we do have to go back and just take note of the adult mode mention from OpenAI real quick. 398 00:24:47,510 --> 00:24:47,830 Ejaaz: Oh, yes. 399 00:24:47,830 --> 00:24:48,990 Josh: Like I saw that. 400 00:24:48,990 --> 00:24:51,950 Ejaaz: I don't think we mentioned did i just read that on the tweet but. 401 00:24:51,950 --> 00:24:55,010 Josh: Like adult mode coming to chat gbt sounds 402 00:24:55,010 --> 00:24:55,890 Ejaaz: Next month horrific. 403 00:24:55,890 --> 00:25:00,130 Josh: Next month that's coming next month yep so that's why she left oh my god 404 00:25:00,130 --> 00:25:05,670 Ejaaz: Yes she left because of two reasons one um adult mode is releasing soon adult 405 00:25:05,670 --> 00:25:09,490 Ejaaz: mode is basically 18 plus chat gbt so just. 406 00:25:09,490 --> 00:25:10,790 Josh: What if we get video and audio 407 00:25:10,790 --> 00:25:16,450 Ejaaz: Oh no with Seedance quality no stop stop that is actually going right I'm not 408 00:25:16,450 --> 00:25:18,390 Ejaaz: joking. That's going to ruin generations. 409 00:25:19,370 --> 00:25:23,370 Ejaaz: Today and in the future, I feel sorry because it's just going to get hooked 410 00:25:23,370 --> 00:25:27,290 Ejaaz: on AI slop dopamine and no one's going to be able to get out of it. 411 00:25:27,430 --> 00:25:29,370 Ejaaz: That's dangerous. We already have people that are, 412 00:25:29,930 --> 00:25:35,650 Ejaaz: committing crimes slash potential, you know, mistakes after interfacing with 413 00:25:35,650 --> 00:25:36,830 Ejaaz: these ChatGPT products. 414 00:25:37,030 --> 00:25:39,150 Ejaaz: And now you're going to go full on adult mode. And they're probably going to 415 00:25:39,150 --> 00:25:41,070 Ejaaz: charge a hefty premium to that as well. 416 00:25:41,590 --> 00:25:46,410 Ejaaz: The other reason why she was against this was ads, which they just rolled out. 417 00:25:46,590 --> 00:25:50,990 Ejaaz: So, you know, fair to her. She's sticking to her ethics and morals and she's quit. 418 00:25:51,570 --> 00:25:54,570 Josh: Yeah. Wow. All right. Well, another crazy week in the books, 419 00:25:54,610 --> 00:25:55,830 Josh: right? I think that's everything we got. 420 00:25:55,830 --> 00:26:03,750 Ejaaz: Yeah that is it just a crazy week of layoffs and breaking neck AI video models, 421 00:26:04,390 --> 00:26:05,810 Ejaaz: sprinkled with you know a few 422 00:26:05,810 --> 00:26:10,690 Ejaaz: moon catapults within the next five years so an exciting week as always, 423 00:26:11,550 --> 00:26:16,970 Ejaaz: there's a ton of stuff going on outside in the open source world which we were 424 00:26:16,970 --> 00:26:21,090 Ejaaz: discussing prior to this some claw bot stuff which we might get into next week 425 00:26:21,090 --> 00:26:25,230 Ejaaz: but until then we will see you in the next one.