1 00:00:00,100 --> 00:00:04,120 Ejaaz: Four days ago, Anthropic asked the world's leading AI labs to slow down their 2 00:00:04,120 --> 00:00:09,720 Ejaaz: AI research out of fear that the AI models were getting so good that they would escape human control. 3 00:00:09,940 --> 00:00:13,940 Ejaaz: Now, just yesterday, the same company released the most powerful model the world 4 00:00:13,940 --> 00:00:18,440 Ejaaz: has ever seen, but it comes with a twist. It's one model, but there's two versions of it. 5 00:00:18,750 --> 00:00:21,560 Ejaaz: Claude Fable 5 is the model that everyone gets to use. 6 00:00:21,560 --> 00:00:24,870 Ejaaz: It's a class that sits above Opus, but it's heavily restricted. 7 00:00:24,870 --> 00:00:28,480 Ejaaz: It comes with a lot of safeguards around it because version 2, 8 00:00:28,480 --> 00:00:34,310 Ejaaz: called Mythos 5 is the unrestricted version which poses itself as a higher cybersecurity 9 00:00:34,310 --> 00:00:38,570 Ejaaz: risk than its predecessor, Mythos Preview. It also 10 00:00:39,010 --> 00:00:43,870 Ejaaz: excels at creating biological compounds, which could potentially be used as 11 00:00:43,870 --> 00:00:48,690 Ejaaz: bioweapons. So it's only accessible by vetted partners that Anthropic approves. 12 00:00:48,940 --> 00:00:53,200 Ejaaz: So this brings us to a fork in the road where the smartest, most intelligent 13 00:00:53,200 --> 00:00:56,190 Ejaaz: AI model isn't accessible to everyone. 14 00:00:56,440 --> 00:00:59,800 Ejaaz: Typically, for the entire history of software, you'd be able to pay for access 15 00:00:59,800 --> 00:01:02,490 Ejaaz: to the same level of access that an institutional would. 16 00:01:02,830 --> 00:01:06,050 Ejaaz: But this time, the new metric isn't how smart is your AI model, 17 00:01:06,050 --> 00:01:09,750 Ejaaz: it's the gap between how smart is your AM model that you get access to and the 18 00:01:09,750 --> 00:01:11,620 Ejaaz: ones others also get access to. 19 00:01:11,620 --> 00:01:16,180 Josh: Well, how amazing is it that we have Mythos now available in our apps today? 20 00:01:16,620 --> 00:01:20,660 Josh: I think that's the really exciting takeaway is not only is it better than the 21 00:01:20,660 --> 00:01:23,090 Josh: Mythos preview, but it's available inside of your app. 22 00:01:23,090 --> 00:01:27,320 Josh: So if you are listening to this, you can go and try CloudFable 5 right now with 23 00:01:27,320 --> 00:01:28,230 Josh: those two restrictions. 24 00:01:28,230 --> 00:01:31,120 Josh: It won't go anywhere near biology. It won't go anywhere near cybersecurity, 25 00:01:31,350 --> 00:01:35,840 Josh: but this is the best model in the world and it is now available to everyone 26 00:01:35,840 --> 00:01:39,780 Josh: to go try out and it's like every once in a while there there's this new technology that 27 00:01:40,150 --> 00:01:43,990 Josh: kind of forces you to to reframe how you engage with the technology and i think 28 00:01:43,990 --> 00:01:46,230 Josh: that's been my experience so far using fable 29 00:01:46,580 --> 00:01:50,490 Josh: is that it's so different than any other model we've used that it kind of forces 30 00:01:50,490 --> 00:01:55,820 Josh: you to reframe how you engage with them and to showcase that i want us to go through some demos of 31 00:01:56,210 --> 00:02:00,690 Josh: attempts that people have done to kind of showcase the powers and capabilities 32 00:02:00,690 --> 00:02:05,760 Josh: of fable 5 this new frontier model starting with a really bizarre demo in which 33 00:02:05,760 --> 00:02:11,000 Josh: uh dan shipper who is a prominent poster on x he recreated the library of babble 34 00:02:11,370 --> 00:02:14,610 Josh: and help me understand what's going on here you just because i'm seeing a lot 35 00:02:14,610 --> 00:02:19,090 Josh: of visuals um and i understand that it's not a image or video generating model 36 00:02:19,290 --> 00:02:22,190 Josh: so how is it able to create a real world's, 37 00:02:22,660 --> 00:02:24,920 Josh: emulation so accurate so quickly? 38 00:02:25,390 --> 00:02:31,140 Ejaaz: So the greatest part about this is it took one prompt and he basically asked 39 00:02:31,140 --> 00:02:36,540 Ejaaz: it to read a book, the book of Babel or whatever the title of the book is. 40 00:02:36,770 --> 00:02:38,260 Ejaaz: And he said, I want you to read this book. 41 00:02:38,580 --> 00:02:41,840 Ejaaz: And then I want you to recreate one of the concepts that is described throughout 42 00:02:41,840 --> 00:02:45,070 Ejaaz: the book, which is this library of Babel or Babel. 43 00:02:45,460 --> 00:02:48,640 Ejaaz: And it did so in just under an hour, I believe. 44 00:02:48,850 --> 00:02:54,030 Ejaaz: And if you look visually on the screen, it is this high fidelity 3D representation 45 00:02:54,030 --> 00:02:57,720 Ejaaz: of what this library looks like. And it's infinite. So you'll notice in this 46 00:02:57,720 --> 00:03:03,090 Ejaaz: video that he looks down, he looks up, and it's just infinite books he can access himself. 47 00:03:03,300 --> 00:03:06,900 Ejaaz: And he even asked it to include some of the essays that he's written himself. 48 00:03:06,900 --> 00:03:11,490 Ejaaz: This guy authors a bunch of analysis on AI, and he pulls open a piece that he 49 00:03:11,490 --> 00:03:14,970 Ejaaz: wrote that is in the Library of Babel. The idea of the library is that it contains 50 00:03:14,970 --> 00:03:19,010 Ejaaz: every piece of text that has been ever written. And so he gets access to it. 51 00:03:19,010 --> 00:03:19,920 Ejaaz: It's a pretty cool example. 52 00:03:20,580 --> 00:03:24,350 Josh: Yeah, that's super cool. The other ones that I've really noticed that it accelerated 53 00:03:24,350 --> 00:03:29,140 Josh: in is kind of 3D world building, which is funny because you don't think of Anthropic 54 00:03:29,140 --> 00:03:32,120 Josh: as a world building model. It's not a world model. 55 00:03:32,390 --> 00:03:35,020 Josh: In fact, it doesn't have image generation capabilities. In fact, 56 00:03:35,020 --> 00:03:36,740 Josh: it doesn't have video generation capabilities. 57 00:03:36,740 --> 00:03:40,800 Josh: So how is it creating all this realistic visual design assets? 