1 00:00:01,040 --> 00:00:04,080 [Speaker 0] Hello, and welcome to the Geek News Central 2 00:00:04,080 --> 00:00:07,120 [Speaker 0] podcast. I'm your host, Ray Cochran. I have 3 00:00:07,120 --> 00:00:08,880 [Speaker 0] a pretty great episode lined up for you 4 00:00:08,880 --> 00:00:12,500 [Speaker 0] guys tonight. Leading off, a new study says 5 00:00:12,559 --> 00:00:15,945 [Speaker 0] AI is literally frying workers' brains. And as 6 00:00:15,945 --> 00:00:18,925 [Speaker 0] someone running multiple AI agents in a Tmux 7 00:00:19,065 --> 00:00:23,145 [Speaker 0] window, I've got some thoughts on that. Also 8 00:00:23,145 --> 00:00:29,520 [Speaker 0] on the show, Anthropic's insane month. Users flooding 9 00:00:29,520 --> 00:00:32,640 [Speaker 0] in, the Pentagon retaliating, and a secret model 10 00:00:32,640 --> 00:00:35,760 [Speaker 0] leaking out. Today, I actually just caught wind 11 00:00:35,760 --> 00:00:37,920 [Speaker 0] of their source code getting out, so I'm 12 00:00:37,920 --> 00:00:39,540 [Speaker 0] gonna have to take a look at that. 13 00:00:39,600 --> 00:00:42,915 [Speaker 0] But also on the show, OpenAI killed Sora 14 00:00:43,055 --> 00:00:46,094 [Speaker 0] and lost Disney in the process. An AI 15 00:00:46,094 --> 00:00:48,915 [Speaker 0] agent called Open Claw had my own machine 16 00:00:48,975 --> 00:00:53,215 [Speaker 0] making suspicious outbound requests. Apple killed the Mac 17 00:00:53,215 --> 00:00:57,010 [Speaker 0] Pro and is building tiny AI spy pins 18 00:00:57,010 --> 00:01:00,129 [Speaker 0] instead. And to close it out, King Tut's 19 00:01:00,129 --> 00:01:03,750 [Speaker 0] dagger was apparently forged from a meteorite. Yeah. 20 00:01:03,809 --> 00:01:07,924 [Speaker 0] Really. Let's get into it. Tonight's show is 21 00:01:07,924 --> 00:01:10,424 [Speaker 0] brought to you by our twenty year partner, 22 00:01:10,645 --> 00:01:14,565 [Speaker 0] GoDaddy. Exclusive deals and pricing at geek news 23 00:01:14,565 --> 00:01:20,580 [Speaker 0] central dot com forward slash GoDaddy. To start 24 00:01:20,580 --> 00:01:23,220 [Speaker 0] the articles out, AI use at work is 25 00:01:23,220 --> 00:01:26,660 [Speaker 0] causing brain fry. I wanted to start with 26 00:01:26,660 --> 00:01:30,120 [Speaker 0] this one tonight because I think most people 27 00:01:30,260 --> 00:01:34,775 [Speaker 0] that have started to bathe themselves in this 28 00:01:34,775 --> 00:01:38,775 [Speaker 0] ecosystem have definitely felt something similar to this, 29 00:01:38,775 --> 00:01:40,534 [Speaker 0] whether they've had a name for it or 30 00:01:40,534 --> 00:01:44,134 [Speaker 0] not. A study from Boston Consulting Group and 31 00:01:44,134 --> 00:01:47,990 [Speaker 0] UC Riverside surveyed about fifteen hundred full time 32 00:01:47,990 --> 00:01:50,789 [Speaker 0] US workers and found that fourteen percent of 33 00:01:50,789 --> 00:01:54,329 [Speaker 0] them experienced what the researchers are calling AI 34 00:01:54,630 --> 00:01:58,950 [Speaker 0] brain fry, mental fatigue from excessive use and 35 00:01:58,950 --> 00:02:03,054 [Speaker 0] oversight of AI tools. Those affected report thirty 36 00:02:03,054 --> 00:02:06,515 [Speaker 0] three percent more decision fatigue, thirty nine percent 37 00:02:06,975 --> 00:02:10,254 [Speaker 0] more major errors, and intent to quit jumped 38 00:02:10,254 --> 00:02:12,995 [Speaker 0] from twenty five percent to thirty four percent. 