1 00:00:00,020 --> 00:00:03,540 Ejaaz: The most powerful model in the world is here right now. In fact, 2 00:00:03,640 --> 00:00:06,340 Ejaaz: it's so good that it beats Claude mythos. 3 00:00:06,600 --> 00:00:12,760 Ejaaz: OpenAI just released ChatGPT 5.5 and it crushes Claude on every single benchmark. 4 00:00:12,920 --> 00:00:17,420 Ejaaz: It's the new number one coding model. It can do 20 hour tasks that expert software 5 00:00:17,420 --> 00:00:18,900 Ejaaz: engineers sometimes can't do. 6 00:00:19,040 --> 00:00:24,160 Ejaaz: It's already discovered groundbreaking solutions in maths and frontier sciences such as genetics. 7 00:00:24,600 --> 00:00:30,460 Ejaaz: And it's cheaper than GPT 5.4. This is the result of two years worth of frontier 8 00:00:30,460 --> 00:00:32,700 Ejaaz: research released in this one single model. 9 00:00:32,820 --> 00:00:36,340 Ejaaz: In fact, it's so good that an NVIDIA engineer said, and I quote, 10 00:00:36,720 --> 00:00:40,800 Ejaaz: losing access to GPT 5.5 feels like I've had a limb amputated. 11 00:00:40,940 --> 00:00:43,640 Josh: I think a lot of people are going to compare this to Opus 4.7, 12 00:00:43,760 --> 00:00:48,120 Josh: and that's fair, but I really think the true comparison is to Mythos because 13 00:00:48,120 --> 00:00:51,920 Josh: Sam Elman recently, he just posted something as the model was coming out that 14 00:00:51,920 --> 00:00:53,700 Josh: felt very much like a jab at Mythos. 15 00:00:53,800 --> 00:00:56,340 Josh: And we're going to get into the benchmarks comparing them, many of which will 16 00:00:56,340 --> 00:00:59,520 Josh: actually beat the Claude model. But what I find most interesting about this 17 00:00:59,520 --> 00:01:03,220 Josh: post is the second paragraph where he says, we believe in democratization. 18 00:01:03,380 --> 00:01:07,720 Josh: And he mentioned specifically, we have been tracking cybersecurity as a preparedness 19 00:01:07,720 --> 00:01:12,280 Josh: category for a long time and have built mitigations we believe in that enable 20 00:01:12,280 --> 00:01:14,400 Josh: us to make capable models broadly available. 21 00:01:14,400 --> 00:01:17,220 Josh: So this is very much a dig at Mythos, which is, as we all know, 22 00:01:17,360 --> 00:01:21,300 Josh: privately available, only gated to the companies that are given allowance to it. 23 00:01:22,000 --> 00:01:25,040 Josh: ChatGPT and OpenAI are like, hey, we're going to give you the powerful cybersecurity. 24 00:01:25,240 --> 00:01:29,300 Josh: We're just going to bake in the precautions into the model so that everyone could have it. 25 00:01:29,540 --> 00:01:32,980 Josh: And it ends by saying it's this really sweet thing. It's like we love you and 26 00:01:32,980 --> 00:01:35,560 Josh: we want you to win. We believe in everyone having access to this intelligence. 27 00:01:35,720 --> 00:01:38,380 Josh: And I really respect that. And I think it's an awesome way to set the precedence 28 00:01:38,380 --> 00:01:40,980 Josh: for what the next generation of these models is going to look like. 29 00:01:41,080 --> 00:01:44,540 Josh: But before we go any further, let's talk about the model itself. It's out right now. 30 00:01:44,680 --> 00:01:47,860 Josh: If you have a chat GPT membership, you can go and use it, go and play with it. 31 00:01:48,140 --> 00:01:50,820 Josh: EJS, what's the TLDR? What are the high-level things that everyone should know? 32 00:01:51,000 --> 00:01:53,880 Josh: What's most new and noteworthy about GPT 5.5? 33 00:01:54,140 --> 00:01:58,020 Ejaaz: Okay, so inspired by your mythos comparison, the first question that pops into 34 00:01:58,020 --> 00:02:02,300 Ejaaz: my head is I use Claude Opus 4.7 every single day. So I'm like, 35 00:02:02,600 --> 00:02:06,520 Ejaaz: is it better than this? Like, should I be switching back to ChatGPT right now? 36 00:02:06,800 --> 00:02:10,560 Ejaaz: The answer might be yes. So if we look at the benchmark score right here, 37 00:02:10,880 --> 00:02:15,760 Ejaaz: GPT 5.5 on the left over here absolutely crushes all the standard benchmarks 38 00:02:15,760 --> 00:02:17,820 Ejaaz: that these frontier models are weighted against. 39 00:02:17,880 --> 00:02:21,540 Ejaaz: And if you look on the right over here, Claude Opus 4.7, it either doesn't even 40 00:02:21,540 --> 00:02:26,380 Ejaaz: measure in a particular category, or it's completely beaten by GPT 5.5. 41 00:02:26,380 --> 00:02:31,800 Ejaaz: In fact, the only stat that GPT 5.5 doesn't beat Opus 4.7 in is something called 42 00:02:31,800 --> 00:02:35,780 Ejaaz: Software Engineering Benchmark Verified Pro or something like that. 43 00:02:35,860 --> 00:02:38,120 Ejaaz: It's like the pro software coding situation. 44 00:02:39,050 --> 00:02:42,290 Ejaaz: But there's a footnote at the bottom of this blog where OpenAI states, 45 00:02:42,830 --> 00:02:46,330 Ejaaz: Anthropic has publicly said that they might have gamed that particular benchmark 46 00:02:46,330 --> 00:02:47,830 Ejaaz: and they need to be re-evaluated. 47 00:02:47,890 --> 00:02:52,370 Ejaaz: So we might have a complete clean sweep for 5.5 as we see today. 48 00:02:52,490 --> 00:02:53,910 Ejaaz: So it's an incredibly powerful model. 49 00:02:54,030 --> 00:02:58,330 Ejaaz: But a question that popped to my head is, does it actually beat Mythos? 50 00:02:58,450 --> 00:03:00,110 Ejaaz: And we have a direct comparison right here. 51 00:03:00,310 --> 00:03:03,170 Josh: Yeah, so it shows that it does across some benchmarks. Now, again, 52 00:03:03,250 --> 00:03:04,270 Josh: these benchmarks are pretty fuzzy. 53 00:03:04,470 --> 00:03:08,030 Josh: We don't know which ones are gamed to do what. But there is a world in which 54 00:03:08,030 --> 00:03:12,930 Josh: GPT 5.5 will outperform Mythos on some things, which ones we're not entirely sure. 55 00:03:13,130 --> 00:03:18,470 Josh: I think as we kind of figure out ways to describe GPT 5.5, it seems as if it's 56 00:03:18,470 --> 00:03:23,010 Josh: their first attempt at making a model built for autonomy instead of answers. 57 00:03:23,210 --> 00:03:26,550 Josh: I think a lot of the benchmarks that they're working on is in agent decoding, 58 00:03:26,970 --> 00:03:30,010 Josh: things like it handles tasks that are 20 hours long. We'll get into that. 59 00:03:30,170 --> 00:03:33,650 Josh: It's doing 85% of OpenAI's internal work already. 60 00:03:33,650 --> 00:03:36,550 Josh: And it also helped rewrite the infrastructure that 61 00:03:36,550 --> 00:03:39,790 Josh: built it there was this amazing quote in the blog post it said open ai 62 00:03:39,790 --> 00:03:42,630 Josh: says 5.5 itself helped optimize the stack 63 00:03:42,630 --> 00:03:45,750 Josh: that serves it codex analyzed weeks of production traffic and 64 00:03:45,750 --> 00:03:49,630 Josh: wrote custom heuristics for load balancing that boosted token generation speed 65 00:03:49,630 --> 00:03:53,670 Josh: by over 20 so they're using the model to actually build the model and make it 66 00:03:53,670 --> 00:03:57,150 Josh: maximally efficient based on the data that it's collected from users like us 67 00:03:57,150 --> 00:04:00,150 Josh: who are interacting with the model on a daily basis so it's very smart it's 68 00:04:00,150 --> 00:04:02,730 Josh: very clever it's not just there to give you answers. 69 00:04:02,750 --> 00:04:06,850 Josh: It's there to think deeply and actually solve problems for you in a way that 70 00:04:06,850 --> 00:04:10,750 Josh: I think Mythos and a lot of these other frontier models are kind of pivoting towards now. 71 00:04:11,090 --> 00:04:15,210 Ejaaz: The great thing about this model release is it reveals a few things that OpenAI 72 00:04:15,210 --> 00:04:18,570 Ejaaz: has as an advantage against, say, a frontier lab like Anthropic. 