1 00:00:01,800 --> 00:00:05,840 Speaker 1: All Zone Media. Hi, and welcome to the very first 2 00:00:05,880 --> 00:00:08,080 Speaker 1: Better Offline Monologue. This is going to be a short 3 00:00:08,080 --> 00:00:10,559 Speaker 1: weekly episode where I take a quick look at something 4 00:00:10,560 --> 00:00:13,240 Speaker 1: going on in the tech industry doesn't quite warrant a 5 00:00:13,240 --> 00:00:16,360 Speaker 1: full episode. One might say, they're like quick bites of 6 00:00:16,400 --> 00:00:18,959 Speaker 1: content quibis if you will, and this is a business 7 00:00:18,960 --> 00:00:22,360 Speaker 1: model that's proven successful time and time again. This week, 8 00:00:22,400 --> 00:00:24,200 Speaker 1: I'm going to give you a distilled rundown of a 9 00:00:24,239 --> 00:00:27,320 Speaker 1: recent situation at Rock both the economy and the AI world. 10 00:00:27,520 --> 00:00:29,440 Speaker 1: For those of you that either need a refresh or 11 00:00:29,480 --> 00:00:39,720 Speaker 1: rejected the notion of a TUPAC podcast. At the end 12 00:00:39,720 --> 00:00:43,120 Speaker 1: of January, something happened that radically overturned not just the 13 00:00:43,159 --> 00:00:46,360 Speaker 1: AI industry status quo, but also called into question the 14 00:00:46,360 --> 00:00:50,320 Speaker 1: dominance of the American tech industry. Our story starts on 15 00:00:50,400 --> 00:00:53,240 Speaker 1: January twentieth, when a little known Chinese company called deep 16 00:00:53,280 --> 00:00:56,760 Speaker 1: Seek released It's our one AI model, terrifying the Western 17 00:00:56,800 --> 00:01:00,160 Speaker 1: tech behemoths that applowed over two hundred billion dollars combined 18 00:01:00,360 --> 00:01:03,920 Speaker 1: into data centers in industrial grade graphics processing units GPUs 19 00:01:04,040 --> 00:01:07,280 Speaker 1: for others to power generative AI models like those behind chat, 20 00:01:07,319 --> 00:01:12,000 Speaker 1: GPT and anthropics. Claud like open aizo one model, deep 21 00:01:12,000 --> 00:01:15,120 Speaker 1: seeks are one model is a reasoning model, which is 22 00:01:15,160 --> 00:01:16,959 Speaker 1: a way to say that it works through problems step 23 00:01:16,959 --> 00:01:19,360 Speaker 1: by step, showing the users the steps it took to 24 00:01:19,360 --> 00:01:22,800 Speaker 1: reach its conclusion. Generally, when you make a request of 25 00:01:22,840 --> 00:01:26,440 Speaker 1: a generative model, it generates an answer probabilistically, meaning it's 26 00:01:26,480 --> 00:01:29,480 Speaker 1: guessing at each next bit based on the request you've made. 27 00:01:29,640 --> 00:01:32,119 Speaker 1: In the case of open aizo one model, and indeed 28 00:01:32,160 --> 00:01:35,440 Speaker 1: deep seeks are one model, the model thinks. They use 29 00:01:35,480 --> 00:01:38,240 Speaker 1: that term loosely. These models do not know anything. They're 30 00:01:38,280 --> 00:01:41,399 Speaker 1: not thinking. They have no consciousness, but I think through 31 00:01:41,520 --> 00:01:44,440 Speaker 1: each step by generating it piece by piece and reviewing 32 00:01:44,440 --> 00:01:46,520 Speaker 1: it piece by piece with separate parts of the model. 