58 00:03:40,800 --> 00:03:45,090 Josh: And it's just really good at math. And this begs the question is how important 59 00:03:45,090 --> 00:03:49,530 Josh: is it to focus on image gen on video gen if kind of at its core it could do 60 00:03:49,530 --> 00:03:52,230 Josh: all of this with math and what we're seeing on screen here is how. 61 00:03:52,970 --> 00:03:58,310 Josh: A virtual one-to-one recreation of Yosemite National Park that was done with 62 00:03:58,310 --> 00:04:01,640 Josh: a simple prompt asking it to create a recreation of it. 63 00:04:01,850 --> 00:04:05,660 Josh: And then the model was smart enough to go off and understand the context required 64 00:04:05,660 --> 00:04:08,330 Josh: in order to build an accurate representation and pulled that off. 65 00:04:08,550 --> 00:04:13,700 Josh: It did things like it scanned the satellite imagery to figure out what the elevations 66 00:04:13,700 --> 00:04:17,680 Josh: were like. It pulled topographic maps that it found to figure out specifically what the heights were. 67 00:04:17,930 --> 00:04:21,650 Josh: It found imagery that any imagery that it could find about the park so that 68 00:04:21,650 --> 00:04:24,810 Josh: it can reference it and show you i mean look at this resuming in on a waterfall 69 00:04:24,810 --> 00:04:27,270 Josh: it feels like it's a one-to-one replica, 70 00:04:27,920 --> 00:04:31,310 Josh: lower fidelity but something that can run inside of your browser it's really 71 00:04:31,310 --> 00:04:34,850 Josh: impressive how far the model can go on one prompt and i think that's one of 72 00:04:34,850 --> 00:04:39,830 Josh: the places in which fable stands out in particular is its ability to to reason through 73 00:04:40,190 --> 00:04:44,850 Josh: your requests in a way that hasn't been done before we filmed an episode yesterday 74 00:04:44,990 --> 00:04:48,490 Josh: that i would highly recommend listening to about how we've kind of progressed, 75 00:04:49,170 --> 00:04:52,220 Josh: with our interaction of the model kind of moving up the extraption layer, 76 00:04:52,420 --> 00:04:55,060 Josh: where first we engage with models, then agents, then harnesses. 77 00:04:55,320 --> 00:04:57,920 Josh: Now we're just creating these loops where the agent and the. 78 00:04:58,750 --> 00:05:00,120 Josh: Underlying intelligence is smart 79 00:05:00,120 --> 00:05:03,110 Josh: enough to actually understand what's required to get you to your goal. 80 00:05:03,400 --> 00:05:06,080 Josh: And I think that's what this is such a great example of this Yosemite one in 81 00:05:06,080 --> 00:05:09,630 Josh: particular is, hey, I want you to recreate Yosemite for me so that way I can 82 00:05:09,630 --> 00:05:12,030 Josh: fly around and I can enjoy it in a one-to-one replica. 83 00:05:12,330 --> 00:05:15,760 Josh: And it does all the rest for you. And I think that level of critical thinking 84 00:05:15,760 --> 00:05:19,530 Josh: is something that's novel with Fable 5 that we've never seen in any other model before. 85 00:05:19,530 --> 00:05:23,550 Ejaaz: I mean, the breakthrough that we're talking about is visual and spatial reasoning. 86 00:05:23,550 --> 00:05:27,590 Ejaaz: And I think it's important to explain the difference between this and another 87 00:05:27,910 --> 00:05:31,490 Ejaaz: favorite version of a model that we speak about a lot on this show, which is world models. 88 00:05:32,130 --> 00:05:35,590 Ejaaz: Typically with world models, it recreates the physical world around us. 89 00:05:35,590 --> 00:05:39,630 Ejaaz: But most importantly, it understands that physical reality, understands how 90 00:05:39,630 --> 00:05:42,990 Ejaaz: gravity works, it understands how different forces of nature works, 91 00:05:43,220 --> 00:05:46,990 Ejaaz: and it applies it when an object has an action. So let's say you kind of like 92 00:05:47,290 --> 00:05:50,270 Ejaaz: punch a puddle of water, it splashes, the droplets come up. 93 00:05:50,570 --> 00:05:54,170 Ejaaz: This isn't exactly the same thing. It's still based on theory. 94 00:05:54,170 --> 00:05:58,910 Ejaaz: This is still an LLM that ingests a lot of text and understands kind of like 95 00:05:59,090 --> 00:06:03,590 Ejaaz: how the physics works in theory and then recreates what its version of it might 96 00:06:03,590 --> 00:06:05,610 Ejaaz: be. And this is what we're looking at on screen. It's kind of like, 97 00:06:05,610 --> 00:06:08,210 Ejaaz: it's known as spatial reasoning or visual intelligence. 98 00:06:08,770 --> 00:06:12,320 Ejaaz: It's close to the thing, but it's not quite the same thing. Now, 99 00:06:12,510 --> 00:06:16,030 Ejaaz: another example that I really enjoyed was from Ethan Moloch. 100 00:06:16,550 --> 00:06:20,090 Ejaaz: Ethan Moloch is one of my favorite AI researchers that analyzes a lot of these 101 00:06:20,090 --> 00:06:23,830 Ejaaz: new models, but he builds or tests it in really interesting ways. 102 00:06:24,070 --> 00:06:26,650 Ejaaz: One of these ways was he rebuilt Snake. 103 00:06:27,270 --> 00:06:31,300 Ejaaz: Now, I am kind of ashamed to admit how much time I spent playing this particular 104 00:06:31,300 --> 00:06:34,530 Ejaaz: game only because it was like the best version of Snake that we see. 105 00:06:34,530 --> 00:06:38,240 Ejaaz: And like I'm showing you on the screen right now, it's like incredibly high 106 00:06:38,240 --> 00:06:42,880 Ejaaz: fidelity. It looks way better than the game I used to play on the Nokia that I had as a kid. 107 00:06:43,110 --> 00:06:46,610 Ejaaz: But the point is, it's pretty cool. It introduces new power-ups and obviously 108 00:06:46,610 --> 00:06:48,300 Ejaaz: like you know you can die in usual things 109 00:06:48,880 --> 00:06:54,070 Ejaaz: Also, forget about creating the game. Claude Fable 5 is really good at playing 110 00:06:54,070 --> 00:06:58,130 Ejaaz: the game itself. What you're seeing is an accelerated version of it playing 111 00:06:58,130 --> 00:07:02,330 Ejaaz: Pokemon Fire Red, and it completes the game in, I believe, 50 minutes. 112 00:07:02,540 --> 00:07:07,490 Ejaaz: And the way that it works is it takes screenshots of the game at any single 113 00:07:07,490 --> 00:07:11,540 Ejaaz: point, and it basically makes a decision as to which button it wants to click, 114 00:07:11,540 --> 00:07:15,120 Ejaaz: which step it wants to take, and it is just kind of like spread through a software 115 00:07:15,120 --> 00:07:16,450 Ejaaz: run where it's able to do that. 116 00:07:16,450 --> 00:07:20,180 Ejaaz: Now, as a kid growing up and watching, you know, a lot of Pokemon, 117 00:07:20,180 --> 00:07:23,520 Ejaaz: playing a lot of Pokemon, trading the cards, this is kind of nostalgic, 118 00:07:23,520 --> 00:07:26,460 Ejaaz: but also kind of scary. We were joking before we started recording, 119 00:07:26,660 --> 00:07:29,480 Ejaaz: Josh plays Cod quite a bit or plays a lot of computer games. 120 00:07:29,790 --> 00:07:33,140 Ejaaz: And I wonder about the time in the near future where you're going to be one 121 00:07:33,140 --> 00:07:37,040 Ejaaz: of the buying a AI agent and it might actually be better than you. 122 00:07:37,400 --> 00:07:40,580 Josh: Yeah, that's going to be a little traumatic for me on a personal note. 123 00:07:40,980 --> 00:07:44,130 Josh: Just hurting my ego that I'm losing my game to an AI. 124 00:07:44,320 --> 00:07:47,260 Josh: I saw another great example on the YouTube channel, actually, 125 00:07:47,260 --> 00:07:50,690 Josh: where they were Mythos or Fable 5 was playing Factorio. 