73 00:04:18,910 --> 00:04:21,970 Ejaaz: It's clear looking at these benchmarks compared to Mythos, which, 74 00:04:22,090 --> 00:04:25,190 Ejaaz: by the way, the entire world is spiraling because of this model, 75 00:04:25,230 --> 00:04:29,570 Ejaaz: because it's going to have the cybersecurity ability to take over any kind of government system. 76 00:04:30,220 --> 00:04:33,840 Ejaaz: This model is pretty close, and Sam is going to be releasing this publicly, 77 00:04:34,080 --> 00:04:36,760 Ejaaz: or OpenAI is going to be releasing it publicly for everyone to use. 78 00:04:36,960 --> 00:04:41,520 Ejaaz: So a question that pops to my head is, does this mean that it's a matter of 79 00:04:41,520 --> 00:04:43,880 Ejaaz: compute, and OpenAI just simply has more of them? 80 00:04:44,020 --> 00:04:48,440 Ejaaz: Certainly, if you compare Sam Altman's ability to acquire compute and spend 81 00:04:48,440 --> 00:04:51,440 Ejaaz: all these trillions of dollars to acquire it versus Anthropic, 82 00:04:51,800 --> 00:04:54,040 Ejaaz: Anthropic has been extremely conservative, and now they're struggling. 83 00:04:54,320 --> 00:04:58,100 Ejaaz: They recently signed a $5 billion deal with Amazon, which we'll get to later 84 00:04:58,100 --> 00:05:01,180 Ejaaz: on. But the point is, this is a tale of two stories. 85 00:05:01,420 --> 00:05:05,600 Ejaaz: Either OpenAI has enough compute and they're about to leapfrog Claude because of that. 86 00:05:05,740 --> 00:05:09,780 Ejaaz: And they're proving that through this model that is a very good answer to Mythos. 87 00:05:10,100 --> 00:05:15,480 Ejaaz: Or, and this is the alternative side, Anthropic's Mythos model is just plainly better than 5.5. 88 00:05:15,620 --> 00:05:19,300 Ejaaz: And these benchmarks are actually verified, which is technically kind of true 89 00:05:19,300 --> 00:05:21,240 Ejaaz: because I don't know how official these things are. 90 00:05:21,320 --> 00:05:24,220 Ejaaz: These are just through tests that a small set of users have done. 91 00:05:24,500 --> 00:05:27,580 Ejaaz: So it's a game of both. I'm sure Anthropic is watching this and thinking, 92 00:05:27,580 --> 00:05:30,240 Ejaaz: hmm, maybe we should roll out Mythos, but they don't have to compute. 93 00:05:30,700 --> 00:05:33,220 Josh: Yeah, they don't have the inference. In fact, speaking of the inference, 94 00:05:33,420 --> 00:05:36,080 Josh: Sam actually made a post saying that he's really... 95 00:05:36,260 --> 00:05:39,640 Josh: Excellent work by the inference team to serve this model so efficiently he 96 00:05:39,640 --> 00:05:43,100 Josh: wanted to really highlight the fact that to a significant degree they've 97 00:05:43,100 --> 00:05:45,720 Josh: become an ai inference company now and i think that's a 98 00:05:45,720 --> 00:05:48,960 Josh: really big difference than what was previously stated like anthropic has really 99 00:05:48,960 --> 00:05:51,920 Josh: tough time serving compute and we see that and even if they had mythos available 100 00:05:51,920 --> 00:05:55,820 Josh: in a way that was safe they can't serve it open ai can and we see it reflected 101 00:05:55,820 --> 00:05:59,460 Josh: in pricing because i mean we have some pricing for this model right and it seems 102 00:05:59,460 --> 00:06:03,320 Josh: as if it's roughly at par with 4.7 if not slightly better? 103 00:06:03,500 --> 00:06:06,720 Ejaaz: Slightly. It's slightly more expensive, but not by much. 104 00:06:06,860 --> 00:06:13,540 Ejaaz: So for every million tokens input, it's both the same for Anthropic Opus 4.7 and GPT 5.5. 105 00:06:13,660 --> 00:06:22,280 Ejaaz: It's $5 in, but the output is $30 for 5.5 per million tokens and $25 per million tokens for 4.7. 106 00:06:22,980 --> 00:06:27,820 Ejaaz: So it's a little more expensive, but here's where you actually have more of 107 00:06:27,820 --> 00:06:32,920 Ejaaz: a bargain using the more expensive model 5.5. It is cheaper than GPT 5.4, 108 00:06:33,000 --> 00:06:36,960 Ejaaz: and it uses tokens way more efficiently to think. 109 00:06:37,160 --> 00:06:41,660 Ejaaz: So what does that mean if you are an enterprise that wants to plug in this AI 110 00:06:41,660 --> 00:06:45,220 Ejaaz: model and not worry about it and just have it power your entire profit engine? 111 00:06:45,440 --> 00:06:50,520 Ejaaz: Well, you end up using less tokens, so you hit your rate limits in a much slower 112 00:06:50,520 --> 00:06:54,300 Ejaaz: rate, which means that you end up getting more bang for your buck as long as 113 00:06:54,300 --> 00:06:57,140 Ejaaz: you use the model like 24-7 or you use it effectively. 114 00:06:57,140 --> 00:07:01,780 Ejaaz: Way if you are just kind of out there using 5.5 to like ask questions that you 115 00:07:01,780 --> 00:07:06,640 Ejaaz: should maybe be asking google this is probably not the model for you but otherwise super powerful one 116 00:07:07,180 --> 00:07:09,460 Josh: Yeah. And if these prices don't mean anything to you, that's fine. 117 00:07:09,460 --> 00:07:11,660 Josh: As long as you have a $20 a month subscription. 118 00:07:11,920 --> 00:07:15,380 Josh: In fact, this is going to be available to freezers fairly soon, I believe. 119 00:07:15,580 --> 00:07:18,440 Josh: But anyone who is a subscriber has access to this. You don't need to use the 120 00:07:18,440 --> 00:07:19,420 Josh: API. There's nothing fancy. 121 00:07:19,600 --> 00:07:21,920 Josh: You open up your app on your phone, you go to the web browser, 122 00:07:21,920 --> 00:07:23,840 Josh: it's there, it's available, ready to go. 123 00:07:23,940 --> 00:07:26,900 Josh: Now, there's a few interesting things that you can do with this model that haven't 124 00:07:26,900 --> 00:07:27,780 Josh: previously been possible. 125 00:07:28,060 --> 00:07:31,020 Josh: And although we don't quite have access to it just yet, we're recording this 126 00:07:31,020 --> 00:07:32,680 Josh: right as the model got launched. 127 00:07:32,820 --> 00:07:37,140 Josh: We do have a blog post from OpenAI themselves who are showcasing a few demos 128 00:07:37,140 --> 00:07:40,340 Josh: so again take these with a grain of salt these are straight from open ai but 129 00:07:40,340 --> 00:07:43,900 Josh: they are seemingly pretty impressive and pretty noteworthy as to what they're 130 00:07:43,900 --> 00:07:47,940 Josh: capable of doing starting with this space mission application which is um pretty 131 00:07:47,940 --> 00:07:51,200 Josh: cool and very reminiscent of the moon mission that we just had yeah. 132 00:07:51,200 --> 00:07:55,740 Ejaaz: Um so if you guys don't know um josh has a secret he has many secrets on this 133 00:07:55,740 --> 00:08:00,460 Ejaaz: show one is he's a massive space fan and when he's not hanging out with me he's 134 00:08:00,460 --> 00:08:04,760 Ejaaz: doing uh space simulations uh on whatever he can do right well okay maybe maybe 135 00:08:04,760 --> 00:08:05,920 Ejaaz: be part of that is a bit of a lie. 136 00:08:06,100 --> 00:08:09,340 Ejaaz: But with this new app that we're seeing in front of us right now, 137 00:08:09,440 --> 00:08:12,300 Ejaaz: this was completely vibe coded using 5.5. 138 00:08:12,540 --> 00:08:15,460 Ejaaz: And it's used to simulate a specific space mission. 139 00:08:15,560 --> 00:08:18,740 Ejaaz: Now, if this looks very similar, it's because we just had a space mission for 140 00:08:18,740 --> 00:08:21,900 Ejaaz: some we visited or went back to the moon in 53 years, pretty big deal. 141 00:08:22,120 --> 00:08:25,200 Ejaaz: And we can see a pretty accurate simulation going on right here. 142 00:08:25,300 --> 00:08:28,640 Ejaaz: So as you can see, there's various different toggles, the physics of the entire 143 00:08:28,640 --> 00:08:30,180 Ejaaz: thing is very important. 