33 00:01:47,360 --> 00:01:50,800 Speaker 1: In theory, this ability to reason means it's well suited 34 00:01:50,800 --> 00:01:53,440 Speaker 1: for tasks where there's a definitive right and wrong answer, 35 00:01:53,760 --> 00:01:57,120 Speaker 1: like logic and maths. It's also what it makes it 36 00:01:57,160 --> 00:01:59,960 Speaker 1: different from the standard CHAT GPT or GPT four US, 37 00:02:00,600 --> 00:02:03,680 Speaker 1: which is considerably faster, as it doesn't undertake this step 38 00:02:03,720 --> 00:02:06,680 Speaker 1: by step thinking and thus is better suited for more 39 00:02:06,760 --> 00:02:09,400 Speaker 1: open ended questions such as what would it be like 40 00:02:09,440 --> 00:02:12,720 Speaker 1: if Garfield had a gun? To be clear, this doesn't 41 00:02:12,720 --> 00:02:15,519 Speaker 1: mean the answers are any good now. Just a few 42 00:02:15,520 --> 00:02:18,640 Speaker 1: weeks earlier, Deep Sea could release another model, albeit a 43 00:02:18,680 --> 00:02:21,400 Speaker 1: far less fanfare, likely due to it being launched there 44 00:02:21,400 --> 00:02:24,360 Speaker 1: after Christmas, of course, but nevertheless, it was called V 45 00:02:24,400 --> 00:02:27,800 Speaker 1: three and it was still pretty impressive. V three competes 46 00:02:27,840 --> 00:02:30,680 Speaker 1: with the same model that powers chat GPTs I just mentioned, 47 00:02:30,720 --> 00:02:32,880 Speaker 1: which at the time of recording this is called GPT 48 00:02:33,080 --> 00:02:36,079 Speaker 1: four zero, and that's a more general purpose kind of product. 49 00:02:36,240 --> 00:02:38,560 Speaker 1: It can write code and solve maths problems, but it's 50 00:02:38,560 --> 00:02:41,400 Speaker 1: better suited for tasks that are rooted in language, writing 51 00:02:41,440 --> 00:02:44,480 Speaker 1: that term paper, summarizing a document, whatever it is you 52 00:02:44,600 --> 00:02:47,600 Speaker 1: do with this. And it's also important to know that 53 00:02:47,680 --> 00:02:50,880 Speaker 1: this is the most commonly used style of model. You're 54 00:02:50,919 --> 00:02:53,520 Speaker 1: not really getting reasoning in everything, at least not yet, 55 00:02:53,680 --> 00:02:56,960 Speaker 1: and I don't know how prevalent it'll ever be now. 56 00:02:56,960 --> 00:02:59,959 Speaker 1: Deep seeks Tech didn't just match open ai and capabilities. 57 00:03:00,080 --> 00:03:02,880 Speaker 1: It was also purportedly cheaper to train and to operate, 58 00:03:03,400 --> 00:03:06,600 Speaker 1: whereas open AI's GPT four model reportedly costs one hundred 59 00:03:06,680 --> 00:03:10,000 Speaker 1: million dollars to train. Some experts estimate the deep Seek's 60 00:03:10,000 --> 00:03:13,560 Speaker 1: reasoning model, called R one cost a lot less than that, 61 00:03:14,000 --> 00:03:16,680 Speaker 1: and their V three model actually costs less than six 62 00:03:16,800 --> 00:03:19,959 Speaker 1: million dollars to train. This figure is open to some debate, 63 00:03:20,760 --> 00:03:22,880 Speaker 1: but the big thing is about these models is they're 64 00:03:22,960 --> 00:03:26,920 Speaker 1: dramatically cheaper. They can be run on your computer, though 65 00:03:27,080 --> 00:03:29,760 Speaker 1: much slower, or they can be run another cloud infrastructure. 66 00:03:30,280 --> 00:03:32,160 Speaker 1: And in the case of the V three model, the 67 00:03:32,160 --> 00:03:35,160 Speaker 1: one that competes with chat GPT, it was actually about 68 00:03:35,200 --> 00:03:38,800 Speaker 1: fifty times cheaper, and the Reasoning model are one about 69 00:03:38,880 --> 00:03:41,480 Speaker 1: thirty which is crazy. Now, these are the prices that 70 00:03:41,520 --> 00:03:43,840 Speaker 1: are run on the servers where deep Seak runs, but 71 00:03:43,880 --> 00:03:46,080 Speaker 1: we're very quickly going to see as other people host 72 00:03:46,160 --> 00:03:48,560 Speaker 1: them exactly how much cheaper they are. And they're more 73 00:03:48,600 --> 00:03:52,360 Speaker 1: efficient too, which is crazy. They's so much more efficient. 