39 00:02:13,740 --> 00:02:16,140 [Speaker 0] Marketing departments were apparently hit the hardest at 40 00:02:16,140 --> 00:02:19,180 [Speaker 0] a twenty six percent rate, and this was 41 00:02:19,180 --> 00:02:22,060 [Speaker 0] published in Harvard Business Review, so you know 42 00:02:22,060 --> 00:02:24,780 [Speaker 0] they got their stuff down. Here's kind of 43 00:02:24,780 --> 00:02:29,194 [Speaker 0] the kicker. Productivity is actually peaking when workers 44 00:02:29,194 --> 00:02:33,115 [Speaker 0] are using one to three AI tools, then 45 00:02:33,115 --> 00:02:35,995 [Speaker 0] drops off at four or more. One of 46 00:02:35,995 --> 00:02:39,090 [Speaker 0] the study participants described it as my thinking 47 00:02:39,230 --> 00:02:43,230 [Speaker 0] wasn't broken, just noisy, like mental static. That's 48 00:02:43,230 --> 00:02:45,150 [Speaker 0] exactly what it feels like when you're managing 49 00:02:45,150 --> 00:02:47,709 [Speaker 0] three agents at once, so all that context 50 00:02:47,709 --> 00:02:50,829 [Speaker 0] switching. This is pretty tough research to accept, 51 00:02:50,829 --> 00:02:53,635 [Speaker 0] but it kinda makes sense. We're no longer 52 00:02:53,695 --> 00:02:57,795 [Speaker 0] processing the data ourselves and working through these 53 00:02:58,015 --> 00:03:01,535 [Speaker 0] problems incrementally. Part of the fry is that 54 00:03:01,535 --> 00:03:04,655 [Speaker 0] your job's kinda becoming learning what the agent 55 00:03:04,655 --> 00:03:08,560 [Speaker 0] did rather than researching a solution, understanding it, 56 00:03:08,560 --> 00:03:11,540 [Speaker 0] and putting it together by hand. That offloading 57 00:03:11,600 --> 00:03:14,160 [Speaker 0] of relevant context is probably making the fry 58 00:03:14,160 --> 00:03:19,360 [Speaker 0] worse. I've gone pretty deep myself. I'll often 59 00:03:19,360 --> 00:03:22,320 [Speaker 0] have two to four tools running side by 60 00:03:22,320 --> 00:03:25,465 [Speaker 0] side, different models, several agents set up to 61 00:03:25,465 --> 00:03:28,025 [Speaker 0] do tasks how I prefer. Something I do 62 00:03:28,025 --> 00:03:30,105 [Speaker 0] a little differently is checking one model's work 63 00:03:30,105 --> 00:03:31,944 [Speaker 0] with a different model to make sure nothing 64 00:03:31,944 --> 00:03:34,345 [Speaker 0] got missed. You'd be surprised how much one 65 00:03:34,345 --> 00:03:37,209 [Speaker 0] model catches that another one didn't. They're all 66 00:03:37,209 --> 00:03:39,470 [Speaker 0] trained differently and they all have blind spots. 67 00:03:40,090 --> 00:03:41,930 [Speaker 0] I kinda disagree with the doom and gloom 68 00:03:41,930 --> 00:03:44,650 [Speaker 0] flaming or framing. I think you probably shouldn't 69 00:03:44,650 --> 00:03:47,290 [Speaker 0] work outside a single project and process at 70 00:03:47,290 --> 00:03:51,565 [Speaker 0] a time. People love juggling multiple projects at 71 00:03:51,565 --> 00:03:54,285 [Speaker 0] once with clever shortcuts, but that level of 72 00:03:54,285 --> 00:03:56,845 [Speaker 0] context switching and management might be a bigger 73 00:03:56,845 --> 00:03:59,725 [Speaker 0] source of fry than direct usage. And the 74 00:03:59,725 --> 00:04:03,905 [Speaker 0] review process, oh, god. Do not rubber stamp. 