144 00:08:30,340 --> 00:08:33,300 Ejaaz: And that's another point I want to make about this model. it is being 145 00:08:33,300 --> 00:08:36,520 Ejaaz: used for frontier research not just in ai but in 146 00:08:36,520 --> 00:08:40,700 Ejaaz: mathematics in genetics like it made frontier progression on both of these fronts 147 00:08:40,700 --> 00:08:45,100 Ejaaz: and so what we're showing here is this is a model that goes way beyond just 148 00:08:45,100 --> 00:08:49,720 Ejaaz: text and telling you what could be it actually implements this into a lot of 149 00:08:49,720 --> 00:08:52,760 Ejaaz: different things and understands the world around it which is extremely powerful 150 00:08:52,760 --> 00:08:55,940 Ejaaz: but we have another one here we have a we have an earthquake tracker 151 00:08:56,490 --> 00:08:59,490 Josh: For anyone who wants to make websites, it's so good at making websites. 152 00:08:59,610 --> 00:09:01,910 Josh: And this appears to be one of the strong suits. In this case, 153 00:09:02,050 --> 00:09:04,090 Josh: there's a few things to highlight on this Earthquake tracker. 154 00:09:04,350 --> 00:09:07,590 Josh: One of them being that it's one, just like a pretty elegantly designed website. 155 00:09:07,730 --> 00:09:11,110 Josh: But two, all of the graphics are interactive. You'll notice that they update 156 00:09:11,110 --> 00:09:14,390 Josh: dynamically as you hover over them, as you click. It looks very clean. 157 00:09:14,570 --> 00:09:18,950 Josh: I assume that it is pulling up-to-date information from an API somewhere that it set up. 158 00:09:19,070 --> 00:09:23,170 Josh: It is just truly competent and capable of doing these kind of longer tail tasks 159 00:09:23,170 --> 00:09:25,830 Josh: that are a bit more complicated than a static landing page, 160 00:09:25,830 --> 00:09:29,470 Josh: but have dynamic data have the richness that 161 00:09:29,470 --> 00:09:32,550 Josh: you would expect from a high-end high-quality polished website 162 00:09:32,550 --> 00:09:35,390 Josh: except just built with an ai model from someone who doesn't 163 00:09:35,390 --> 00:09:38,290 Josh: need to know anything about coding at all and then 164 00:09:38,290 --> 00:09:41,190 Josh: for the gamers also there's another great example of a dungeon game 165 00:09:41,190 --> 00:09:44,250 Josh: which is they're describing as a playable 3d dungeon arena prototype 166 00:09:44,250 --> 00:09:47,130 Josh: built with codex and gpt models now 167 00:09:47,130 --> 00:09:50,110 Josh: i think this is something novel to this setup where codex handles 168 00:09:50,110 --> 00:09:54,070 Josh: the game architecture the combat systems the enemy encounters and then the character 169 00:09:54,070 --> 00:09:58,110 Josh: models the character textures and animations those were created with third-party 170 00:09:58,110 --> 00:10:03,190 Josh: asset generation tools using something like image gen 2.0 so this is also one 171 00:10:03,190 --> 00:10:06,050 Josh: of the earlier signs where you can actually merge a lot of these tools together 172 00:10:06,050 --> 00:10:09,290 Josh: to build something dynamic in a way that you previously couldn't have done before. 173 00:10:09,290 --> 00:10:13,350 Ejaaz: Yet actually the quality of this game looks like something out of uh league 174 00:10:13,350 --> 00:10:15,770 Ejaaz: of legends or something like that at least that's what it reminds me of like 175 00:10:15,770 --> 00:10:19,730 Ejaaz: the these games are getting way more high def than i expected i know it's just 176 00:10:19,730 --> 00:10:22,290 Ejaaz: like it's pretty basic for anyone that's watching this they can kind of like 177 00:10:22,290 --> 00:10:25,830 Ejaaz: pick with a finer eye but it's cool but for those of you who prefer like the 178 00:10:25,830 --> 00:10:27,050 Ejaaz: more traditional side of games 179 00:10:27,490 --> 00:10:30,270 Ejaaz: this might be something that you can kind of vibe code in a 180 00:10:30,270 --> 00:10:33,750 Ejaaz: couple of minutes now it may look basic but theoretically 181 00:10:33,750 --> 00:10:38,390 Ejaaz: this is like a 3D spatially aware game and that's not something you could achieve 182 00:10:38,390 --> 00:10:42,190 Ejaaz: at least very easily with previous models what I love about this as well is 183 00:10:42,190 --> 00:10:46,290 Ejaaz: it's also they've also created or included the prompt for all of these things 184 00:10:46,290 --> 00:10:49,470 Ejaaz: so this is something that you can try right now like look at this And the prompt 185 00:10:49,470 --> 00:10:50,550 Ejaaz: is no more than like, what's it? 186 00:10:50,730 --> 00:10:54,130 Ejaaz: One, two, three, four, like 12 lives. 12 lives, dude. 187 00:10:54,470 --> 00:10:58,010 Ejaaz: And you can have like a fully functioning game. You can probably then add an 188 00:10:58,010 --> 00:11:00,970 Ejaaz: extra step or extra prompt saying, hey, can you deploy this to Vercel? And- 189 00:11:01,250 --> 00:11:03,810 Ejaaz: Send that to your friends. Now you can use, you have a game. 190 00:11:03,890 --> 00:11:05,090 Ejaaz: You're a game creator. You're a game developer. 191 00:11:05,250 --> 00:11:09,790 Ejaaz: So the applications for this model cannot be understated. 192 00:11:10,170 --> 00:11:13,770 Ejaaz: I'm going to be very honest. I thought this model was going to be just an iterative 193 00:11:13,770 --> 00:11:16,470 Ejaaz: upgrade. I didn't think it would get anywhere near Claude Mythos. 194 00:11:17,210 --> 00:11:21,490 Ejaaz: Two stories have now revealed themselves, which is, one, it's the answer to Claude Mythos. 195 00:11:21,710 --> 00:11:25,910 Ejaaz: And two, it's really damn good. I am now convinced that compute is everything, 196 00:11:26,050 --> 00:11:28,750 Ejaaz: but not in the way that I thought it would be useful. I thought it would be 197 00:11:28,750 --> 00:11:31,450 Ejaaz: largely for pre-training. 198 00:11:31,870 --> 00:11:36,770 Ejaaz: But to Sam's tweet earlier on, and also in Greg Brockman, the president of OpenAI's 199 00:11:36,770 --> 00:11:40,210 Ejaaz: recent interview, they're going all in on inference, test time compute, 200 00:11:40,530 --> 00:11:43,410 Ejaaz: which just means that if you have more compute and if you have a good enough 201 00:11:43,410 --> 00:11:44,810 Ejaaz: model, it can do the thing. 202 00:11:44,930 --> 00:11:48,550 Ejaaz: This thing, like I said, built itself. It's a self-improving model. Very, very impressive. 203 00:11:48,870 --> 00:11:51,970 Josh: It's good for solving hard problems. It's good for thinking for a long time. 204 00:11:52,090 --> 00:11:56,030 Josh: In fact, they marketed it as a model that can now think for 20 hours coherently. 205 00:11:56,030 --> 00:11:58,790 Josh: Great which is almost a full day it can work 206 00:11:58,790 --> 00:12:01,690 Josh: on a problem yeah and what you're noticing from this prompt that's on screen is 207 00:12:01,690 --> 00:12:04,530 Josh: it doesn't take that much to get it going you don't need to 208 00:12:04,530 --> 00:12:07,390 Josh: kind of spoon feed it all the way through anymore it can make 209 00:12:07,390 --> 00:12:10,090 Josh: decisions on its own it can infer conclusions on what 210 00:12:10,090 --> 00:12:12,810 Josh: you want just based on the the knowledge architecture that it 211 00:12:12,810 --> 00:12:15,690 Josh: currently has it's amazingly impressive in fact one of the people who got access to 212 00:12:15,690 --> 00:12:18,770 Josh: it early just posted on x that he's posting 213 00:12:18,770 --> 00:12:21,850 Josh: live as his um prompt is seven hours 214 00:12:21,850 --> 00:12:24,850 Josh: into his task it has been running for over seven hours 215 00:12:24,850 --> 00:12:27,650 Josh: he said this has literally never happened before the models would maybe run 216 00:12:27,650 --> 00:12:30,430 Josh: for 30 minutes or so wow or or if you 217 00:12:30,430 --> 00:12:34,410 Josh: really shouted them after two to three hours but he's on seven plus hours i 218 00:12:34,410 --> 00:12:37,630 Josh: think this is going to be fun for people with complicated things if you really 219 00:12:37,630 --> 00:12:42,370 Josh: want to make a triple a feeling video game or a simulator or a really complex 220 00:12:42,370 --> 00:12:46,670 Josh: website this is the model to try out and to use it with codex and see how all 221 00:12:46,670 --> 00:12:48,650 Josh: these things kind of piece together it's really i mean 222 00:12:49,080 --> 00:12:54,240 Josh: I wasn't, I didn't have my hopes very high based on the Opus 4.7 to 4.6 incremental improvement. 