74 00:03:53,720 --> 00:03:56,400 Speaker 1: And it's also important to note that they train these 75 00:03:56,400 --> 00:03:59,360 Speaker 1: models using older generation N video chips because they had 76 00:03:59,400 --> 00:04:01,600 Speaker 1: sanctions on them from China. They got some of the 77 00:04:01,640 --> 00:04:05,360 Speaker 1: newer ones too through weird resellers, but nevertheless this made 78 00:04:05,400 --> 00:04:08,640 Speaker 1: it much harder for them to get GPUs in general, 79 00:04:09,120 --> 00:04:11,480 Speaker 1: and thus they were able to kind of squeeze more 80 00:04:11,520 --> 00:04:13,200 Speaker 1: power out than they had to come up with really 81 00:04:13,280 --> 00:04:16,479 Speaker 1: interesting kind of assembly language level stuff where they did 82 00:04:16,520 --> 00:04:19,279 Speaker 1: extra things with the GPUs, the well, the fat and 83 00:04:19,360 --> 00:04:22,520 Speaker 1: happy tech executives never thought of, and Sam Altman and 84 00:04:22,560 --> 00:04:25,160 Speaker 1: his ILK from open ai never really thought of, because well, 85 00:04:25,320 --> 00:04:27,200 Speaker 1: why would they have to be why would they have 86 00:04:27,240 --> 00:04:29,680 Speaker 1: to think of that they had the unlimited money cheap 87 00:04:29,720 --> 00:04:32,080 Speaker 1: from the hyperscalers, like in the case of open Ai 88 00:04:32,320 --> 00:04:35,120 Speaker 1: funded by Microsoft, in the case of Anthropic funded by 89 00:04:35,240 --> 00:04:38,720 Speaker 1: Amazon and Google. And this is where the narrative has 90 00:04:38,760 --> 00:04:41,000 Speaker 1: begun to kind of fall apart, because all of this 91 00:04:41,040 --> 00:04:43,839 Speaker 1: has made it much harder to justify these companies building 92 00:04:43,839 --> 00:04:47,279 Speaker 1: new data centers and buying new in video GPUs. This 93 00:04:47,640 --> 00:04:50,440 Speaker 1: entire AI boom has been based off of the assumption 94 00:04:50,480 --> 00:04:52,880 Speaker 1: that the only way to build powerful models was to 95 00:04:52,920 --> 00:04:55,560 Speaker 1: get the biggest, most hugest chips from in video each year, 96 00:04:55,960 --> 00:04:57,560 Speaker 1: and that there was just no way to make these 97 00:04:57,640 --> 00:05:01,640 Speaker 1: models cheaper. Now as an aside, lost five billion dollars 98 00:05:01,680 --> 00:05:04,400 Speaker 1: in twenty twenty four and all of their products are unprofitable, 99 00:05:04,520 --> 00:05:07,520 Speaker 1: even their two hundred dollars a month open ai Chat 100 00:05:07,560 --> 00:05:11,240 Speaker 1: GPT pro subscription. I hate these terms, by the way, 101 00:05:11,400 --> 00:05:15,640 Speaker 1: They're all different. Nevertheless, everyone assumed that there was never 102 00:05:15,680 --> 00:05:18,360 Speaker 1: going to be a more efficient model and I personally 103 00:05:18,440 --> 00:05:20,600 Speaker 1: made the mistake of saying, well, if it was going 104 00:05:20,680 --> 00:05:22,599 Speaker 1: to be more efficient, surely they would want it to 105 00:05:22,640 --> 00:05:25,760 Speaker 1: be or they could do that, right, right, Maybe they 106 00:05:25,839 --> 00:05:27,839 Speaker 1: just have to do this stuff even though it's stupid. 