1 00:00:02,480 --> 00:00:06,800 Speaker 1: Alzo Media. Hello and welcome to your first Better Offline 2 00:00:06,840 --> 00:00:09,680 Speaker 1: monologue of twenty twenty six. I'm your host ed Zitron. 3 00:00:17,760 --> 00:00:19,800 Speaker 1: We're back from CES and we've already had a great 4 00:00:19,840 --> 00:00:23,000 Speaker 1: conversation with the wonderful Steve Burke of Games Nexus, and 5 00:00:23,040 --> 00:00:25,119 Speaker 1: this felt like a good time to take stark and 6 00:00:25,160 --> 00:00:29,080 Speaker 1: talk about what I see happening in the year ahead. Also, legitimately, 7 00:00:29,120 --> 00:00:31,200 Speaker 1: thank you for all for listening at CES. It was 8 00:00:31,600 --> 00:00:33,320 Speaker 1: one of the best weeks of my life, and we 9 00:00:33,360 --> 00:00:36,040 Speaker 1: made some amazing stuff and you've all sent wonderful emails 10 00:00:36,040 --> 00:00:38,760 Speaker 1: and such love hearing from you. But let's talk about 11 00:00:38,760 --> 00:00:41,559 Speaker 1: this year. I have a strong gut instinct that this 12 00:00:41,720 --> 00:00:44,240 Speaker 1: is the year of reckonings for the AI bubble. Nothing 13 00:00:44,280 --> 00:00:46,320 Speaker 1: I'm seeing in the news, the markets, or hearing from 14 00:00:46,320 --> 00:00:48,480 Speaker 1: people I talked to behind the scenes suggests that things 15 00:00:48,479 --> 00:00:50,920 Speaker 1: are improving at all. In fact, it kind of feels 16 00:00:50,920 --> 00:00:52,680 Speaker 1: like when your dog takes a shit somewhere in your 17 00:00:52,720 --> 00:00:55,480 Speaker 1: apartment but you can't quite find it. There's that little stink, 18 00:00:55,520 --> 00:00:57,280 Speaker 1: but it's not strong enough to give you an idea 19 00:00:57,320 --> 00:00:59,560 Speaker 1: of where it is. But you know, one day, without fail, 20 00:00:59,600 --> 00:01:02,080 Speaker 1: you're going to a nasty brown delight under your foot 21 00:01:02,560 --> 00:01:06,319 Speaker 1: colorful Simile's aside, though Generative AI is not becoming cheaper, 22 00:01:06,400 --> 00:01:09,360 Speaker 1: more effective, or more efficient in Vidio's new Vira Ruben, 23 00:01:09,400 --> 00:01:12,000 Speaker 1: GPUs are likely to be more expensive than ever, in 24 00:01:12,080 --> 00:01:14,560 Speaker 1: part thanks to the ever increasing price of RAM and 25 00:01:14,600 --> 00:01:16,880 Speaker 1: in part because of its single vendor mod monopoly, which 26 00:01:16,920 --> 00:01:19,800 Speaker 1: allows them to set prices well to stun. And the 27 00:01:19,840 --> 00:01:21,880 Speaker 1: other thing with them as well is they're going to 28 00:01:21,959 --> 00:01:24,640 Speaker 1: allow people to use the same racks from Blackwell, so 29 00:01:24,760 --> 00:01:27,640 Speaker 1: the things you put the GPUs into, but only for 30 00:01:27,680 --> 00:01:30,680 Speaker 1: the beginning of Verra Ruben. Then they're going to move 31 00:01:30,680 --> 00:01:33,080 Speaker 1: to these giant I think they're called Kaiber racks. The 32 00:01:33,120 --> 00:01:35,960 Speaker 1: current ones are called Oberon. Someone will email and correct me, 33 00:01:36,040 --> 00:01:38,200 Speaker 1: but it's one of those two. The new one is 34 00:01:38,240 --> 00:01:42,119 Speaker 1: going to require an entirely new cooling system architecture. It's 35 00:01:42,120 --> 00:01:44,959 Speaker 1: all very good, but on the subject of things that 36 00:01:45,000 --> 00:01:47,760 Speaker 1: are also very good, I believe we're also on the 37 00:01:47,840 --> 00:01:49,960 Speaker 1: verge of the multiple con jobs of the AI bubble 38 00:01:49,960 --> 00:01:53,120 Speaker 1: collapsing too. In July of last year, former open AI 39 00:01:53,200 --> 00:01:55,800 Speaker 1: CTO Mira Marati, who barely appear