223 00:12:54,480 --> 00:12:57,400 Josh: This seems like a very solid improvement over 5.4. 224 00:12:57,600 --> 00:13:01,700 Ejaaz: Absolutely. And listen, if you are listening to this and you're like, listen, I'm not a gamer. 225 00:13:01,900 --> 00:13:04,900 Ejaaz: I can't waste my time with that. I focus on more serious things. 226 00:13:05,060 --> 00:13:09,360 Ejaaz: Well, for you serious people, if you're a manager at a top company or whatever 227 00:13:09,360 --> 00:13:13,100 Ejaaz: that might be, this isn't just a toy or a model used for coders. 228 00:13:13,160 --> 00:13:15,340 Ejaaz: A lot of the examples that we just gave are around coding. 229 00:13:15,520 --> 00:13:20,400 Ejaaz: You can use this for just admin stuff or managerial work, like the capability 230 00:13:20,400 --> 00:13:24,860 Ejaaz: of this model to think more strategically and long-term and understand the context 231 00:13:24,860 --> 00:13:26,320 Ejaaz: of the tasks that you're working towards. 232 00:13:26,320 --> 00:13:31,900 Ejaaz: Like we said earlier, for coding specifically, it can work on 20-hour-long expert tasks. 233 00:13:32,200 --> 00:13:36,340 Ejaaz: That also applies for administrative stuff or things that are more generalized, 234 00:13:36,520 --> 00:13:38,060 Ejaaz: white-collar worker work. 235 00:13:38,240 --> 00:13:41,700 Ejaaz: And so in this example, Noam Brown says, I'm a manager at OpenAI, 236 00:13:41,840 --> 00:13:45,140 Ejaaz: but I'm using this model to basically manage my entire team and make sure we're 237 00:13:45,140 --> 00:13:45,980 Ejaaz: focused on the right things. 238 00:13:46,020 --> 00:13:49,520 Ejaaz: And guess what? but the output of this team and this product has been pretty amazing. 239 00:13:49,660 --> 00:13:54,880 Ejaaz: So all around really excellent work by the entire team and the inference team 240 00:13:54,880 --> 00:13:56,420 Ejaaz: specifically, as Sam Altman says here. 241 00:13:56,620 --> 00:13:59,680 Ejaaz: And yeah, I'm looking forward to using this thing. I don't have access to it right now. 242 00:13:59,800 --> 00:14:04,700 Ejaaz: I've refreshed my account probably like five times at this point and it hasn't 243 00:14:04,700 --> 00:14:06,000 Ejaaz: appeared. So maybe it's like a slow rollout. 244 00:14:06,080 --> 00:14:08,760 Ejaaz: But if you're listening to this and you've tried it out, let us know what you're 245 00:14:08,760 --> 00:14:11,420 Ejaaz: using it for. Let us know what amazes you. I really want to hear more. 246 00:14:11,920 --> 00:14:14,700 Josh: Yeah, OpenAX had a pretty incredible week. And this comes on the back of their 247 00:14:14,700 --> 00:14:17,460 Josh: new ImageGen model that they just released, which was also unbelievable. 248 00:14:17,480 --> 00:14:20,060 Josh: If you haven't seen that episode, we just recorded it yesterday. 249 00:14:20,160 --> 00:14:23,400 Josh: So I would go advise you to see because, oh my God, it is amazing. 250 00:14:23,560 --> 00:14:27,440 Josh: We also recorded an episode on Apple's new CEO this week and what that means 251 00:14:27,440 --> 00:14:31,800 Josh: for the company, as well as the hardware race and how this, I mean, 252 00:14:32,020 --> 00:14:34,820 Josh: this model, Opus, no, not Opus, this is GPT. 253 00:14:34,940 --> 00:14:40,700 Josh: GPT 5.5 is very much part of the AGI class of models that is built on Blackwell 254 00:14:40,700 --> 00:14:42,880 Josh: chips. and we've recorded an entire episode all about that. 255 00:14:43,120 --> 00:14:46,160 Josh: Very interesting, very fascinating. Also interesting and fascinating because 256 00:14:46,160 --> 00:14:49,480 Josh: as always, this is the weekly roundup. We have a few other topics to talk about. 257 00:14:49,620 --> 00:14:52,920 Josh: We have some news out of SpaceX, which is a pseudo acquisition. 258 00:14:53,200 --> 00:14:56,860 Josh: Now they haven't quite acquired Cursor being the company in question, 259 00:14:56,860 --> 00:15:01,560 Josh: but they have at least partnered with them with the option to buy Cursor for 260 00:15:01,560 --> 00:15:05,480 Josh: either $60 billion or pay 10 billion for the right to actually work together. 261 00:15:05,640 --> 00:15:06,800 Josh: This seems like a big deal. 262 00:15:07,100 --> 00:15:11,440 Josh: This seems like, I mean, XAI, we could call it SpaceX, but SpaceX AI is taking 263 00:15:11,440 --> 00:15:13,540 Josh: AI very seriously. They're currently behind. 264 00:15:13,760 --> 00:15:17,460 Josh: They clearly don't want to be behind. This is a huge step and a huge kind of 265 00:15:17,460 --> 00:15:21,700 Josh: trust of support in Cursor with this minimum of $10 billion into accelerating 266 00:15:21,700 --> 00:15:23,600 Josh: their progress and trying to get themselves into this game. 267 00:15:24,000 --> 00:15:30,480 Ejaaz: This is actually a genius deal, and there are a few stories why it makes that so. So let me explain. 268 00:15:30,780 --> 00:15:34,420 Ejaaz: If you're SpaceX AI, which by the way is a ridiculous name now, 269 00:15:34,540 --> 00:15:38,460 Ejaaz: like we'll just call them XAI, you are currently harboring... 270 00:15:39,180 --> 00:15:47,260 Ejaaz: One to 1.5 million of the frontier GPUs, mainly NVIDIA, in a warehouse. There's one issue. 271 00:15:47,580 --> 00:15:51,840 Ejaaz: You're not really utilizing all of it because XAI has had a bit of a slow start 272 00:15:51,840 --> 00:15:54,120 Ejaaz: to training their models. What's a genius idea? 273 00:15:54,880 --> 00:15:59,120 Ejaaz: If I rent those out to another company to train their own model, 274 00:15:59,340 --> 00:16:02,720 Ejaaz: then we can make money from that. Okay, so that's win number one for SpaceX. 275 00:16:02,980 --> 00:16:06,880 Ejaaz: But then they've thought of another thing which is huh grok 276 00:16:06,880 --> 00:16:09,600 Ejaaz: isn't really good at coding and we are 277 00:16:09,600 --> 00:16:12,420 Ejaaz: losing the race every single day we don't update our model 278 00:16:12,420 --> 00:16:15,760 Ejaaz: at coding because anthropic and chat gpt 279 00:16:15,760 --> 00:16:18,860 Ejaaz: 5.5 is completely running away with it so 280 00:16:18,860 --> 00:16:23,180 Ejaaz: how did they leapfrog and get ahead they should acquire the company that is 281 00:16:23,180 --> 00:16:27,400 Ejaaz: using their own gpus to train a frontier coding model so then the question becomes 282 00:16:27,400 --> 00:16:31,960 Ejaaz: well who the hell is cursor what what's the mode that they have like why do 283 00:16:31,960 --> 00:16:36,100 Ejaaz: they have a good shot of training a better coding model than Anthropic and GPT-505? 284 00:16:36,300 --> 00:16:39,860 Ejaaz: Aren't those two companies way ahead? Well, the answer is not quite so. 285 00:16:40,020 --> 00:16:44,620 Ejaaz: Cursor, for the longest time, was the number one platform and tool for people 286 00:16:44,620 --> 00:16:46,240 Ejaaz: to use to do their Vibe coding. Why? 287 00:16:46,520 --> 00:16:50,440 Ejaaz: Not only did they have access to Frontier coding models from Claude and ChatGPT, 288 00:16:50,620 --> 00:16:53,600 Ejaaz: they also had something called an agent harness. 