126 00:07:50,950 --> 00:07:53,890 Josh: And Factorio is a game that I really enjoy, that I've been playing for a long time. 127 00:07:54,210 --> 00:07:56,910 Josh: And it was doing it very, very well. And I'm like, oh, dude, 128 00:07:56,910 --> 00:07:59,260 Josh: you're getting a little too close to home with this. I don't love it. 129 00:07:59,600 --> 00:08:04,980 Josh: But it's incredibly capable. And we have another example here that shows a use case that is not a game. 130 00:08:04,980 --> 00:08:08,010 Josh: Instead, it is a it's so cool. 131 00:08:08,010 --> 00:08:12,090 Josh: So there's this guy named Todd Saunders. He's on X and he he posts this tweet 132 00:08:12,090 --> 00:08:16,720 Josh: saying fable slash mythos is unbelievable was on a customer call today and had 133 00:08:16,720 --> 00:08:20,770 Josh: Claude transcribing in the background and on screen we're showing a visual of what that looks like. 134 00:08:21,250 --> 00:08:23,910 Josh: As they were telling me about the features they wish their current software 135 00:08:23,910 --> 00:08:27,140 Josh: had, Claude was building the features in real time. 136 00:08:27,540 --> 00:08:30,990 Josh: By the end of the call, I was able to show a fully working product with the 137 00:08:30,990 --> 00:08:33,930 Josh: exact workflow they mentioned 15 minutes earlier. 138 00:08:34,300 --> 00:08:37,540 Josh: Autonomous looped building triggers from a customer call. 139 00:08:37,720 --> 00:08:41,620 Josh: And this is one of the most amazing things about the model is that it's able 140 00:08:41,620 --> 00:08:45,380 Josh: to go off and do a lot of the hard work yourself, where it feels like recently 141 00:08:45,380 --> 00:08:49,640 Josh: you've had to kind of be in the loop. You had to continue to prompt the agent to give it more context. 142 00:08:50,000 --> 00:08:55,070 Josh: And with this model, it's very easy to give it a goal and give it a verifiable 143 00:08:55,070 --> 00:08:56,680 Josh: outcome that it can match against the goal. 144 00:08:57,020 --> 00:08:59,830 Josh: And then it will just go off and do those things. So as is on the customer call, 145 00:09:00,060 --> 00:09:03,370 Josh: like how cool is that for a salesperson where you're listening to customers, 146 00:09:03,370 --> 00:09:05,940 Josh: you're listening to complaints and in real time, you're fixing their problems. 147 00:09:06,230 --> 00:09:07,710 Josh: You're building new software on top of it. 148 00:09:08,410 --> 00:09:11,590 Josh: It's unbelievably capable. This was one of the demos that I found most interesting too. 149 00:09:11,860 --> 00:09:14,960 Josh: And then this final example, it's just fun for networking nerds or just computer 150 00:09:14,960 --> 00:09:19,000 Josh: science nerds in general. We're seeing a highway on screen with cars and buses and vans. 151 00:09:19,320 --> 00:09:22,990 Josh: And those cars are not random. They are actually associated with specific packets 152 00:09:22,990 --> 00:09:26,180 Josh: that are being pushed across the network. So it's a really fun and interactive 153 00:09:26,180 --> 00:09:28,210 Josh: way to visualize network traffic. 154 00:09:28,470 --> 00:09:32,160 Josh: And I could imagine this being great for a lot of educational purposes. 155 00:09:32,370 --> 00:09:36,460 Josh: And one of the things that I did notice also on Snake Ejaz is that there was audio, there was sound. 156 00:09:36,570 --> 00:09:40,520 Josh: It actually generated sounds and audio too. So there's a lot of modalities in 157 00:09:40,520 --> 00:09:42,210 Josh: which it's actually performing pretty well. 158 00:09:43,110 --> 00:09:46,100 Ejaaz: Can I add a bonus Easter egg, which we didn't see on the demo, 159 00:09:46,100 --> 00:09:48,280 Ejaaz: but I know because I played it too much yesterday. 160 00:09:48,580 --> 00:09:54,250 Ejaaz: There are power-ups that pop up, but the power-ups are software patches. 161 00:09:54,500 --> 00:09:59,670 Ejaaz: So if you don't hit the software patch power-up, you end up with a random error 162 00:09:59,900 --> 00:10:04,130 Ejaaz: in the game that you must avoid, like a random wormhole that could like suck 163 00:10:04,130 --> 00:10:05,640 Ejaaz: you out of the game and like cause you to lose. 164 00:10:05,940 --> 00:10:09,750 Ejaaz: So like it's coding in real time but you could fix it in real time and if you 165 00:10:09,750 --> 00:10:13,280 Ejaaz: hit the power up it it like implements the code fix like immediately it was 166 00:10:13,280 --> 00:10:16,790 Ejaaz: just that's very nerdy but like very cool and very creative not something that 167 00:10:16,790 --> 00:10:19,510 Ejaaz: we've seen before now if you want to like move away from the 168 00:10:19,970 --> 00:10:24,210 Ejaaz: The toy aspect or the retail adoption of how you can use this particular model 169 00:10:24,480 --> 00:10:28,500 Ejaaz: there's of course a lot of enterprise use cases and the number one version of 170 00:10:28,650 --> 00:10:32,430 Ejaaz: enterprises using Claude is through code specifically and there were a few examples 171 00:10:32,430 --> 00:10:35,660 Ejaaz: that were included in the official blog my favorite one was Stripe, 172 00:10:36,010 --> 00:10:40,870 Ejaaz: who did a code migration of 50 million lines of Ruby, which typically would 173 00:10:40,870 --> 00:10:44,940 Ejaaz: take around two months and several software engineering teams. 174 00:10:45,360 --> 00:10:50,170 Ejaaz: It took them less than a day using Fable. I believe they used two to three instances 175 00:10:50,170 --> 00:10:53,220 Ejaaz: of Fable, but still, that's like two to three software engineers that kind of 176 00:10:53,220 --> 00:10:57,150 Ejaaz: worked night and day continuously to be able to achieve this. 177 00:10:57,150 --> 00:11:02,400 Ejaaz: I saw another example of a company which had a software engineering team of 178 00:11:02,400 --> 00:11:06,410 Ejaaz: Opus 4.8 working on projects that would typically take him two weeks. 179 00:11:06,630 --> 00:11:09,970 Ejaaz: He can now do it in less than a day as well. So the point is there's a massive 180 00:11:09,970 --> 00:11:14,300 Ejaaz: leap in intelligence for coding specifically with Fable, which is the publicly 181 00:11:14,300 --> 00:11:17,340 Ejaaz: accessible model. Now, before we continue, 182 00:11:17,990 --> 00:11:21,570 Ejaaz: We also wanted to like not just look at the demos that other people have recorded. 183 00:11:21,570 --> 00:11:24,730 Ejaaz: We want to try our own. So we have a few prepared for you today. 