to be able to 40 00:01:55,840 --> 00:01:58,120 Speaker 1: get out of fucking sentence in every interview. I've seen 41 00:01:58,360 --> 00:02:00,920 Speaker 1: founded Thinking Machines raising two two billion dollars at a 42 00:02:00,960 --> 00:02:04,600 Speaker 1: twelve billion dollar valuation, all while refusing to tell investors 43 00:02:04,600 --> 00:02:07,040 Speaker 1: what product they'd make, or what the business plan was, 44 00:02:07,160 --> 00:02:09,640 Speaker 1: or how they'd make money or anything like that. Naturally, 45 00:02:09,680 --> 00:02:12,120 Speaker 1: investors were like, take my money, please, I need to 46 00:02:12,120 --> 00:02:14,560 Speaker 1: give this to your mirror. I am an idiot, I'm 47 00:02:14,560 --> 00:02:17,959 Speaker 1: a moron, I'm a goofboar. I just love wasting money. Now, 48 00:02:17,960 --> 00:02:20,559 Speaker 1: when this happened, they were able to poach open AISVP 49 00:02:20,639 --> 00:02:23,679 Speaker 1: of Research for Post Training, Barrett Zov, among two other 50 00:02:23,720 --> 00:02:28,200 Speaker 1: staff as I just really can't engineer the excitement to name. 51 00:02:28,720 --> 00:02:30,560 Speaker 1: And then, a few days ago, as reported by a 52 00:02:30,560 --> 00:02:33,560 Speaker 1: friend of the show, Kylie Robinson of Core Memory, Zof 53 00:02:33,680 --> 00:02:36,200 Speaker 1: was fired from Thinking Machines after telling Marati he was 54 00:02:36,240 --> 00:02:39,720 Speaker 1: considering leaving with wiredsmac Zef reporting in a follow up 55 00:02:40,080 --> 00:02:45,359 Speaker 1: that the firing involved Zov sharing confidential information with competitors. 56 00:02:45,960 --> 00:02:49,320 Speaker 1: What confidential information? What possible information could he have had 57 00:02:49,680 --> 00:02:53,400 Speaker 1: about the things they're not building? About the bullshit? Who knows? 58 00:02:53,720 --> 00:02:55,880 Speaker 1: I think she just fired him because she was mad 59 00:02:55,919 --> 00:02:58,680 Speaker 1: that he was leaving. What the fuck is Thinking Machines 60 00:02:58,680 --> 00:03:01,800 Speaker 1: building exactly would have had a single AI thing last 61 00:03:01,840 --> 00:03:04,680 Speaker 1: year for and I quote, controlling every aspect of model 62 00:03:04,720 --> 00:03:07,680 Speaker 1: training and fine tuning or some such bollocks. I realize 63 00:03:07,680 --> 00:03:09,720 Speaker 1: I'm being a little dismissive, but I just do not 64 00:03:09,880 --> 00:03:12,440 Speaker 1: have it in me anymore to feign excitement over every 65 00:03:12,520 --> 00:03:16,720 Speaker 1: single one of these two or three or whatever fraudulent Okay, 66 00:03:16,960 --> 00:03:19,320 Speaker 1: not literally fraudulent. I have no evidence of that. This 67 00:03:19,400 --> 00:03:21,800 Speaker 1: is a libel blah blah blah. It's not fraudulent. It 68 00:03:21,840 --> 00:03:25,720 Speaker 1: just looks like bollocks. I can't be excited about the 69 00:03:25,840 --> 00:03:28,840 Speaker 1: new asshole who raised a billion or two billion dollars 70 00:03:28,880 --> 00:03:30,800 Speaker 1: to do nothing. I don't have it in me anymore. 71 00:03:31,120 --> 00:03:33,880 Speaker 1: I'm not going to read a sentence about post training 72 00:03:33,960 --> 00:03:37,800 Speaker 1: or fine tuning and pretend that means anything. It doesn't 73 00:03:37,800 --> 00:03:40,960 Speaker 1: mean anything anymore. Nothing has changed, nothing is going on. 74 00:03:41,080 --> 00:03:45,160 Speaker 1: Everyone is wasting money, and I can't care about who 75 00:03:45,200 --> 00:03:49,120 Speaker 1: looted billions of dollars from venture capitalists last nor do 76 00:03:49,200 --> 00:03:51,400 Speaker 1: I believe that anything should happen to this venture capitalists 77 00:03:51,400 --> 00:03:54,080 Speaker 1: other than being tard and feathered by the limited partners. 