289 00:16:53,760 --> 00:16:58,520 Ejaaz: Now, you'll notice in GPT-505, it's really good at coding because of something 290 00:16:58,520 --> 00:17:02,880 Ejaaz: called agentic That is something that Cursor pretty much pioneered. 291 00:17:03,020 --> 00:17:07,540 Ejaaz: It's basically the harness, the prompts, the environment that they mold the model, 292 00:17:07,820 --> 00:17:12,680 Ejaaz: or rather that they mold around the model that makes it so good and intuitive 293 00:17:12,680 --> 00:17:15,860 Ejaaz: and remembers the context across every single project, like menial things, 294 00:17:16,000 --> 00:17:20,220 Ejaaz: like understanding your GitHub branches and working on separate flows at the same time. 295 00:17:20,580 --> 00:17:25,560 Ejaaz: A lot of the top software engineers in the world right now use tools like Curse 296 00:17:25,560 --> 00:17:27,200 Ejaaz: and Argentic Coding to be able to pull this off. 297 00:17:27,520 --> 00:17:32,020 Ejaaz: So Elon Musk thought, hmm, if I give you the GPUs to train a better coding model, 298 00:17:32,160 --> 00:17:35,980 Ejaaz: which gives you a better product, I should have the option to acquire you. 299 00:17:36,200 --> 00:17:40,340 Ejaaz: In acquiring you, I can integrate you with Grok and Grok somehow becomes the 300 00:17:40,340 --> 00:17:44,460 Ejaaz: number one coding model over the next year or so, depending on if this deal goes. 301 00:17:44,640 --> 00:17:47,880 Ejaaz: And if the deal falls through and they create a really bad model, 302 00:17:48,260 --> 00:17:50,480 Ejaaz: well, you pay me $10 billion for the service. 303 00:17:50,580 --> 00:17:52,720 Ejaaz: Well i pay you not a bad deal not a bad deal 304 00:17:52,720 --> 00:17:55,620 Josh: Yeah it seems like they're they're going to be continuing to 305 00:17:55,620 --> 00:17:58,320 Josh: work with other companies to accelerate in places that 306 00:17:58,320 --> 00:18:01,120 Josh: they're weak at currently because i mean they they're so strong at 307 00:18:01,120 --> 00:18:04,120 Josh: building out the hardware and creating these huge data centers they need 308 00:18:04,120 --> 00:18:06,920 Josh: someone who could take advantage of all those gpus hopefully this will 309 00:18:06,920 --> 00:18:09,640 Josh: help serve that cause and that's not the only spacex news 310 00:18:09,640 --> 00:18:12,540 Josh: this week the other is that they have officially filed an s1 which 311 00:18:12,540 --> 00:18:15,740 Josh: for those who are not familiar it means they're going public it's officially official 312 00:18:15,740 --> 00:18:18,640 Josh: 100 they will be going public this year if there 313 00:18:18,640 --> 00:18:21,880 Josh: were any doubts please let them be relinquished here we 314 00:18:21,880 --> 00:18:24,740 Josh: have it spacex will be going published the most interesting thing from 315 00:18:24,740 --> 00:18:28,300 Josh: this was i think the share structure of 316 00:18:28,300 --> 00:18:32,320 Josh: how they're going to be organizing this for daddy elon who's going to be getting 317 00:18:32,320 --> 00:18:37,020 Josh: quite a big payday if he does well so we have on screen here just a series of 318 00:18:37,020 --> 00:18:39,880 Josh: some of the financials i mean we know starlink as a business has been doing 319 00:18:39,880 --> 00:18:44,280 Josh: unbelievable they have about 25 billion dollars in cash 92 billion assets 50 320 00:18:44,280 --> 00:18:45,580 Josh: billion liabilities that's. 321 00:18:45,580 --> 00:18:47,900 Ejaaz: Quite a lot of liabilities on this my god 322 00:18:47,900 --> 00:18:51,540 Josh: They got a lot of debt man i don't know we'll see we'll see once they finally 323 00:18:51,540 --> 00:18:54,060 Josh: publish everything i'm very excited for the first earnings report where you 324 00:18:54,060 --> 00:18:57,440 Josh: really get a true peek behind the scenes of what's going on there but it looks 325 00:18:57,440 --> 00:19:02,300 Josh: like it's going to be going public at a 1.75 trillion dollar valuation now in terms of pay structure. 326 00:19:02,800 --> 00:19:05,680 Josh: Elon is posed to get 60 million shares 327 00:19:05,680 --> 00:19:09,120 Josh: which is 11 tranches vesting in 328 00:19:09,120 --> 00:19:16,000 Josh: 500 billion dollar market cap increments from 1.1 trillion to 6.6 trillion dollar 329 00:19:16,000 --> 00:19:21,160 Josh: share price oh um so for those unfamiliar with the current ceiling i think it's 330 00:19:21,160 --> 00:19:25,220 Josh: nvidia nvidia is what five trillion under five trillion close to five trillion 331 00:19:25,220 --> 00:19:29,000 Josh: it's like 4.3 yeah okay so not even close they're like 20 away from five trillion. 332 00:19:29,500 --> 00:19:33,760 Josh: SpaceX needs to be, what is that? Like 20 something percent more valuable than 333 00:19:33,760 --> 00:19:34,780 Josh: the most valuable company in the world. 334 00:19:34,880 --> 00:19:38,500 Josh: But if they do, Elon gets 60 million shares. 335 00:19:38,620 --> 00:19:41,320 Josh: Now I haven't done the math on exactly how much that is. 336 00:19:41,700 --> 00:19:47,600 Josh: But if we make some assumptions here, the total value at Vest looks like it 337 00:19:47,600 --> 00:19:49,140 Josh: could be about a quarter of a trillion dollars. 338 00:19:49,340 --> 00:19:52,920 Josh: So pretty good payday for Elon. I think the most important thing is that he's 339 00:19:52,920 --> 00:19:54,320 Josh: getting a lot of control over this. 340 00:19:54,480 --> 00:19:58,220 Josh: It seems as if he's going to have 40 something percent control of 341 00:19:58,220 --> 00:20:00,940 Josh: the company which is really ultimately what was most important to 342 00:20:00,940 --> 00:20:04,500 Josh: him as they went public so really exciting news i am hopeful that it happens 343 00:20:04,500 --> 00:20:08,140 Josh: this june which we can expect and it's without a shadow of a doubt going to 344 00:20:08,140 --> 00:20:11,480 Josh: be the largest ipo in history i think everyone's going to be talking about it 345 00:20:11,480 --> 00:20:14,640 Josh: there is a new vehicle in which some people are investing in we're actually 346 00:20:14,640 --> 00:20:18,560 Josh: going to have the founder on the show soon so keep an eye out for that one. 347 00:20:19,020 --> 00:20:20,920 Josh: And yeah, the SpaceX news is very exciting. 348 00:20:21,180 --> 00:20:27,520 Ejaaz: Now, in the world of AI hardware, many people think that NVIDIA has run away with the win. 349 00:20:27,660 --> 00:20:31,320 Ejaaz: And you could argue that with a $4.300 market cap, not many people are competing, 350 00:20:31,480 --> 00:20:33,720 Ejaaz: except that there is one company, Google. 351 00:20:34,040 --> 00:20:36,500 Ejaaz: Now, you might be thinking, Google does all my search engines and stuff. 352 00:20:36,640 --> 00:20:43,480 Ejaaz: Well, Google is the only vertically integrated Mag 7 company that is involved 353 00:20:43,480 --> 00:20:47,120 Ejaaz: or has a frontier capability at every single layer of the AI stack. 354 00:20:47,280 --> 00:20:51,120 Ejaaz: Now, right at the bottom are these things called Google TPUs, 355 00:20:51,180 --> 00:20:52,640 Ejaaz: Tensor Processing Units. 356 00:20:52,740 --> 00:20:55,840 Ejaaz: And they're their version of the GPU. In fact, fun fact, 357 00:20:56,290 --> 00:21:00,470 Ejaaz: Google's Gemini models has never trained on an NVIDIA GPU. 358 00:21:00,770 --> 00:21:04,490 Ejaaz: It's all been their own internal warehoused infrastructure. And they've been 359 00:21:04,490 --> 00:21:05,930 Ejaaz: working on this thing for 10 years. 360 00:21:06,270 --> 00:21:10,650 Ejaaz: Now, just today, or rather this week, they released their latest generation 361 00:21:10,650 --> 00:21:14,710 Ejaaz: of TPUs, the TPU-8T and the TPU-8i. 362 00:21:14,990 --> 00:21:18,830 Ejaaz: Now, the TPU-8T, T stands for training or pre-training. 