184 00:11:25,070 --> 00:11:27,610 Josh: Okay, so I just, I actually have no idea what you've been working on with these 185 00:11:27,610 --> 00:11:31,770 Josh: demos. I know you are building a demo. Please share with the class what the 186 00:11:31,770 --> 00:11:35,000 Josh: prompt was, what you're building and what the outputs of this thing are. 187 00:11:35,140 --> 00:11:39,050 Ejaaz: For sure. Okay, so one of my favorite breakthroughs with this model is the visual 188 00:11:39,050 --> 00:11:41,200 Ejaaz: and spatial intelligence that we referenced earlier on. 189 00:11:41,550 --> 00:11:46,720 Ejaaz: So what I did was I found this hand sketch, this hand drawn version of a floor 190 00:11:46,720 --> 00:11:49,760 Ejaaz: plan of a blueprint, which I'm showing you on the screen here. 191 00:11:49,760 --> 00:11:53,040 Ejaaz: It gives you the layout of someone's home. It has a garage. 192 00:11:53,340 --> 00:11:57,100 Ejaaz: It has bedrooms. It's kind of clunkily drawn. It's not really high fidelity. 193 00:11:57,270 --> 00:12:01,290 Ejaaz: If an architect looked at this, they'd be like, this is probably physically inaccurate. 194 00:12:01,680 --> 00:12:06,100 Ejaaz: And then I fed it to Fable and I said, listen, here's a photo of a hand-drawn floor plan. 195 00:12:06,100 --> 00:12:10,740 Ejaaz: I want you to rebuild it as a single, clean, self-contained SVG that is architecturally 196 00:12:10,990 --> 00:12:15,310 Ejaaz: accurate. I want you to improve it where you can, improve it in a way that can 197 00:12:15,310 --> 00:12:20,200 Ejaaz: like, you know, reinforce the wall structures, et cetera, like really high detail things. 198 00:12:20,480 --> 00:12:24,700 Ejaaz: And it produced this, what we're seeing on screen right here, 199 00:12:24,860 --> 00:12:27,670 Ejaaz: which is this really high fidelity floor plan. You've got the garage. 200 00:12:27,740 --> 00:12:28,690 Josh: Architectural grade. 201 00:12:28,690 --> 00:12:31,610 Ejaaz: Yeah. You've got the surface area measured in meter squared. 202 00:12:31,800 --> 00:12:34,820 Ejaaz: You've got the swing angle of each door. 203 00:12:35,030 --> 00:12:37,980 Ejaaz: You've got the entire like kind of like layout of this entire thing, 204 00:12:37,980 --> 00:12:42,860 Ejaaz: including potential mock furniture, which gave me a second idea, 205 00:12:42,860 --> 00:12:47,700 Ejaaz: which was like, okay, I asked it, I want to purchase this sofa. 206 00:12:47,850 --> 00:12:49,470 Ejaaz: These are the dimensions of this sofa. 207 00:12:50,030 --> 00:12:53,420 Ejaaz: And I want to place it in the dining room or in the lounge. Can you tell me 208 00:12:53,420 --> 00:12:56,420 Ejaaz: if I can feasibly do this in this full plan or will it get stuck? 209 00:12:56,640 --> 00:12:59,080 Ejaaz: Like, how do I, how do I do this? Do we have enough doors? Do we have enough 210 00:12:59,080 --> 00:13:02,340 Ejaaz: space to like maneuver it? Like how would this work? 211 00:13:02,710 --> 00:13:07,590 Ejaaz: And it said verdict, yes, but not flat. You're going to need to pivot the sofa 212 00:13:07,590 --> 00:13:10,140 Ejaaz: on its side and kind of like pull it in like vertically. 213 00:13:10,420 --> 00:13:14,100 Ejaaz: And it gave me this really cool mock-up of how I would do it. 214 00:13:14,100 --> 00:13:17,380 Ejaaz: It would gave me route A where I take it in through the front door. 215 00:13:17,640 --> 00:13:20,280 Ejaaz: And this is kind of like the sofa that you see here, but I need to turn it on 216 00:13:20,280 --> 00:13:23,490 Ejaaz: its side and kind of like shift it through this gap that I'm highlighting on 217 00:13:23,490 --> 00:13:24,790 Ejaaz: the screen here in green. 218 00:13:24,990 --> 00:13:28,890 Ejaaz: Or that there's route B where I can take it in from the outside and I have a 219 00:13:29,040 --> 00:13:32,810 Ejaaz: two meter, very spacious door opening, which I can bring it straight into the 220 00:13:32,810 --> 00:13:36,410 Ejaaz: lounge and place it right there. So it's really physically accurate going off 221 00:13:36,410 --> 00:13:37,660 Ejaaz: the comments that we made earlier. 222 00:13:37,860 --> 00:13:42,330 Ejaaz: It understands the kind of like reasoning behind physics really, really well. 223 00:13:42,590 --> 00:13:45,910 Josh: It's such a great companion. And this is this gets back to what we talked about 224 00:13:45,910 --> 00:13:49,370 Josh: earlier, which is like the most complicated and difficult part about this model 225 00:13:49,370 --> 00:13:54,230 Josh: is figuring out how to engage with it, what to ask it, because it's so capable of doing these things. 226 00:13:54,450 --> 00:13:58,620 Josh: We love making artifacts for the show as a way to share kind of the ideas that we're talking about. 227 00:13:58,870 --> 00:14:02,290 Josh: And this might be a good time to get into the benchmarks of how good this model 228 00:14:02,290 --> 00:14:08,000 Josh: actually is relative to other models. And I was looking at this post that I 229 00:14:08,000 --> 00:14:13,520 Josh: saw on X showcasing in particular Claude Fable 5 versus GPT 5.5. 230 00:14:13,790 --> 00:14:18,750 Josh: And my first reaction is, holy shit, that's a huge leap. 231 00:14:18,970 --> 00:14:28,280 Josh: So Fable 5 low mode is scoring over 10%, whereas GPT 5.5 extra high is getting about 5.7%. 232 00:14:28,460 --> 00:14:31,930 Josh: Now, this is on Frontier Code Benchmark. This is a specific particular coding 233 00:14:31,930 --> 00:14:36,030 Josh: benchmark. This is not across the board, but it gives you a sense of how much 234 00:14:36,030 --> 00:14:38,830 Josh: more powerful this model really is versus, 235 00:14:39,570 --> 00:14:43,420 Josh: all the others second coding benchmark that i'd say this is probably one of 236 00:14:43,420 --> 00:14:47,530 Josh: the gold standards this is what a lot of models will use to benchmark themselves against each other 237 00:14:47,940 --> 00:14:57,030 Josh: um this is the swb bench pro and fable 5 scores 22 points higher than gpt 5.5 which was already, 238 00:14:57,640 --> 00:15:03,030 Josh: what is that 11 points below opus 4.8 so it's currently looking like gemini 239 00:15:03,030 --> 00:15:09,340 Josh: 3.1 pro gpt 5.5 opus 4.8 and then fable is running away with it and this seems 240 00:15:09,340 --> 00:15:11,190 Josh: to be the case with almost all of these other. 