107 00:05:28,680 --> 00:05:31,760 Speaker 1: That was never the case, and deep Seek proved in crucially, 108 00:05:31,800 --> 00:05:34,560 Speaker 1: deep Seak released its models under an open source license, 109 00:05:34,640 --> 00:05:37,520 Speaker 1: meaning any company can reuse and repurpose its tech without 110 00:05:37,560 --> 00:05:40,480 Speaker 1: having to pay anyone anything, any license fees or anything, 111 00:05:40,640 --> 00:05:43,960 Speaker 1: or ask anyone for permission. Open Ai, by contrast, keeps 112 00:05:43,960 --> 00:05:46,840 Speaker 1: its technology under lock and key. Despite their name, open 113 00:05:46,880 --> 00:05:50,080 Speaker 1: ai is a deeply secretive organization open in name only. 114 00:05:50,839 --> 00:05:53,800 Speaker 1: In summary, deep Seek has created a viable alternative to 115 00:05:53,839 --> 00:05:58,240 Speaker 1: open AI's tech and indeed anthropics that's equally capable, vastly cheaper, 116 00:05:58,360 --> 00:06:00,680 Speaker 1: an open source and proven that you don't need the 117 00:06:00,680 --> 00:06:03,640 Speaker 1: most expensive and powerful chips to do so. And they 118 00:06:03,720 --> 00:06:06,520 Speaker 1: kind of came out of nowhere. Well, deep Seek isn't 119 00:06:06,560 --> 00:06:10,280 Speaker 1: exactly a tiny little startup. They're also not a Silicon 120 00:06:10,360 --> 00:06:13,880 Speaker 1: Valley giant with billions of dollars of venture capital, or 121 00:06:14,120 --> 00:06:16,880 Speaker 1: someone who's backed by one of the many different companies 122 00:06:16,880 --> 00:06:19,680 Speaker 1: with a three trillion dollar market cap. They started off 123 00:06:19,680 --> 00:06:21,880 Speaker 1: as a side project from a Chinese hedge fund. No, 124 00:06:22,000 --> 00:06:25,480 Speaker 1: I'm not kidding now, still an eight billion dollars under 125 00:06:25,480 --> 00:06:29,520 Speaker 1: management hedge fund. They're not small at all. It's so strange. 126 00:06:29,920 --> 00:06:32,880 Speaker 1: It's a kind of cynical version of David versus Goliath, 127 00:06:32,960 --> 00:06:37,040 Speaker 1: where David is a hedge fund baby and Goliath is 128 00:06:37,600 --> 00:06:42,640 Speaker 1: several different hyperscalers taped together with a bad idea. But anyway, 129 00:06:42,680 --> 00:06:45,039 Speaker 1: put yourself in the shoes of open Ai CEO and 130 00:06:45,080 --> 00:06:48,160 Speaker 1: co founder Sam Mortmon. You've crafted this public perception of 131 00:06:48,200 --> 00:06:51,080 Speaker 1: yourself as a visionary that isn't just bringing generative AI 132 00:06:51,120 --> 00:06:53,360 Speaker 1: to the massives, but you're on the path that will 133 00:06:53,360 --> 00:06:56,359 Speaker 1: bring about artificial general intelligence, which is to say, an 134 00:06:56,400 --> 00:06:59,400 Speaker 1: AI that's as capable as a human being. You've crafted 135 00:06:59,400 --> 00:07:01,679 Speaker 1: this myth not just about yourself, but about your company 136 00:07:01,680 --> 00:07:03,520 Speaker 1: and what you'll do, and this has allowed you to, 137 00:07:03,680 --> 00:07:05,760 Speaker 1: in essence, to fire the laws of physics when it 138 00:07:05,760 --> 00:07:08,080 Speaker 1: comes to business. You can burn money at a rate 139 00:07:08,160 --> 00:07:11,440 Speaker 1: unlike any tech company in history, with no hope of 140 00:07:11,480 --> 00:07:13,160 Speaker 1: making a profit, or at least not in the short 141 00:07:13,200 --> 00:07:16,400 Speaker 1: to medium term, and no real expectation that you'll do so, 142 00:07:16,720 --> 00:07:19,400 Speaker 1: as investors will still line up to give you more money. 143 00:07:19,400 --> 00:07:22,560 Speaker 1: With your company valued and even more ludicrous numbers seemingly 144 00:07:22,600 --> 00:07:25,760 Speaker 1: every other month, you can say these outlandish things like 145 00:07:25,800 --> 00:07:28,680 Speaker 1: you need seven trillion dollars to build the infrastructure and 146 00:07:28,760 --> 00:07:31,400 Speaker 1: chip manufacturing capacity to bring your plans to life, and 147 00:07:31,440 --> 00:07:33,280 Speaker 1: you don't get laughed out of the room if I 148 00:07:33,360 --> 00:07:35,880 Speaker 1: said this shit, they'd asked me if I had a concussion. 