363 00:21:19,030 --> 00:21:23,570 Ejaaz: It is highly optimized for the pre-training part of an AI model. 364 00:21:23,570 --> 00:21:27,690 Ejaaz: So this is like the bulk, arguably the more expensive part of training a model. 365 00:21:27,830 --> 00:21:29,790 Ejaaz: It's like teaching it like, hey, these are words. 366 00:21:29,950 --> 00:21:32,970 Ejaaz: These are the general fundamental set of facts that you need to know before 367 00:21:32,970 --> 00:21:36,810 Ejaaz: we can kind of like put you out into the world and present you to our users. 368 00:21:37,490 --> 00:21:42,050 Ejaaz: TPU AI is specialized or hyper-specialized in inference specifically. 369 00:21:42,450 --> 00:21:46,350 Ejaaz: Now, the important part about inference is it's being used for so many different things. 370 00:21:46,470 --> 00:21:49,170 Ejaaz: Number one, it's to answer all your different prompts. Whenever you write a 371 00:21:49,170 --> 00:21:52,330 Ejaaz: prompt and you submit it to an AI model, it is known as inference. 372 00:21:52,510 --> 00:21:55,770 Ejaaz: It's getting inference. It needs to query the model and make sure it does the 373 00:21:55,770 --> 00:21:57,670 Ejaaz: right types of thinking and gives you the right answer. 374 00:21:57,790 --> 00:22:01,770 Ejaaz: But the other part of inference is post-training, where a lot of people train 375 00:22:01,770 --> 00:22:06,270 Ejaaz: the model, and then they do more training after the fact by using it to help 376 00:22:06,270 --> 00:22:11,270 Ejaaz: the model reason and think of other alternative facts before it presents you the actual answer. 377 00:22:11,510 --> 00:22:15,690 Ejaaz: And that's what that second TPU is. Now, Google's TPUs have been used extensively. 378 00:22:15,850 --> 00:22:19,390 Ejaaz: In fact, their largest customer is a little-known AI lab known as Anthropic, 379 00:22:19,510 --> 00:22:25,150 Ejaaz: which currently runs 1.5 million TPUs. So the argument can be made that TPUs 380 00:22:25,150 --> 00:22:28,050 Ejaaz: are largely responsible for Claude's and Opus' success. 381 00:22:28,210 --> 00:22:31,330 Ejaaz: So very impressive all around, but there's some other facts about this, right? 382 00:22:31,910 --> 00:22:36,790 Josh: Yeah, well, I love the dual architecture training setup that they have here being hyper specific. 383 00:22:37,050 --> 00:22:40,630 Josh: I mean, the AT chip in particular, it's built to reduce frontier model development 384 00:22:40,630 --> 00:22:42,970 Josh: cycles, they said, from months to weeks. 385 00:22:43,170 --> 00:22:46,490 Josh: And then we have the AI, which is the reasoning engine, which is specifically 386 00:22:46,490 --> 00:22:51,350 Josh: served for agentic use to deliver tokens really quick, as fast as possible. 387 00:22:51,530 --> 00:22:54,430 Josh: And as we know, Anthropic is working closely with them. And also, 388 00:22:54,550 --> 00:22:55,910 Josh: I mean, Google is making these for themselves. 389 00:22:55,910 --> 00:23:00,350 Josh: So I think whoever is working with Google, whoever's kind of focused on these 390 00:23:00,350 --> 00:23:05,650 Josh: accelerators, is probably in for a nice little windfall as it relates to increased 391 00:23:05,650 --> 00:23:08,850 Josh: velocity of the training and also increased ability to distribute these models. 392 00:23:08,890 --> 00:23:11,310 Josh: As we know, Anthropik is having a very difficult time with this. 393 00:23:11,610 --> 00:23:14,850 Josh: Now, NVIDIA and Jensen are probably feeling a little shook. 394 00:23:14,990 --> 00:23:17,770 Josh: They got to be feeling a little bit of pressure here. and it seems as if 395 00:23:17,770 --> 00:23:20,690 Josh: that's why they're pushing to be open source because if you 396 00:23:20,690 --> 00:23:23,550 Josh: are a in a closed source world where everyone is 397 00:23:23,550 --> 00:23:26,590 Josh: making closed source models on their own architecture then the 398 00:23:26,590 --> 00:23:29,730 Josh: nvidia edge very quickly disappears and i mean i'm looking at these chips in 399 00:23:29,730 --> 00:23:32,690 Josh: hand they look beautiful they're ready to be they're taped out ready to be manufactured 400 00:23:32,690 --> 00:23:36,750 Josh: and i i think you could start getting kind of excited about this new world of 401 00:23:36,750 --> 00:23:40,370 Josh: accelerated hardware and we're seeing this happen again and again because amazon 402 00:23:40,370 --> 00:23:45,810 Josh: just made another big investment in who else other than Anthropic. 403 00:23:46,110 --> 00:23:48,930 Josh: And the deal, I think, is like this has to be close to a record deal. 404 00:23:49,070 --> 00:23:50,510 Josh: They're owning a tremendous amount of this company now. 405 00:23:50,850 --> 00:23:57,390 Ejaaz: Yep. So the news here is Amazon announced they're investing $5 billion into Anthropic. 406 00:23:57,890 --> 00:23:59,790 Ejaaz: They've just raised $5 billion. Congrats. 407 00:24:00,510 --> 00:24:04,430 Ejaaz: And so the reason why this is important is, well, there's a few reasons. 408 00:24:04,730 --> 00:24:07,450 Ejaaz: Number one, Anthropic knows that they don't have enough compute. 409 00:24:07,710 --> 00:24:10,310 Ejaaz: The argument could be made that's why Claude Mythos hasn't been rolled out. 410 00:24:10,570 --> 00:24:13,430 Ejaaz: Well, hey, hey, presto, now you have $5 billion worth more of compute. 411 00:24:14,130 --> 00:24:19,250 Ejaaz: Now, for those of you who didn't know, Amazon is a primary investor already in Anthropic. 412 00:24:19,630 --> 00:24:23,250 Ejaaz: Before this announcement, they owned around 17% of Anthropic. 413 00:24:23,450 --> 00:24:28,070 Ejaaz: After this announcement, it's closer to 20%. So we're talking about one company 414 00:24:28,070 --> 00:24:31,890 Ejaaz: that's publicly tradable right now that owns a fifth. 415 00:24:32,150 --> 00:24:35,410 Ejaaz: Is my math right? Yeah, a fifth of the world's leading 416 00:24:35,980 --> 00:24:40,260 Ejaaz: AI lab, which is pretty crazy. Now, if we look into the stats of this, 417 00:24:40,460 --> 00:24:45,840 Ejaaz: this is a five gigawatt deal, which is more than any single data center that's currently live. 418 00:24:46,320 --> 00:24:52,080 Ejaaz: It's actually a multiple of five. I think SpaceX AI's Colossus 2 is the largest 419 00:24:52,080 --> 00:24:53,480 Ejaaz: right now with their 1 million TB. 420 00:24:53,620 --> 00:24:57,180 Ejaaz: So it's going to be 5X larger than the average data center that we're seeing 421 00:24:57,180 --> 00:24:58,560 Ejaaz: right now for AI specifically. 422 00:24:58,820 --> 00:25:01,900 Ejaaz: And they're aiming to get one gigawatt online by the end of the year. 423 00:25:01,980 --> 00:25:05,660 Ejaaz: Now, Now, the reason why this is so good for both teams is Anthropic already 424 00:25:05,660 --> 00:25:09,720 Ejaaz: has a close relationship with AWS and Amazon's cloud computing department. 425 00:25:09,920 --> 00:25:13,280 Ejaaz: So spinning up more compute clusters is gonna be so easy for them. 426 00:25:13,320 --> 00:25:14,280 Ejaaz: They have a working relationship. 427 00:25:14,480 --> 00:25:17,820 Ejaaz: They're used to training cloud models on this, so it shouldn't be too hard to 428 00:25:17,820 --> 00:25:19,460 Ejaaz: ramp this up. If you're Amazon, 429 00:25:20,360 --> 00:25:23,940 Ejaaz: Welcome back. That $5 billion is going to come right back to you. 430 00:25:24,060 --> 00:25:27,620 Ejaaz: So I don't know what kind of like circle economy this is, but it's back and 431 00:25:27,620 --> 00:25:28,800 Ejaaz: it's very impressive for them. 432 00:25:29,300 --> 00:25:32,980 Josh: Is it ironic that today Amazon hit an all-time high? No, maybe, maybe not. 433 00:25:33,080 --> 00:25:34,400 Ejaaz: I'm holding stock. I got the stock. 