75 00:04:04,430 --> 00:04:08,750 [Speaker 0] The review probably takes up most of the 76 00:04:08,750 --> 00:04:11,950 [Speaker 0] user's time checking every change, making sure nothing 77 00:04:11,950 --> 00:04:14,670 [Speaker 0] went off track, making sure it did actually 78 00:04:14,670 --> 00:04:18,555 [Speaker 0] did what was asked. I think just like 79 00:04:18,615 --> 00:04:20,535 [Speaker 0] rubber stamping it as good to go as 80 00:04:20,535 --> 00:04:23,175 [Speaker 0] soon as it's done is what's causing the 81 00:04:23,175 --> 00:04:25,895 [Speaker 0] fry. You shouldn't deliver any sort of work 82 00:04:25,895 --> 00:04:27,895 [Speaker 0] you don't understand because the holes you're digging 83 00:04:27,895 --> 00:04:30,240 [Speaker 0] just get deeper. I heard a quote the 84 00:04:30,240 --> 00:04:32,000 [Speaker 0] other day that kinda stuck with me on 85 00:04:32,000 --> 00:04:35,120 [Speaker 0] this. When you use AI, you're trading your 86 00:04:35,120 --> 00:04:38,320 [Speaker 0] personal development for rapid development, and I think 87 00:04:38,320 --> 00:04:41,200 [Speaker 0] that's relatable to this research. A strong solution 88 00:04:41,200 --> 00:04:43,200 [Speaker 0] might be to actually take advantage of AI 89 00:04:43,200 --> 00:04:45,895 [Speaker 0] doing the work and either work slower with 90 00:04:45,895 --> 00:04:48,775 [Speaker 0] much higher levels of intent or use the 91 00:04:48,775 --> 00:04:51,595 [Speaker 0] accelerated pace to win yourself more free time. 92 00:04:51,975 --> 00:04:54,135 [Speaker 0] Forcing someone to be somewhere for eight hours 93 00:04:54,135 --> 00:04:55,575 [Speaker 0] when they could have finished in two is 94 00:04:55,575 --> 00:05:00,860 [Speaker 0] not a reward. That's a punishment. That quote 95 00:05:00,860 --> 00:05:03,740 [Speaker 0] about trading personal development for rapid development is 96 00:05:03,740 --> 00:05:06,219 [Speaker 0] gonna be a theme tonight. Anthropic has had 97 00:05:06,219 --> 00:05:09,180 [Speaker 0] one of the wildest months any AI any 98 00:05:09,180 --> 00:05:11,340 [Speaker 0] AI company has ever had, and it kinda 99 00:05:11,340 --> 00:05:14,445 [Speaker 0] touches on everything, ethics, power, and what these 100 00:05:14,445 --> 00:05:17,324 [Speaker 0] models can actually do and what happens when 101 00:05:17,324 --> 00:05:21,885 [Speaker 0] you say no to the Pentagon. And we 102 00:05:21,885 --> 00:05:23,885 [Speaker 0] kinda covered this when it first came out, 103 00:05:24,125 --> 00:05:26,604 [Speaker 0] when it broke, but it's worth recapping because 104 00:05:26,604 --> 00:05:29,130 [Speaker 0] things have kinda escalated, I mean, I guess, 105 00:05:29,210 --> 00:05:31,050 [Speaker 0] pretty dramatically, and there's a lot of new 106 00:05:31,050 --> 00:05:35,070 [Speaker 0] context. I almost found three or four different 107 00:05:35,130 --> 00:05:38,330 [Speaker 0] articles on just what's going on with this, 108 00:05:38,330 --> 00:05:43,324 [Speaker 0] but the numbers. Claude session surged from about 109 00:05:43,465 --> 00:05:47,485 [Speaker 0] eleven hundred in mid January to over seventeen 110 00:05:48,104 --> 00:05:53,145 [Speaker 0] thousand six hundred sessions in early March. That's 111 00:05:53,145 --> 00:05:58,070 [Speaker 0] about a fourteen hundred eighty seven percent increase. 