241 00:15:11,190 --> 00:15:15,330 Ejaaz: Benchmarks i have a i have a better one for you right so a lot of people listening 242 00:15:15,330 --> 00:15:18,490 Ejaaz: to this might be thinking okay well i don't code i'm not a software engineer 243 00:15:18,490 --> 00:15:21,780 Ejaaz: why does this apply to me well there's another benchmark called gdp 244 00:15:22,180 --> 00:15:27,620 Ejaaz: vow which tests it against real world tasks that take human experts like knowledge 245 00:15:27,620 --> 00:15:31,100 Ejaaz: workers that you know do back-end admin excel sheets all that kind of stuff 246 00:15:31,100 --> 00:15:33,500 Ejaaz: hours to do and they compare it to the model 247 00:15:33,950 --> 00:15:37,790 Ejaaz: take a look at this so fable mythos 5 248 00:15:38,130 --> 00:15:42,220 Ejaaz: basically achieves the highest benchmark score it's actually almost completed 249 00:15:42,220 --> 00:15:44,830 Ejaaz: the entire benchmark so they're probably going to have to recreate an entire 250 00:15:44,830 --> 00:15:49,620 Ejaaz: new benchmark for this but basically what this means is probabilistically if you were to 251 00:15:50,180 --> 00:15:55,270 Ejaaz: blind test or blind pick the output work of a expert human that is really good 252 00:15:55,270 --> 00:16:00,100 Ejaaz: at a particular knowledge work task versus this particular model over 50% of 253 00:16:00,100 --> 00:16:04,600 Ejaaz: the time, you're going to be picking this model, which is just an insane stat to see. 254 00:16:05,070 --> 00:16:07,860 Josh: Yeah, it's pretty unbelievable. And you have to like, I'm looking at these charts, 255 00:16:07,860 --> 00:16:11,340 Josh: and you have to ask yourself the question, as Anthropica saturating benchmark, 256 00:16:11,340 --> 00:16:17,050 Josh: are they running away with it? Like, where is OpenAI in this conversation now? 257 00:16:17,050 --> 00:16:19,490 Josh: I have to imagine that GBT 5.6 is coming soon. 258 00:16:20,380 --> 00:16:24,200 Josh: Is it okay maybe it's better than opus 4.8 but can it can it eclipse fable no, 259 00:16:24,760 --> 00:16:29,380 Josh: and i mean we know that anthropic released mythos months ago so you have to 260 00:16:29,380 --> 00:16:32,650 Josh: assume that like they've been continuing progress and iterative development on 261 00:16:33,030 --> 00:16:36,860 Josh: new frontier models that are even more powerful than this and and is this is 262 00:16:36,860 --> 00:16:40,930 Josh: this beginning to become a runway or are they actually still competitive with 263 00:16:40,930 --> 00:16:43,340 Josh: each other or maybe we just don't have enough information to tell we kind of 264 00:16:43,340 --> 00:16:44,970 Josh: have to see what the response from open ai is. 265 00:16:45,080 --> 00:16:48,630 Ejaaz: Well you look at the cadence between model releases right like what was the 266 00:16:48,630 --> 00:16:51,760 Ejaaz: time since 4.8 was released. It was like less than, I think, 267 00:16:51,760 --> 00:16:53,520 Ejaaz: 30 days ago. So the cadence is getting... 268 00:16:53,520 --> 00:16:54,920 Josh: Yeah, we filmed an episode on this not too long ago. 269 00:16:54,920 --> 00:16:57,960 Ejaaz: Yeah, like I remember that episode, right? And we spoke about it and went through 270 00:16:58,310 --> 00:16:59,420 Ejaaz: its benchmarks back then. 271 00:16:59,420 --> 00:17:04,100 Ejaaz: So the point is, these model releases are happening faster, but the capability 272 00:17:04,100 --> 00:17:09,570 Ejaaz: gaps are even greater, which tells me one thing, which is we're getting closer 273 00:17:09,570 --> 00:17:12,210 Ejaaz: and closer to the AI models just building itself. 274 00:17:12,210 --> 00:17:15,470 Ejaaz: They haven't been private about this either. Anthropic has publicly claimed 275 00:17:15,470 --> 00:17:19,060 Ejaaz: that they have been using Mythos Preview to build this new version that we're 276 00:17:19,060 --> 00:17:20,660 Ejaaz: talking about today, Fable, right? 277 00:17:21,080 --> 00:17:24,980 Ejaaz: So we think we've reached a point where you could maybe call it a breakaway 278 00:17:24,980 --> 00:17:28,630 Ejaaz: from Anthropic, where they basically have recursive self-improvement almost 279 00:17:28,630 --> 00:17:32,540 Ejaaz: achieved, where the model can do all the research, figure out its own issues, 280 00:17:32,540 --> 00:17:34,180 Ejaaz: and build a better version of itself. 281 00:17:34,450 --> 00:17:39,900 Ejaaz: Now, I do want to ground us at this point in this episode, Josh, which is, 282 00:17:40,480 --> 00:17:44,510 Ejaaz: Fable 5 is an amazing model, but it's one version of the amazing model. 283 00:17:44,510 --> 00:17:49,510 Ejaaz: There is another version of this model, which is technically better than Fable 284 00:17:49,510 --> 00:17:51,930 Ejaaz: 5, but it is not publicly accessible. 285 00:17:51,930 --> 00:17:56,810 Ejaaz: It is restricted because it poses itself as a cybersecurity risk, 286 00:17:56,810 --> 00:18:00,310 Ejaaz: not just a cybersecurity risk, but also a bioweapons risk. 287 00:18:00,310 --> 00:18:06,090 Ejaaz: It is so good at biology and chemistry that it could feasibly create compounds 288 00:18:06,090 --> 00:18:11,460 Ejaaz: and a biological weapon that could pose a risk to any sort of nation state. 289 00:18:11,640 --> 00:18:17,030 Ejaaz: And so for that reason, it is under heavy restrictions and safeguards in the 290 00:18:17,030 --> 00:18:19,520 Ejaaz: version of Fable where you can't get access to any of this. 291 00:18:19,670 --> 00:18:24,390 Ejaaz: And only vetted partners and cleared government security initiatives are able 292 00:18:24,390 --> 00:18:26,570 Ejaaz: to get access to this Mythos 5 thing. 293 00:18:26,820 --> 00:18:33,190 Ejaaz: Now i put this to the test josh um and i i did a very simple example which was 294 00:18:33,190 --> 00:18:39,080 Ejaaz: um can you explain how the mitochondria works do you want to bet what its answer was 295 00:18:39,260 --> 00:18:42,610 Josh: Oh i'm gonna be best it's not touching that it's not touching biology yeah i 296 00:18:42,610 --> 00:18:46,080 Josh: want to take a second to actually explain the nuances between the models because 297 00:18:46,460 --> 00:18:49,780 Josh: when you say the word better i'm not sure it's better it's just more complete 298 00:18:49,780 --> 00:18:54,930 Josh: one model is a complete model one model is a heavily restricted model and in the case of Mythos, 299 00:18:55,350 --> 00:18:57,470 Josh: it's available for cybersecurity. 300 00:18:57,470 --> 00:19:00,080 Josh: It's available for biology. And that's what we've seen with Project Glasswing, 301 00:19:00,360 --> 00:19:05,440 Josh: where they're working privately with companies to kind of fix security vulnerabilities and figure out bio. 302 00:19:05,440 --> 00:19:09,960 Josh: And in the case of the system card, I saw that it's accelerating some bio experiments 303 00:19:09,960 --> 00:19:13,300 Josh: at a full order of magnitude, 10 times better. So it's really capable there. 304 00:19:14,110 --> 00:19:19,190 Josh: The compromise that we had to make in order to receive it was it can't touch 305 00:19:19,190 --> 00:19:20,730 Josh: bio, it can't touch cyber. 306 00:19:21,000 --> 00:19:23,560 Josh: So it's just as capable everywhere else. It will not do that. 307 00:19:23,560 --> 00:19:27,170 Josh: What happens is if you ask in the case like you did, how does mitochondria work? 308 00:19:27,530 --> 00:19:32,830 Josh: It will route through Opus 4.8 for that answer and then come back and give you a response. 