434 00:25:34,520 --> 00:25:37,540 Josh: Clearly, clearly they're doing something right. Amazon is a phenomenal company. 435 00:25:37,680 --> 00:25:38,900 Josh: They're the largest shareholder in Anthropic. 436 00:25:39,140 --> 00:25:42,360 Josh: It's hard not to be bullish on them. It's hard not to be bullish on the accelerated 437 00:25:42,360 --> 00:25:45,040 Josh: computing stack. And I think that's probably what Jensen is getting nervous 438 00:25:45,040 --> 00:25:46,620 Josh: about. That's why NVIDIA is pushing open source. 439 00:25:46,820 --> 00:25:50,200 Josh: And the good news is, is he has some help. He has some assistance. 440 00:25:50,360 --> 00:25:54,300 Josh: From the folks overseas in China who have been pumping out unbelievable models 441 00:25:54,300 --> 00:25:58,400 Josh: all week long as it relates to Kimi and Quen, our Chinese favorites. 442 00:25:58,780 --> 00:26:04,300 Josh: We have Kimi K2.6 and Quen 3.6. There's a lot of digits and numbers. 443 00:26:04,500 --> 00:26:09,520 Josh: All you need to know is that the best open source models in the world didn't exist last week. 444 00:26:09,640 --> 00:26:12,760 Josh: They now exist this week and they are better at pretty much everything, 445 00:26:12,920 --> 00:26:14,580 Josh: but exceptional at coding. 446 00:26:14,760 --> 00:26:18,680 Josh: In fact, word on the street is that some of these models are as good as GPT 447 00:26:18,680 --> 00:26:21,260 Josh: 5.4 was and only a few points off of Claude. 448 00:26:21,460 --> 00:26:23,460 Josh: I mean, these are pretty amazing open source models that, again, 449 00:26:23,640 --> 00:26:27,820 Josh: are free to run locally on your machine if you have the machine capability of doing so. 450 00:26:28,020 --> 00:26:30,160 Josh: That's a big, that's a big game changer. 451 00:26:30,420 --> 00:26:33,960 Ejaaz: Okay, so typically the story we tell with these open source models is, 452 00:26:34,080 --> 00:26:35,940 Ejaaz: wow, aren't they so amazing? 453 00:26:36,240 --> 00:26:39,800 Ejaaz: Yeah, they're the good younger brother. They're not as good as the Frontier AI Labs. 454 00:26:40,060 --> 00:26:45,120 Ejaaz: That completely changed this week. So Kimi K 2.6 is the latest model from a 455 00:26:45,120 --> 00:26:49,160 Ejaaz: Chinese lab called Moonshot Labs. I believe it's Moonshot or Moonshot AI. 456 00:26:49,500 --> 00:26:54,080 Ejaaz: And they released their model, which ends up being as good as coding or at coding. 457 00:26:54,540 --> 00:26:57,620 Ejaaz: As opus 4.7 and it's 100 open source 458 00:26:57,620 --> 00:27:00,380 Ejaaz: like you mentioned josh which means that maybe you could run 459 00:27:00,380 --> 00:27:03,200 Ejaaz: this on a local device now the answer that you would typically get back from 460 00:27:03,200 --> 00:27:05,960 Ejaaz: this is hey like listen it's uh it's too 461 00:27:05,960 --> 00:27:09,420 Ejaaz: large to run on my laptop and that is true but with the latest quen model which 462 00:27:09,420 --> 00:27:16,060 Ejaaz: is a 3.6 version you can run it as an 18 gigabyte sized model slightly quantized 463 00:27:16,060 --> 00:27:20,920 Ejaaz: on your laptop today so the point that i want to make about these models isn't 464 00:27:20,920 --> 00:27:23,460 Ejaaz: exactly the specifics but across all benchmarks, 465 00:27:23,700 --> 00:27:27,240 Ejaaz: it's not as good as the Frontier AI Labs, but it's a few points. 466 00:27:27,480 --> 00:27:31,860 Ejaaz: That difference and gap has closed massively over the last couple of months, 467 00:27:31,880 --> 00:27:33,200 Ejaaz: which tells me two things. 468 00:27:33,340 --> 00:27:37,220 Ejaaz: Number one, China has figured out some kind of groundbreaking way to train their 469 00:27:37,220 --> 00:27:40,340 Ejaaz: models that they haven't told the West about, and they're going to keep it closed 470 00:27:40,340 --> 00:27:43,900 Ejaaz: guard and eventually close source their model releases going forwards. 471 00:27:44,240 --> 00:27:47,880 Ejaaz: And number two, they've figured out a new way to use inference to their benefit. 472 00:27:48,060 --> 00:27:51,760 Ejaaz: Like one thing I'm going to highlight here is this new Kimi K 2.6 model can 473 00:27:51,760 --> 00:27:56,160 Ejaaz: code continuously for 12 hours straight using 300 agents. 474 00:27:56,440 --> 00:28:00,360 Ejaaz: So the unlock here isn't one model itself. It's spitting up 300 versions of 475 00:28:00,360 --> 00:28:02,340 Ejaaz: itself and getting it to attack the problem. 476 00:28:02,460 --> 00:28:05,300 Ejaaz: That's something Sam realized and what he's implementing in 5.5. 477 00:28:05,520 --> 00:28:09,140 Ejaaz: That's something Opus 4.7 realized and is doing probably similarly with Mythos. 478 00:28:09,400 --> 00:28:12,820 Ejaaz: So I have this question here, which is like, how do they have to try to do this? 479 00:28:12,980 --> 00:28:16,500 Ejaaz: Well, I think every three months that there's a new open model that gets released, 480 00:28:16,680 --> 00:28:19,360 Ejaaz: they're making these jumps because they're using these models to train themselves 481 00:28:19,360 --> 00:28:23,960 Ejaaz: we proved that with kimmy k 2.5 there's too many two point whatevers um and 482 00:28:23,960 --> 00:28:27,180 Ejaaz: the same thing is happening with quen it's just all around pretty amazing stuff um 483 00:28:27,180 --> 00:28:31,120 Josh: Yeah chen is crushing okay so before we go we have two quick things to hit the 484 00:28:31,120 --> 00:28:34,720 Josh: first being one that we missed last week which we need to touch on quickly anthropic 485 00:28:34,720 --> 00:28:38,120 Josh: has a design tool now if you are a designer if you are interested in building 486 00:28:38,120 --> 00:28:43,800 Josh: web pages videos graphics slideshows pitch decks any type of visual asset you're 487 00:28:43,990 --> 00:28:47,030 Josh: claude now has an entire design suite built just 488 00:28:47,030 --> 00:28:50,330 Josh: for this purpose it's called claude design it exists separately 489 00:28:50,330 --> 00:28:53,610 Josh: you can access it through the desktop app or on your browser and 490 00:28:53,610 --> 00:28:56,290 Josh: it basically allows you to build visual assets in a way that 491 00:28:56,290 --> 00:28:59,330 Josh: you couldn't previously previously with claude you had artifacts an 492 00:28:59,330 --> 00:29:02,030 Josh: artifact you could generate something dynamic it could kind of 493 00:29:02,030 --> 00:29:05,910 Josh: build you a web page this takes it to a whole new level you could generate wireframes 494 00:29:05,910 --> 00:29:09,670 Josh: if you want to try it to use less tokens you could fill it out and create properly 495 00:29:09,670 --> 00:29:13,070 Josh: created prototypes that are actually clickable it's amazing the video we're 496 00:29:13,070 --> 00:29:16,770 Josh: seeing on screen highlights a few of them unfortunately there was a big loser 497 00:29:16,770 --> 00:29:21,690 Josh: in this because uh this sounds like a lot of what that little design company named figma does. 498 00:29:21,690 --> 00:29:22,750 Ejaaz: Yeah the little company 499 00:29:22,750 --> 00:29:26,650 Josh: Stock market did not love the reaction to that did it nope. 500 00:29:26,650 --> 00:29:31,690 Ejaaz: Nope it is down almost 20 on the week um i actually tracked the stock price 501 00:29:31,690 --> 00:29:34,910 Ejaaz: after the announcement was made so like it wasn't even readily available it 502 00:29:34,910 --> 00:29:38,370 Ejaaz: was literally just the tweet 20 minutes after it was tweeted the stock was down 503 00:29:38,370 --> 00:29:40,570 Ejaaz: six percent so So the point being, 504 00:29:40,710 --> 00:29:43,030 Ejaaz: whether this is market speculation or not, like, listen, 505 00:29:43,470 --> 00:29:45,530 Ejaaz: Claude Design isn't as good as Figma. 