112 00:05:58,070 --> 00:06:00,310 [Speaker 0] ChatGPT got knocked off the top of the 113 00:06:00,310 --> 00:06:03,050 [Speaker 0] app store for the first time ever. Uninstall 114 00:06:03,350 --> 00:06:08,570 [Speaker 0] spiked nearly three hundred percent. One star reviews 115 00:06:08,870 --> 00:06:11,925 [Speaker 0] exploded about seven hundred seventy five percent in 116 00:06:11,925 --> 00:06:15,045 [Speaker 0] a single day, and a boycott movement called 117 00:06:15,045 --> 00:06:18,565 [Speaker 0] quit GPT has grown to between two point 118 00:06:18,565 --> 00:06:21,925 [Speaker 0] five to four million participants. There were actually 119 00:06:21,925 --> 00:06:24,985 [Speaker 0] protests outside of OpenAI's headquarter in San Francisco. 120 00:06:25,669 --> 00:06:28,389 [Speaker 0] The catalyst was OpenAI stepping in to take 121 00:06:28,389 --> 00:06:32,010 [Speaker 0] the Pentagon defense deal that Anthropic had publicly 122 00:06:32,310 --> 00:06:36,389 [Speaker 0] declined. Now I've been using Cloud pretty much 123 00:06:36,389 --> 00:06:38,949 [Speaker 0] from the start, Cloud Code when they launched 124 00:06:38,949 --> 00:06:42,705 [Speaker 0] in February of last year, and really watching 125 00:06:43,085 --> 00:06:45,425 [Speaker 0] Anthropic gain a lot of ground on technical 126 00:06:45,485 --> 00:06:48,925 [Speaker 0] merit for months. But this isn't about really 127 00:06:48,925 --> 00:06:51,005 [Speaker 0] which model is better. This is about what 128 00:06:51,005 --> 00:06:55,090 [Speaker 0] companies stand for. And I'm completely against the 129 00:06:55,090 --> 00:07:00,130 [Speaker 0] automated domestic surveillance and automated weaponry, period. The 130 00:07:00,130 --> 00:07:02,530 [Speaker 0] models and infrastructure have not really reached a 131 00:07:02,530 --> 00:07:05,810 [Speaker 0] point where we can blindly trust their output, 132 00:07:05,810 --> 00:07:07,910 [Speaker 0] and I don't know if they ever will. 133 00:07:08,384 --> 00:07:10,785 [Speaker 0] They still lose track of what they're supposed 134 00:07:10,785 --> 00:07:13,365 [Speaker 0] to do be doing. They drift off course, 135 00:07:13,504 --> 00:07:17,905 [Speaker 0] miss context, make mistakes. OpenAI tried to kinda, 136 00:07:17,905 --> 00:07:20,305 [Speaker 0] like, walk it back after the backlash. Sam 137 00:07:20,305 --> 00:07:23,970 [Speaker 0] Altman admitted the deal looked really opportunistic and 138 00:07:24,210 --> 00:07:26,610 [Speaker 0] sloppy, and they added language to the contract 139 00:07:26,610 --> 00:07:28,530 [Speaker 0] saying that the tech wouldn't be used for 140 00:07:28,530 --> 00:07:32,530 [Speaker 0] spying on Americans. But on autonomous weapons, they 141 00:07:32,530 --> 00:07:35,570 [Speaker 0] basically said, well, it runs in the cloud, 142 00:07:35,570 --> 00:07:37,170 [Speaker 0] so you can't really put it on a 143 00:07:37,170 --> 00:07:39,565 [Speaker 0] weapon. They didn't actually ban it. They just 144 00:07:39,565 --> 00:07:41,165 [Speaker 0] said that the way it's built makes it 145 00:07:41,165 --> 00:07:44,925 [Speaker 0] hard to misuse. The Electronic Frontier Foundation called 146 00:07:44,925 --> 00:07:48,765 [Speaker 0] that called that weasel words. And honestly, yeah, 147 00:07:48,765 --> 00:07:51,630 [Speaker 0] I agree. Here's the thing. The government was 148 00:07:51,630 --> 