309 00:19:33,010 --> 00:19:36,800 Josh: So it is as capable everywhere. It's just don't ask about bio, 310 00:19:36,800 --> 00:19:40,980 Josh: don't ask about cyber because from my experience so far trying it and EGS, it seems like yours. 311 00:19:41,300 --> 00:19:45,750 Josh: Anytime you get remotely close to those topics, it is just completely shut down, 312 00:19:45,970 --> 00:19:47,520 Josh: routed through Opus 4.8 instead. 313 00:19:47,650 --> 00:19:52,490 Ejaaz: Yeah, I think it's it's too aggressive, personally, like, as a former science 314 00:19:52,490 --> 00:19:57,100 Ejaaz: nerd, I still spend a lot of time trying to digest like some of the latest scientific advancements. 315 00:19:57,100 --> 00:20:00,460 Ejaaz: And like, listen, I'm not reading research papers. So I work with my best pal 316 00:20:00,460 --> 00:20:03,430 Ejaaz: Claude to try and figure out, you know, what the latest takeaways are. 317 00:20:03,700 --> 00:20:07,290 Ejaaz: Now, typically, I could slam that into Opus 4.8. And it would give me an amazing 318 00:20:07,290 --> 00:20:09,020 Ejaaz: summary. And I could like ask it questions. 319 00:20:09,260 --> 00:20:12,800 Ejaaz: Now, if I want to use Fable 5, it just simply won't read the paper. 320 00:20:12,800 --> 00:20:15,790 Ejaaz: As soon as it sees anything related to chemistry or biology, 321 00:20:15,790 --> 00:20:17,690 Ejaaz: it switches off and reroutes to 4.8. 322 00:20:17,690 --> 00:20:25,070 Ejaaz: So I am not able to get access to the frontier LLM intelligence or brain that 323 00:20:25,070 --> 00:20:28,960 Ejaaz: Fable 5 has for me, mythos, even though, you know, my intention isn't to build 324 00:20:28,960 --> 00:20:32,290 Ejaaz: a bio-weapon by any means, I can't get that analysis. 325 00:20:32,290 --> 00:20:36,310 Ejaaz: And so that's one version of it, right? Where like it is too heavily restricted. 326 00:20:36,670 --> 00:20:40,290 Ejaaz: The other version of this is with the more intelligent models get, 327 00:20:40,550 --> 00:20:43,480 Ejaaz: it's not just going to be super intelligent in one particular vertical. 328 00:20:43,480 --> 00:20:47,490 Ejaaz: Like for us, it's like, you know, research and creating artifacts and the best content. 329 00:20:47,870 --> 00:20:51,890 Ejaaz: It should also apply to any other profession, right? Whether you're a scientist, 330 00:20:51,890 --> 00:20:55,290 Ejaaz: whether you're a mathematician, and whether you are building different kinds 331 00:20:55,290 --> 00:20:56,530 Ejaaz: of structures or whatever it might be. 332 00:20:56,860 --> 00:21:00,590 Ejaaz: The fact that it can get triggered so easily or the fact that Anthropic has 333 00:21:00,850 --> 00:21:06,140 Ejaaz: very heavily restricted that capable intelligence in a way that like even people 334 00:21:06,140 --> 00:21:08,330 Ejaaz: that have well intentions can't get access to it. 335 00:21:08,850 --> 00:21:12,140 Ejaaz: In my opinion, is a bit of an issue. And listen, it's V1. I'm sure they're going 336 00:21:12,140 --> 00:21:17,410 Ejaaz: to like release a bunch of versions of the safeguards where it like makes it a lot easier to use. 337 00:21:18,120 --> 00:21:22,370 Ejaaz: But for V1, it's kind of like, I think it's overdone. It's important for people, 338 00:21:22,370 --> 00:21:25,300 Ejaaz: I think, to understand how these safety classifiers work as well. 339 00:21:25,740 --> 00:21:31,620 Ejaaz: Think of Claude Mythos 5 having an AI model or system that is watching it. 340 00:21:32,260 --> 00:21:35,600 Ejaaz: And as soon as one of the red flags that it's been trained on is triggered, 341 00:21:35,800 --> 00:21:39,990 Ejaaz: for example, anything to do with biology or chemistry, it gets switched off 342 00:21:39,990 --> 00:21:42,390 Ejaaz: immediately and rerouted to 4.8. 343 00:21:42,650 --> 00:21:46,720 Ejaaz: Now, there's four particular categories that Fable can't get access to. 344 00:21:46,720 --> 00:21:52,090 Ejaaz: It is cybersecurity, for biology, for chemistry, and for distillation as well. 345 00:21:52,090 --> 00:21:56,350 Ejaaz: And this is a key one which caused a lot of contention in the public ecosystem 346 00:21:56,350 --> 00:22:01,030 Ejaaz: when they launched yesterday, which is if you were to ask about model training 347 00:22:01,030 --> 00:22:04,310 Ejaaz: techniques or even just simple general questions around, hey, 348 00:22:04,310 --> 00:22:06,660 Ejaaz: I have this AI agent, it's pretty clunky, 349 00:22:07,400 --> 00:22:11,250 Ejaaz: how can I improve its harness to kind of make it go quicker or use less tokens? 350 00:22:12,170 --> 00:22:17,220 Ejaaz: Automatically degrades performance. And this is the key change in this fourth 351 00:22:17,220 --> 00:22:18,440 Ejaaz: category with distillation. 352 00:22:18,760 --> 00:22:23,460 Ejaaz: Anything that Anthropic considers to be trying to derive its model to build 353 00:22:23,460 --> 00:22:29,120 Ejaaz: another model, it gives you intentional poor performance. And it doesn't even tell you. 354 00:22:29,380 --> 00:22:34,100 Ejaaz: Now, it says that this happens for 0.3% of cases, but my guess is it's probably 355 00:22:34,320 --> 00:22:37,340 Ejaaz: happening for higher reasons. And listen, it's completely within Anthropic's 356 00:22:37,340 --> 00:22:40,720 Ejaaz: right to do this. I get it. I understand it. You want to remain competitive. 357 00:22:41,060 --> 00:22:44,680 Ejaaz: But it's just interesting to see when like you have this intelligence model 358 00:22:44,680 --> 00:22:48,150 Ejaaz: where, you know, it's meant to kind of like blossom and create and help other 359 00:22:48,150 --> 00:22:49,490 Ejaaz: people build different things. 360 00:22:49,760 --> 00:22:52,960 Ejaaz: But they're being competitive when it comes to other models, I guess. 361 00:22:53,200 --> 00:22:56,730 Josh: Well, we're getting to this unique intersection where like, they have mythos, 362 00:22:56,730 --> 00:22:59,320 Josh: they've had mythos for a little while, and they decided to keep it private. 363 00:22:59,620 --> 00:23:04,940 Josh: And that was okay, and somewhat understood because it was really discovering 364 00:23:04,940 --> 00:23:06,330 Josh: a lot of zero day vulnerabilities. 365 00:23:06,700 --> 00:23:10,030 Josh: And it seems like they worked pretty hard to figure out a way to not only improve 366 00:23:10,030 --> 00:23:13,380 Josh: the quality of the model but actually make it public and i guess like the question 367 00:23:13,380 --> 00:23:18,110 Josh: we're gonna have to start asking as these ai labs continue to create these like unbelievably 368 00:23:18,520 --> 00:23:22,380 Josh: uh forward-looking frontier models is like to what capacity are we just happy 369 00:23:22,380 --> 00:23:26,480 Josh: to have them like how much should we expect out of the labs when it comes to 370 00:23:26,480 --> 00:23:30,200 Josh: delivering these models like in my case i'm pretty stoked to be able to use fable 5, 371 00:23:30,860 --> 00:23:35,110 Josh: And I'm not interested in biology. I'm not interested in distilling the model. 