149 00:07:36,640 --> 00:07:38,880 Speaker 1: You can say stuff like I want to build five 150 00:07:38,920 --> 00:07:41,520 Speaker 1: hundred billion dollars worth of data centers, and instead of 151 00:07:41,520 --> 00:07:44,240 Speaker 1: people rolling their eyes, the world's largest tech companies and 152 00:07:44,400 --> 00:07:47,680 Speaker 1: investors will say, damn man, that's sick, and then it 153 00:07:47,720 --> 00:07:51,200 Speaker 1: turns out that you were wrong. You'd always assume that 154 00:07:51,240 --> 00:07:54,320 Speaker 1: AI must be expensive, that the models used to power 155 00:07:54,440 --> 00:07:58,480 Speaker 1: your apps like chat, GPT and Dally their image generator, 156 00:07:59,720 --> 00:08:02,000 Speaker 1: they always cost more to build, they'd always cost more 157 00:08:02,040 --> 00:08:05,520 Speaker 1: to run, they'd always require more powerful hardware, or maybe 158 00:08:05,520 --> 00:08:07,600 Speaker 1: you just never thought about it too hard because you 159 00:08:07,680 --> 00:08:10,240 Speaker 1: never have to worry about money and to grow to 160 00:08:10,240 --> 00:08:12,920 Speaker 1: build more capable aiye moodels, you assume that you would 161 00:08:12,920 --> 00:08:15,640 Speaker 1: always need more money, and so much more money than 162 00:08:15,680 --> 00:08:19,000 Speaker 1: anyone's ever had, And then here comes this Chinese company 163 00:08:19,040 --> 00:08:23,040 Speaker 1: didn't just replicate the functionality of your model. And on 164 00:08:23,080 --> 00:08:25,320 Speaker 1: top of that, by the way, one is open ayes 165 00:08:25,400 --> 00:08:27,640 Speaker 1: one moat. It was the one thing that people liked. 166 00:08:27,760 --> 00:08:31,760 Speaker 1: It was their most sophisticated AI model. But this company 167 00:08:31,800 --> 00:08:34,440 Speaker 1: came along and did it on a shoestring budget, both 168 00:08:34,520 --> 00:08:37,240 Speaker 1: for actually training it even if the estimates are off 169 00:08:37,280 --> 00:08:39,719 Speaker 1: by like factors of ten. But these things are more 170 00:08:39,720 --> 00:08:42,920 Speaker 1: efficient too. And this company didn't even have access to 171 00:08:42,960 --> 00:08:46,000 Speaker 1: the most capable GPUs. They didn't have the server architecture 172 00:08:46,120 --> 00:08:50,560 Speaker 1: provided by Microsoft or Amazon or Google. And wow, and 173 00:08:50,600 --> 00:08:52,360 Speaker 1: what did they do next with this thing they built 174 00:08:52,360 --> 00:08:55,200 Speaker 1: that's competitive with you only real moat? They gave it away. 175 00:08:56,080 --> 00:08:59,080 Speaker 1: Oh goodness me, Sammy, things aren't looking good at all. 176 00:08:59,679 --> 00:09:02,079 Speaker 1: And this is where Sam Moultman's at. This is where 177 00:09:02,080 --> 00:09:03,920 Speaker 1: open ai and the companies that are backed to it, 178 00:09:03,960 --> 00:09:06,800 Speaker 1: and their competitors, this is where they're all at. The 179 00:09:06,880 --> 00:09:10,200 Speaker 1: decisive lead they once enjoyed has like a puddle on 180 00:09:10,240 --> 00:09:13,360 Speaker 1: a hot day, evaporated. And you'd see that happen a 181 00:09:13,400 --> 00:09:16,400 Speaker 1: lot here in beautiful Las Vegas, Nevada. Now, don't get 182 00:09:16,400 --> 00:09:19,120 Speaker 1: me wrong, open ai still burns money. But now when 183 00:09:19,120 --> 00:09:21,920 Speaker 1: Sam Moretman dusts off his begging bowl. Investors will ask, 184 00:09:22,000 --> 00:09:31,560 Speaker 1: perhaps for the first time, one very simple question, why