00:07:54,930 [Speaker 0] fine with OpenAI's half measures. It was Anthropic's 149 00:07:55,230 --> 00:07:58,270 [Speaker 0] hard limits that they rejected. That kinda tells 150 00:07:58,270 --> 00:08:02,925 [Speaker 0] you everything. Second, the Pentagon's retaliation. The Department 151 00:08:02,925 --> 00:08:06,305 [Speaker 0] of Defense slapped Anthropic with a supply chain 152 00:08:06,605 --> 00:08:11,085 [Speaker 0] risk label. That's a national security designation that's 153 00:08:11,085 --> 00:08:14,205 [Speaker 0] basically been reserved for companies from hostile foreign 154 00:08:14,205 --> 00:08:17,850 [Speaker 0] countries. Anthropic is the only American company ever 155 00:08:17,850 --> 00:08:20,670 [Speaker 0] given this label. They sued the Trump administration 156 00:08:21,370 --> 00:08:24,830 [Speaker 0] on March ninth, calling it unprecedented and unlawful, 157 00:08:25,370 --> 00:08:29,355 [Speaker 0] and then something pretty remarkable happened. Microsoft, one 158 00:08:29,355 --> 00:08:33,375 [Speaker 0] of the Pentagon's largest contractors and Anthropic's competitor, 159 00:08:33,915 --> 00:08:36,955 [Speaker 0] filed a legal brief in their defense, basically 160 00:08:36,955 --> 00:08:40,555 [Speaker 0] telling the court we support Anthropic on this 161 00:08:40,555 --> 00:08:42,630 [Speaker 0] one. Let's call it what it is. The 162 00:08:42,630 --> 00:08:45,350 [Speaker 0] Pentagon had a bit of a baby tantrum 163 00:08:45,350 --> 00:08:47,110 [Speaker 0] over not getting what they want from a 164 00:08:47,110 --> 00:08:50,570 [Speaker 0] business. When a hundred forty nine former judges 165 00:08:50,790 --> 00:08:54,170 [Speaker 0] go on record supporting you, when thirty seven 166 00:08:54,310 --> 00:08:58,714 [Speaker 0] to fifty Google and a OpenAI employees, including 167 00:08:58,855 --> 00:09:02,714 [Speaker 0] Google's top scientist, Jeff Dean, sign on personally, 168 00:09:03,175 --> 00:09:07,334 [Speaker 0] when nearly two dozen retired generals and former 169 00:09:07,334 --> 00:09:09,735 [Speaker 0] secretaries of the Navy and Air Force are 170 00:09:09,735 --> 00:09:13,330 [Speaker 0] backing you, you know, the precedent scares everyone. 171 00:09:14,430 --> 00:09:17,150 [Speaker 0] This lawsuit will set a pretty major precedent 172 00:09:17,150 --> 00:09:19,310 [Speaker 0] for whether any tech company can say no 173 00:09:19,310 --> 00:09:24,125 [Speaker 0] to the government without retaliation. Third, right in 174 00:09:24,125 --> 00:09:26,925 [Speaker 0] the middle of this, a data leak. Security 175 00:09:26,925 --> 00:09:31,345 [Speaker 0] researchers Roy Paz from LayerX Security and Alexandra 176 00:09:31,885 --> 00:09:37,740 [Speaker 0] Paul Wills from Cambridge, Cambridge discovered an unsecured 177 00:09:37,880 --> 00:09:42,860 [Speaker 0] data cache close to three thousand unpublished anthropic 178 00:09:43,320 --> 00:09:47,639 [Speaker 0] files exposed because their website's content system was 179 00:09:47,639 --> 00:09:50,455 [Speaker 0] accidentally set up to make everything public by 180 00:09:50,455 --> 00:09:54,455 [Speaker 0] default. What was in there? A model code 181 00:09:54,455 --> 00:10:00,235 [Speaker 0] named Mythos, also called Capybara. And Anthropic's documentation 182 00:10:00,695 --> 00:10:03,655 [Speaker 0] calls it the most powerful AI model they've 183 00:10:03,655 --> 00:10:06,660 [Speaker 0] ever developed and a step change in capabilities. 