372 00:23:35,110 --> 00:23:37,970 Josh: I'm just like pretty stoked to do my day-to-day work with this capable model. 373 00:23:38,330 --> 00:23:40,650 Josh: And in that sense, it's really fun and exciting and interesting. 374 00:23:41,020 --> 00:23:43,650 Josh: And I think it's the start of a longer conversation. 375 00:23:43,960 --> 00:23:48,120 Josh: We saw some legislation come in a few weeks ago, last week maybe, 376 00:23:48,120 --> 00:23:53,040 Josh: about requiring AI Frontier Labs to kind of showcase the models privately with 377 00:23:53,040 --> 00:23:54,860 Josh: the government to share what's coming down the line. 378 00:23:54,860 --> 00:24:00,640 Josh: And I guess in this essence, I'm more excited to have the model versus not have 379 00:24:00,640 --> 00:24:05,050 Josh: the model and have it have these constraints in the hope that it will slowly become 380 00:24:05,430 --> 00:24:09,950 Josh: unwounded as they kind of improve and iterate on the quality and the kind of 381 00:24:09,950 --> 00:24:11,540 Josh: like security set of this model. 382 00:24:12,070 --> 00:24:16,070 Ejaaz: Yeah, listen, I think Anthropic is ultimately doing the right thing. 383 00:24:16,070 --> 00:24:19,800 Ejaaz: I think that they can't just kind of diffuse this model to anyone and everyone 384 00:24:19,800 --> 00:24:23,990 Ejaaz: because malicious actors, however few they might be, will actually end up doing 385 00:24:23,990 --> 00:24:25,160 Ejaaz: something dangerous with this. 386 00:24:25,460 --> 00:24:31,980 Ejaaz: That being said, I think the subjectivity and who gets to govern that subjectivity is important. 387 00:24:32,230 --> 00:24:37,610 Ejaaz: Like, I can imagine a future version of an Anthropic model that isn't just necessarily 388 00:24:37,610 --> 00:24:40,340 Ejaaz: really good at biology or cybersecurity. 389 00:24:40,760 --> 00:24:44,690 Ejaaz: It might be really good at something such as trading, right, 390 00:24:44,690 --> 00:24:48,950 Ejaaz: for example. And then the question becomes, who gets access to this trading 391 00:24:48,950 --> 00:24:51,960 Ejaaz: model that is so good that it could break the stock market? 392 00:24:52,220 --> 00:24:56,310 Ejaaz: And maybe if you are Citadel, who are closely aligned with Anthropica, 393 00:24:56,310 --> 00:25:00,350 Ejaaz: I'm theorizing here, then they get access to it, but Jane Street won't get access to it. 394 00:25:00,350 --> 00:25:05,750 Ejaaz: And so it becomes this heavily-based nuance that is only dictated by maybe the 395 00:25:05,750 --> 00:25:07,730 Ejaaz: government and maybe Anthropica itself. 396 00:25:07,730 --> 00:25:11,300 Ejaaz: There was talks around like Trump taking a stake in some of these AI labs to 397 00:25:11,300 --> 00:25:14,790 Ejaaz: nationalize it for this exact reason, because it could pose a threat and they 398 00:25:14,790 --> 00:25:17,610 Ejaaz: want to have governance decisions. It just gets a little murky and messy. 399 00:25:17,610 --> 00:25:19,330 Ejaaz: And I think we're at the fork in the road. 400 00:25:19,670 --> 00:25:23,290 Ejaaz: There's no going back at this point. We are now entering a phase where 401 00:25:24,100 --> 00:25:27,740 Ejaaz: The model that you have access to may not be the most intelligent model for 402 00:25:27,740 --> 00:25:31,000 Ejaaz: the specific thing. And listen, it may not be the thing that you necessarily 403 00:25:31,000 --> 00:25:34,150 Ejaaz: do on a day-to-day, but it's a lot of things that other people do day-to-day, 404 00:25:34,150 --> 00:25:37,240 Ejaaz: and they want to get access to this model. How that is governed, I don't know. 405 00:25:37,620 --> 00:25:41,480 Ejaaz: The other restriction, which I found really interesting that I noticed in the 406 00:25:41,480 --> 00:25:47,580 Ejaaz: footnotes of their system card or announcement blog post is on June 22nd, 407 00:25:48,020 --> 00:25:51,120 Ejaaz: we ceased to get access to Fable 5. 408 00:25:51,460 --> 00:25:56,320 Ejaaz: Now i think this was taken massively out of context because i think the reasoning behind this is 409 00:25:56,730 --> 00:26:01,480 Ejaaz: it's because it depends on availability of compute so if by june 22nd anthropic 410 00:26:01,480 --> 00:26:03,390 Ejaaz: has more available compute to distribute to users 411 00:26:03,690 --> 00:26:07,630 Ejaaz: then it wouldn't be the case but on the case that it is it would shift to a 412 00:26:07,630 --> 00:26:11,930 Ejaaz: pay-per-usage model which means that you buy credits and if your credits are 413 00:26:11,930 --> 00:26:15,910 Ejaaz: consumed you then need to buy more credits kind of like the api model is that right 414 00:26:16,460 --> 00:26:19,050 Josh: Yeah, according to the blog post, it says from today through June 22nd, 415 00:26:19,050 --> 00:26:23,480 Josh: Fable 5 is included on Pro, Max, Team, and Seat-based enterprise plans at no extra cost. 416 00:26:23,840 --> 00:26:28,390 Josh: On June 23rd, we'll remove Fable 5 from those plans. Using it after that will require usage credit. 417 00:26:28,820 --> 00:26:32,660 Josh: If capacity allows, we'll extend the included window. After this point, 418 00:26:32,660 --> 00:26:36,800 Josh: when sufficient capacity allows us to do so, we aim to restore Fable 5 as a 419 00:26:36,800 --> 00:26:38,480 Josh: standard part of subscription plans. 420 00:26:38,830 --> 00:26:41,160 Josh: We intend to do this as quickly as we can. 421 00:26:41,520 --> 00:26:46,000 Josh: And yeah, it sounds like Fable 5 consumes a lot of compute. When you load it up inside of the, 422 00:26:46,520 --> 00:26:50,880 Josh: app it says fable is the most capable model and draws down usage twice as fast 423 00:26:50,880 --> 00:26:52,230 Josh: as opus so you have to imagine 424 00:26:52,580 --> 00:26:56,040 Josh: that it consumes a lot of gpus they're clearly using those gpus for a lot of 425 00:26:56,040 --> 00:27:00,190 Josh: things i think the idea is to give a preview and then extend that for as long 426 00:27:00,190 --> 00:27:03,420 Josh: as possible or just continue to extend it perpetually based on compute 427 00:27:03,830 --> 00:27:07,090 Josh: i think to your earlier point we're very much at a fork in the road 428 00:27:07,460 --> 00:27:10,720 Josh: when it comes to these models being capable enough to 429 00:27:11,090 --> 00:27:14,980 Josh: really make a meaningful impact in the world and we've spoken so much about 430 00:27:14,980 --> 00:27:19,560 Josh: alignment and ai safety and it's kind of been this like open-ended fuzzy thing 431 00:27:19,560 --> 00:27:22,910 Josh: where it hasn't really practically applied to anything that's happened before 432 00:27:23,280 --> 00:27:26,230 Josh: and we're finally at a moment in time in which the models are becoming capable 433 00:27:26,230 --> 00:27:28,570 Josh: enough to have that conversation about 434 00:27:28,910 --> 00:27:32,660 Josh: ai alignment ai safety you're starting to see why a lot of the teams are taking 435 00:27:32,660 --> 00:27:36,050 Josh: it so seriously because it is the singular question is like answering what you 436 00:27:36,050 --> 00:27:39,530 Josh: just said who gets access to this model how is it going to be restricted who 437 00:27:39,530 --> 00:27:42,780 Josh: gets to decide that and that's why the alignment and safety conversation. 