78 00:03:54,480 --> 00:03:57,040 Speaker 1: These people aren't building anything. They're huffing their own farts 79 00:03:57,040 --> 00:03:58,960 Speaker 1: and getting paid tens of billions of dollars to do it. 80 00:03:59,120 --> 00:04:02,080 Speaker 1: I'm sick of here wrying about them, and I think 81 00:04:02,160 --> 00:04:04,840 Speaker 1: this is the year we see this kind of con diet. 82 00:04:04,960 --> 00:04:07,320 Speaker 1: We're long past the point at which generative AI has 83 00:04:07,360 --> 00:04:10,400 Speaker 1: made sense. Bloomberg reported last year that there were one 84 00:04:10,520 --> 00:04:14,080 Speaker 1: hundred and seventy eight point five billion dollars of data 85 00:04:14,120 --> 00:04:17,000 Speaker 1: center deals in the US alone in twenty twenty five, 86 00:04:17,120 --> 00:04:19,240 Speaker 1: and based on my rough analysis, there's less than a 87 00:04:19,240 --> 00:04:22,560 Speaker 1: billion dollars of actual compute revenue outside of hyperscalers who 88 00:04:22,560 --> 00:04:24,200 Speaker 1: are just trying to move it off of their balance 89 00:04:24,200 --> 00:04:28,200 Speaker 1: sheet so their earnings look better. The demand isn't arriving, 90 00:04:28,360 --> 00:04:30,600 Speaker 1: and neither of the products and all of this speculative 91 00:04:30,600 --> 00:04:33,440 Speaker 1: money sloshing around is going to zero. I also want 92 00:04:33,440 --> 00:04:36,039 Speaker 1: to be clearer about something else I keep hearing about. 93 00:04:36,640 --> 00:04:39,640 Speaker 1: If open ai dies, the AI bubble bursts with it. 94 00:04:40,240 --> 00:04:43,440 Speaker 1: Mister Sebastian Malabi of the Council on Foreign Relations had 95 00:04:43,440 --> 00:04:44,960 Speaker 1: a piece in The New York Times a few days 96 00:04:45,000 --> 00:04:47,440 Speaker 1: ago saying that he believed open ai would run out 97 00:04:47,480 --> 00:04:50,120 Speaker 1: of money by July twenty twenty seven, adding that and 98 00:04:50,160 --> 00:04:52,800 Speaker 1: I quote an open ai failure wouldn't be an indictment 99 00:04:52,839 --> 00:04:54,839 Speaker 1: of AI, but merely the end of the most hype 100 00:04:54,920 --> 00:04:57,400 Speaker 1: driven builder of it. He also said that the probable 101 00:04:57,440 --> 00:05:00,760 Speaker 1: result is that open ai would be absorbed by Mike, Microsoft, Amazon, 102 00:05:00,800 --> 00:05:03,560 Speaker 1: or another cash rich behemoth, which I am going to 103 00:05:03,640 --> 00:05:06,080 Speaker 1: sit down and reread a lot of my newsletters because 104 00:05:06,080 --> 00:05:08,800 Speaker 1: I'm pretty sure that's fucking lifted from my work, Like 105 00:05:08,920 --> 00:05:11,559 Speaker 1: open ai being absorbed by Microsoft. I know I fucking 106 00:05:11,600 --> 00:05:13,960 Speaker 1: said it. If you can find it, email me easy 107 00:05:14,000 --> 00:05:16,000 Speaker 1: at better Offline dot com and I will give you 108 00:05:16,040 --> 00:05:18,440 Speaker 1: a big thumbs up in a photo. But I want 109 00:05:18,440 --> 00:05:20,720 Speaker 1: to be clear. In any case, if open ai dies, 110 00:05:20,760 --> 00:05:23,560 Speaker 1: the bubble burst aggressively and brutally. Open ai is the 111 00:05:23,560 --> 00:05:26,400 Speaker 1: biggest brand name in AI. Samultman is the only well 112 00:05:26,400 --> 00:05:29,240 Speaker 1: known founder, and chat gpt is the only AI product 113 00:05:29,240 --> 00:05:33,560 Speaker 1: with any consumer penetration, and no Microsoft and Google renaming 114 00:05:33,600 --> 00:05:37,839 Speaker 1: products doesn't count, nor does Nano Banana. I can't believe 115 00:05:37,880 --> 00:05:41,480 Speaker 1: I nearly