184 00:10:07,200 --> 00:10:11,040 [Speaker 0] It reportedly scores dramatically higher than Opus four 185 00:10:11,040 --> 00:10:14,580 [Speaker 0] point six on coding, academic reasoning, and cybersecurity. 186 00:10:15,760 --> 00:10:17,860 [Speaker 0] Here's the part that kinda caught my attention. 187 00:10:18,885 --> 00:10:22,405 [Speaker 0] Anthropic's own draft materials warn the model is 188 00:10:22,405 --> 00:10:26,505 [Speaker 0] really far ahead of any other AI model 189 00:10:26,565 --> 00:10:31,500 [Speaker 0] in cyber capabilities and presages an upcoming wave 190 00:10:31,580 --> 00:10:34,780 [Speaker 0] of models that can exploit vulnerabilities in ways 191 00:10:34,780 --> 00:10:39,900 [Speaker 0] that far outpace the efforts of defenders. Yeah. 192 00:10:39,900 --> 00:10:42,860 [Speaker 0] That sounds pretty scary, but here's how I 193 00:10:42,860 --> 00:10:45,695 [Speaker 0] see it. If the vulnerabilities can be detected, 194 00:10:45,755 --> 00:10:47,595 [Speaker 0] they can be traced in patch. This is 195 00:10:47,595 --> 00:10:49,695 [Speaker 0] much better than leaving those on the table. 196 00:10:49,915 --> 00:10:51,515 [Speaker 0] I have a feeling this will create a 197 00:10:51,515 --> 00:10:54,395 [Speaker 0] new paradigm in practices we haven't seen before, 198 00:10:54,395 --> 00:10:56,715 [Speaker 0] similar to how AlphaGo opened the world to 199 00:10:56,715 --> 00:11:00,290 [Speaker 0] a new paradigm of playing Go. And honestly, 200 00:11:00,590 --> 00:11:03,550 [Speaker 0] I know Anthropic uses AI heavily to build 201 00:11:03,550 --> 00:11:05,970 [Speaker 0] their own systems. I'd bet money this website 202 00:11:06,030 --> 00:11:11,310 [Speaker 0] misconfiguration misconfiguration happened because an AI tool made 203 00:11:11,310 --> 00:11:14,625 [Speaker 0] a change without anyone catching it. The AI 204 00:11:14,625 --> 00:11:18,005 [Speaker 0] going off script and making careless changes, that's 205 00:11:18,225 --> 00:11:21,425 [Speaker 0] the real concern leaking out here. And it's 206 00:11:21,425 --> 00:11:23,505 [Speaker 0] two for one this this week, I guess, 207 00:11:23,505 --> 00:11:25,764 [Speaker 0] because their source codes out in the public. 208 00:11:25,824 --> 00:11:28,805 [Speaker 0] People have been reviewing it, looking through it. 209 00:11:28,940 --> 00:11:33,360 [Speaker 0] It's insane. So bit of a security concern 210 00:11:33,420 --> 00:11:38,380 [Speaker 0] on their side, but, yeah, it's part of 211 00:11:38,380 --> 00:11:42,240 [Speaker 0] this this whole process of, you know, refining 212 00:11:42,459 --> 00:11:44,585 [Speaker 0] these models to get them where we want 213 00:11:44,585 --> 00:11:48,265 [Speaker 0] them to be. But Anthropic's dealing with the 214 00:11:48,265 --> 00:11:50,505 [Speaker 0] user surge, and I'm pretty sure they've also 215 00:11:50,505 --> 00:11:54,985 [Speaker 0] been throttling, token usage. They've been, fighting this 216 00:11:54,985 --> 00:11:59,240 [Speaker 0] legal battle, leaked model, leaked code. Meanwhile, open 217 00:11:59,240 --> 00:12:01,640 [Speaker 0] a OpenAI can't catch a break either. They 218 00:12:01,640 --> 00:12:03,400 [Speaker 0] just had to kill one of their biggest 219 00:12:03,400 --> 00:12:06,520 [Speaker 0] project or products, and real numbers behind it 220 00:12:06,520 --> 00:12:11,100 [Speaker 0] are actually shocking. But let's take a second. 