438 00:27:43,450 --> 00:27:47,300 Josh: Is so important. And while you start to see a lot of the company cultures within 439 00:27:47,300 --> 00:27:51,870 Josh: these companies align around these different priority sets that separate them from each other. 440 00:27:52,060 --> 00:27:55,880 Josh: So this is, it's, it's a new day. It's a new era today. 441 00:27:56,240 --> 00:28:00,370 Josh: We are moving into the, the next frontier of models. 442 00:28:00,370 --> 00:28:03,190 Josh: It was pushed forward a considerable amount and in a way that I don't think 443 00:28:03,190 --> 00:28:06,830 Josh: we've experienced in quite a long time. And it's really exciting to see. 444 00:28:06,830 --> 00:28:10,760 Josh: I'm very excited to play with Fable, spend some time kind of generating outputs, 445 00:28:10,760 --> 00:28:14,290 Josh: figuring out what it's most capable in that could help us in the day-to-day. 446 00:28:14,670 --> 00:28:18,750 Josh: Like for me, if they never told me it wasn't gonna do bio or cyber, 447 00:28:18,750 --> 00:28:23,220 Josh: I'm not sure I'd ever come across it because that's not really within the realm of uses that I have. 448 00:28:23,220 --> 00:28:26,640 Josh: So I'm excited to just kind of play with it and figure out best use cases for 449 00:28:26,640 --> 00:28:30,400 Josh: this. In terms of pricing, what I found really interesting is CloudFable 5 is 450 00:28:30,400 --> 00:28:33,130 Josh: only twice the price of GPT 5.5. 451 00:28:34,000 --> 00:28:40,200 Josh: I believe it's $10 per million input tokens, $50 per million out, and, 452 00:28:40,880 --> 00:28:46,160 Josh: gpt 5.5 is five dollars and thirty dollars out so pretty close and much more 453 00:28:46,160 --> 00:28:50,190 Josh: capable so if the case that it does get removed from subscriptions it is still 454 00:28:50,190 --> 00:28:52,130 Josh: available from api it is not 455 00:28:52,560 --> 00:28:55,730 Josh: as expensive as i think a lot of people thought this is significantly cheaper 456 00:28:55,730 --> 00:28:57,750 Josh: than what i believe mythos preview was early on. 457 00:28:58,370 --> 00:29:02,380 Ejaaz: Yeah it's my new favorite model i've been using it relentlessly for the last 458 00:29:02,380 --> 00:29:04,340 Ejaaz: 10 hours um one thing that 459 00:29:05,430 --> 00:29:09,850 Ejaaz: Uh is a really strong capability that it has is long horizon tasks like this 460 00:29:09,850 --> 00:29:15,510 Ejaaz: model is engineered from the ground up to be able to work like a dog for six 461 00:29:15,510 --> 00:29:18,670 Ejaaz: to 12 hours at a time on whatever project that you have. 462 00:29:18,670 --> 00:29:21,630 Ejaaz: And it has this loop function, which basically says, if you come across a problem, 463 00:29:21,630 --> 00:29:25,050 Ejaaz: don't ask me, try and figure out yourself and do the thing, build the thing. 464 00:29:25,050 --> 00:29:29,280 Ejaaz: That's why we had people build world engines from scratch that we demoed earlier 465 00:29:29,280 --> 00:29:33,000 Ejaaz: and these games from scratch, all from a single prompt, the library of Babel. 466 00:29:33,290 --> 00:29:37,540 Ejaaz: So if you have an idea or if you've been pondering on a project that you've 467 00:29:37,540 --> 00:29:40,420 Ejaaz: been putting on for a while, because you're like, I know I could probably vibe 468 00:29:40,420 --> 00:29:42,430 Ejaaz: code this, but I don't want to spend like an hour doing this. 469 00:29:42,710 --> 00:29:46,130 Ejaaz: Now you just need to write one detailed prompt and you should be able to do 470 00:29:46,130 --> 00:29:50,700 Ejaaz: this. So my prompt for the listeners of the show as we wrap up this episode 471 00:29:50,700 --> 00:29:53,400 Ejaaz: is get access to this model. 472 00:29:53,650 --> 00:29:57,660 Ejaaz: Try it out. And I'm curious what your thoughts are on it. Do you think the restrictions 473 00:29:57,660 --> 00:29:58,860 Ejaaz: affect you specifically? 474 00:29:59,120 --> 00:30:03,110 Ejaaz: Or do you think it's a really good general purpose model and you're happy with 475 00:30:03,110 --> 00:30:06,570 Ejaaz: how it's presented itself. You don't care about Mythos 5 in effect. 476 00:30:07,280 --> 00:30:09,990 Ejaaz: And also, what other projects are you going to be doing with this? 477 00:30:09,990 --> 00:30:13,190 Ejaaz: Are there any kind of demos or use cases that we haven't covered that might 478 00:30:13,190 --> 00:30:17,590 Ejaaz: be specific to you in your leisure or your particular work that you might want to apply this to? 479 00:30:17,820 --> 00:30:22,260 Ejaaz: Let us know the feedback in the comments to this video or DM us on X. 480 00:30:22,260 --> 00:30:23,430 Ejaaz: Our profiles are linked below. 481 00:30:23,630 --> 00:30:28,020 Ejaaz: We want to hear back from you. But I think that brings us to the end of this 482 00:30:28,020 --> 00:30:31,950 Ejaaz: episode. We have now like a new world-leading model? 483 00:30:32,200 --> 00:30:35,920 Ejaaz: OpenAI is going to have to answer to this, but it seems like Anthropic is running 484 00:30:35,920 --> 00:30:37,690 Ejaaz: away with it. Josh, any final thoughts? 485 00:30:38,590 --> 00:30:42,070 Josh: That's it. This is a new, it's a new era today. Like, I feel like we should celebrate. 486 00:30:42,070 --> 00:30:45,740 Josh: This is a new frontier that has been pushed forward very far in an industry 487 00:30:45,740 --> 00:30:48,930 Josh: we care very deeply about. So it's exciting to see. I'm stoked to use it. 488 00:30:48,930 --> 00:30:53,300 Josh: I'm curious to hear what the best types of prompts or use cases are that anyone 489 00:30:53,300 --> 00:30:54,250 Josh: who's listening has found. 490 00:30:54,580 --> 00:30:57,160 Josh: If you enjoyed this episode, don't forget to share it with a friend who might 491 00:30:57,160 --> 00:31:01,120 Josh: also want to try Cloud Fable 5 and experiment and get their feedback on how 492 00:31:01,120 --> 00:31:03,640 Josh: it's being used to improve their life, improve productivity, 493 00:31:03,910 --> 00:31:07,620 Josh: whatever use cases it may be. But as always, thank you all so much for watching 494 00:31:07,620 --> 00:31:09,190 Josh: and we will see you guys in the next one.