got away with saying or thinking that Nano 116 00:05:41,520 --> 00:05:45,839 Speaker 1: Banana for a week. More important, the open ai is 117 00:05:45,839 --> 00:05:48,080 Speaker 1: the only company with any real demand for a compute 118 00:05:48,080 --> 00:05:50,840 Speaker 1: outside of Anthropic, another company that relies on billions of 119 00:05:50,880 --> 00:05:53,640 Speaker 1: dollars of venture capital and subsidies to stop itself from 120 00:05:53,680 --> 00:05:58,120 Speaker 1: falling over and dying. As you can hear them a 121 00:05:58,120 --> 00:06:01,680 Speaker 1: little bit heated on this subject because I'm really really 122 00:06:01,760 --> 00:06:05,479 Speaker 1: sick of hearing this. I'm really sick of hearing things 123 00:06:05,640 --> 00:06:08,920 Speaker 1: from these AI boosters are getting more crazy and more defensive, 124 00:06:08,960 --> 00:06:11,560 Speaker 1: claiming that this is an open AI bubble, not an 125 00:06:11,560 --> 00:06:15,880 Speaker 1: AI bubble. What do you think happens when AI companies 126 00:06:15,920 --> 00:06:19,159 Speaker 1: try and raise money after open ai dies? Do you 127 00:06:19,160 --> 00:06:21,840 Speaker 1: think investors are going to think? You think they're going 128 00:06:21,839 --> 00:06:25,120 Speaker 1: to do the Tobias fun K poly thing from arrested Development, 129 00:06:25,120 --> 00:06:26,840 Speaker 1: where it didn't work for name, it will work for us. 130 00:06:26,960 --> 00:06:29,919 Speaker 1: Now they're going to say, oh shit, oh shit. The 131 00:06:30,000 --> 00:06:32,920 Speaker 1: one company with all the press, all the money, all 132 00:06:32,960 --> 00:06:36,360 Speaker 1: the infrastructure, all the attention, all of the governmental support 133 00:06:36,600 --> 00:06:39,640 Speaker 1: died on its ass. Why would I invest in more 134 00:06:39,680 --> 00:06:42,560 Speaker 1: in this? And I realize venture capitalists are fucking stupid 135 00:06:42,640 --> 00:06:45,960 Speaker 1: at the best of times, but even by this standard, 136 00:06:46,200 --> 00:06:48,120 Speaker 1: I think that they're going to be too scared to 137 00:06:48,160 --> 00:06:50,719 Speaker 1: put the money in. And I don't think anyone else 138 00:06:50,800 --> 00:06:53,000 Speaker 1: is going to be able to justify it either. But 139 00:06:53,080 --> 00:06:54,960 Speaker 1: I do think by the way AI boosters are about 140 00:06:55,000 --> 00:06:57,000 Speaker 1: to get real crazy out there. I think you're going 141 00:06:57,040 --> 00:06:58,960 Speaker 1: to see them say it's an open AI bubble, not 142 00:06:59,000 --> 00:07:01,599 Speaker 1: and a bubble them. We need these investments to beat China. 143 00:07:01,839 --> 00:07:04,920 Speaker 1: Then they're gonna blame people like me AI skeptics. They're 144 00:07:04,920 --> 00:07:08,839 Speaker 1: gonna say luddites destroyed innovation. And if you hear that, 145 00:07:10,280 --> 00:07:13,080 Speaker 1: know that in reality, the people destroying innovation are the 146 00:07:13,080 --> 00:07:15,520 Speaker 1: ones diverting more than half of all venture capital to 147 00:07:15,560 --> 00:07:20,960 Speaker 1: invest in unreliable, unproductive, unsustainable, and unremarkable software in pursuit 148 00:07:21,000 --> 00:07:24,680 Speaker 1: of creating the future of digital slavery. I'm sick of 149 00:07:24,720 --> 00:07:26,760 Speaker 1: these fucking people, but that doesn't mean I won't be 150 00:07:26,800 --> 00:07:29,040 Speaker 1: here to tell you all about their collapse. I'll be 151 00:07:29,080 --> 00:07:31,200 Speaker 1: back next week with a three or four parter. This 152 00:07:31,320 --> 00:07:33,960 Speaker 1: is gonna be a bomb burner. You're gonna love it. 153 00:07:34,120 --> 00:07:34,800 Speaker 1: Zichron out