506 00:29:45,650 --> 00:29:48,150 Ejaaz: They're working with a few of these different partners, such as Canva. 507 00:29:49,280 --> 00:29:55,460 Ejaaz: Two weeks ago, one of Anthropic's former most execs left the board of Figma. 508 00:29:55,920 --> 00:29:59,280 Ejaaz: And the rumors was because they were building a competitor. So it's pretty clear. 509 00:29:59,500 --> 00:30:02,220 Ejaaz: Anthropic is going off to every single sector, whether you're a designer, 510 00:30:02,620 --> 00:30:05,460 Ejaaz: a software engineer, a mathematician, a research scientist, doesn't matter. 511 00:30:05,680 --> 00:30:07,920 Ejaaz: They're going off to everything because the model is applicable to everything. 512 00:30:08,120 --> 00:30:11,520 Ejaaz: And I don't know what this means for certain modes that companies like Figma 513 00:30:11,520 --> 00:30:14,000 Ejaaz: holds, but it's certainly going to affect the stock price. 514 00:30:14,160 --> 00:30:16,900 Josh: Can you do me a favor and click the max button real quick for me just to show the chart? 515 00:30:17,580 --> 00:30:17,980 Ejaaz: Oh! 516 00:30:19,380 --> 00:30:23,540 Josh: Yeah, minus 86% since IPO for those who are not watching on screen. 517 00:30:23,540 --> 00:30:26,360 Josh: It's been a pretty bad, rough run for Figma. 518 00:30:27,300 --> 00:30:31,540 Ejaaz: We have to start naming Anthropic the stock killer, Josh. This is like every 519 00:30:31,540 --> 00:30:33,340 Ejaaz: single tweet is tanking a stock. 520 00:30:33,380 --> 00:30:35,980 Josh: No, it's tough. It's brutal. We had one last thing that you wanted to mention. 521 00:30:36,000 --> 00:30:38,560 Josh: I know. We got to end on this strong. What do we have? 522 00:30:38,780 --> 00:30:44,720 Ejaaz: How good is your accent or impersonation of your president, of our president, Josh? 523 00:30:45,000 --> 00:30:45,580 Josh: Pretty horrible. 524 00:30:46,060 --> 00:30:47,920 Ejaaz: Not good. Okay, well, then we're not going to attempt it. 525 00:30:47,920 --> 00:30:50,940 Josh: I'd love to hear your British take on it, though, if you're feeling ambitious. 526 00:30:51,280 --> 00:30:56,760 Ejaaz: Okay, so my British take on this is, this is, albeit hilarious and somewhat 527 00:30:56,760 --> 00:30:59,620 Ejaaz: terrifying, that the President of the United States is saying this. 528 00:31:00,020 --> 00:31:04,180 Ejaaz: He commented, okay, on the government's relationship with Anthropic. 529 00:31:04,300 --> 00:31:06,020 Ejaaz: Now, if you're wondering why on earth he's commenting on it, 530 00:31:06,220 --> 00:31:07,840 Ejaaz: they're going to be releasing this cold mythos model. 531 00:31:08,000 --> 00:31:11,240 Ejaaz: It might be a security risk. It's probably good for the government to have access 532 00:31:11,240 --> 00:31:13,000 Ejaaz: to this thing and prepare necessarily. 533 00:31:13,520 --> 00:31:16,980 Ejaaz: The government has been having very important conversations with bankers and 534 00:31:16,980 --> 00:31:19,040 Ejaaz: governments all around the world, just try and figure out, you know, 535 00:31:19,120 --> 00:31:20,080 Ejaaz: how best to prepare for this. 536 00:31:20,740 --> 00:31:24,940 Ejaaz: And after having an in-depth discussion with Dario Amore, which by the way, 537 00:31:25,080 --> 00:31:28,900 Ejaaz: he blacklisted that CEO and Anthropic entirely from the government using it. 538 00:31:29,020 --> 00:31:31,940 Ejaaz: He's now rekindling it and saying, maybe there's a deal on the line. 539 00:31:32,020 --> 00:31:34,020 Ejaaz: He goes, and I quote, I'm not going to do the accent. 540 00:31:34,640 --> 00:31:37,820 Ejaaz: We'll get along with Anthropic just fine. Trump said on CNBC. 541 00:31:38,940 --> 00:31:43,340 Josh: We'll get along with Anthropic just fine. I think they can be of great use to us. 542 00:31:43,860 --> 00:31:46,700 Josh: They're high IQ people. Very good. Very good. They tend to be on the left, 543 00:31:46,820 --> 00:31:48,800 Josh: radical left, but We get along with them. 544 00:31:49,530 --> 00:31:51,650 Josh: I don't know. That's all I got. But that is what he said. 545 00:31:51,670 --> 00:31:54,370 Ejaaz: Were you practicing that? That was actually pretty good. I was practicing my head. 546 00:31:54,510 --> 00:31:54,970 Josh: I was rehearsing. 547 00:31:55,110 --> 00:31:58,950 Ejaaz: Damn. I closed my eyes whilst you were doing that, whilst I was laughing. Did it feel right? 548 00:31:59,050 --> 00:31:59,850 Josh: It sounded like him? Good. 549 00:32:00,510 --> 00:32:02,970 Ejaaz: It channeled his spirit. It was there. It was a good effort. 550 00:32:03,990 --> 00:32:07,250 Ejaaz: But I believe that's it. That is the end of the roundup. 551 00:32:07,710 --> 00:32:12,710 Ejaaz: Josh and I are recording this. FYI, it's 4 p.m. over here. Typically, we're morning birds. 552 00:32:12,930 --> 00:32:17,110 Ejaaz: We deliver this in the morning, but we waited for the announcement of Spud GPT 553 00:32:17,110 --> 00:32:22,170 Ejaaz: 5.5 just for you guys. and we're going to be bringing you the cutting edge news every single week. 554 00:32:22,310 --> 00:32:25,710 Ejaaz: As Josh mentioned, we had three other amazing episodes that we filmed earlier 555 00:32:25,710 --> 00:32:28,390 Ejaaz: this week. Definitely go check them out. They're all each 20 minute song. 556 00:32:28,470 --> 00:32:29,170 Ejaaz: It's your commute to work. 557 00:32:29,370 --> 00:32:31,890 Ejaaz: It's your gym session if you're not that active. 558 00:32:32,190 --> 00:32:35,070 Ejaaz: Definitely go check it out and let us know what you think. But yeah, 559 00:32:35,150 --> 00:32:35,930 Ejaaz: Josh, any final thoughts? 560 00:32:36,050 --> 00:32:39,270 Josh: Call me crazy, but I like the afternoon recordings. I got good energy. 561 00:32:39,490 --> 00:32:42,310 Josh: I'm like woken up. I'm 100% right now. I'm rocking and rolling. 562 00:32:42,810 --> 00:32:45,210 Josh: I'm feeling good. So I don't know. Maybe we'll have to lean into this a little 563 00:32:45,210 --> 00:32:46,370 Josh: bit more, but that's everything. 564 00:32:46,550 --> 00:32:48,870 Josh: If you've made it this far, if you're still listening to this and you've heard 565 00:32:48,870 --> 00:32:51,130 Josh: our other episodes, you're caught up. You're done for the week. 566 00:32:51,250 --> 00:32:52,690 Josh: You can go touch grass. Enjoy your weekend. 567 00:32:52,990 --> 00:32:55,470 Josh: There will be a lot more to talk about next weekend. But for now, 568 00:32:55,490 --> 00:33:00,830 Josh: you have fully synchronized with all of the chaos happening on the frontier of AI and technology. 569 00:33:01,150 --> 00:33:03,370 Josh: Thank you so much for watching. As always, we very much appreciate it. 570 00:33:03,430 --> 00:33:05,910 Josh: If you enjoyed this episode or any of our previous episodes from this week, 571 00:33:06,210 --> 00:33:09,470 Josh: don't forget to share them with a friend who you also might enjoy it, possibly. 572 00:33:09,670 --> 00:33:12,030 Josh: We have a newsletter on Substack that goes live twice a week. 573 00:33:12,210 --> 00:33:13,970 Josh: Just went live yesterday, going live again tomorrow. 574 00:33:14,350 --> 00:33:17,170 Josh: The Friday issue is a recap of everything that happens this week, 575 00:33:17,250 --> 00:33:18,350 Josh: which is always fun and exciting. 576 00:33:18,490 --> 00:33:21,350 Josh: In fact, I'm gonna go write that as soon as we finish this episode. 577 00:33:21,490 --> 00:33:22,490 Josh: So thank you all for watching. 578 00:33:22,650 --> 00:33:25,490 Josh: As always, don't forget to subscribe, like, comment, all the good things, 579 00:33:25,530 --> 00:33:26,690 Josh: and we will see you guys next week.