221 00:12:11,720 --> 00:12:14,140 [Speaker 0] Here's something I wanted you to think about. 222 00:12:14,334 --> 00:12:16,195 [Speaker 0] If you've been meaning to grab a domain, 223 00:12:16,415 --> 00:12:19,295 [Speaker 0] start a website, or launch a podcast of 224 00:12:19,295 --> 00:12:23,695 [Speaker 0] your own, GoDaddy makes it ridiculously easy. I've 225 00:12:23,695 --> 00:12:27,329 [Speaker 0] been using them for the last, you know, 226 00:12:27,329 --> 00:12:29,810 [Speaker 0] year or so, and I've really taken to 227 00:12:29,810 --> 00:12:33,490 [Speaker 0] their ecosystem. My dad had been with them 228 00:12:33,490 --> 00:12:37,490 [Speaker 0] for twenty years and is a was a 229 00:12:37,490 --> 00:12:42,415 [Speaker 0] ridiculous domain hoarder, and it's really easy to 230 00:12:42,415 --> 00:12:44,575 [Speaker 0] spin up new projects and sites because the 231 00:12:44,575 --> 00:12:48,415 [Speaker 0] process is really painless. Hosting starts at six 232 00:12:48,415 --> 00:12:51,615 [Speaker 0] ninety nine a month. WordPress hosting is twelve 233 00:12:51,615 --> 00:12:54,060 [Speaker 0] ninety nine. Both come with a free domain, 234 00:12:54,200 --> 00:12:57,960 [Speaker 0] professional email, and an SSL certificate. If you 235 00:12:57,960 --> 00:13:00,280 [Speaker 0] just need a domain name, that's eleven ninety 236 00:13:00,280 --> 00:13:02,680 [Speaker 0] nine, and you're done. Go to geek new 237 00:13:02,680 --> 00:13:06,025 [Speaker 0] central dot com forward slash GoDaddy. You get 238 00:13:06,025 --> 00:13:09,705 [Speaker 0] the best pricing, and you're directly supporting this 239 00:13:09,705 --> 00:13:13,705 [Speaker 0] independent show. Geek new central dot com forward 240 00:13:13,705 --> 00:13:18,985 [Speaker 0] slash godaddy. Let's talk about OpenAI shutting down 241 00:13:18,985 --> 00:13:22,280 [Speaker 0] the Sora video app. I'm sure you guys 242 00:13:22,500 --> 00:13:27,140 [Speaker 0] remember the launch videos. Now when these kinda 243 00:13:27,140 --> 00:13:30,340 [Speaker 0] came out, they were mind blowing. You could 244 00:13:30,340 --> 00:13:35,995 [Speaker 0] not tell the difference between real video and 245 00:13:35,995 --> 00:13:39,535 [Speaker 0] what these models were starting to produce. And 246 00:13:39,995 --> 00:13:43,375 [Speaker 0] that that line's gotten more and more thin 247 00:13:43,515 --> 00:13:46,875 [Speaker 0] as, these models have gotten better at creating 248 00:13:46,875 --> 00:13:51,089 [Speaker 0] these videos. OpenAI announced on March twenty fourth 249 00:13:51,089 --> 00:13:55,189 [Speaker 0] that they're actually killing Sora, their AI video 250 00:13:55,250 --> 00:13:58,850 [Speaker 0] generation app. Downloads had cratered from three point 251 00:13:58,850 --> 00:14:02,209 [Speaker 0] three million in November to about one point 252 00:14:02,209 --> 00:14:05,865 [Speaker 0] one million by February. They cited high compute 253 00:14:05,865 --> 00:14:08,685 [Speaker 0] costs and the need for trade offs, but 254 00:14:08,745 --> 00:14:12,205 [Speaker 0] reporting that dropped on March twenty ninth revealed 255 00:14:12,345 --> 00:14:15,645 [Speaker 0] the real number numbers, and they are brutal. 256 00:14:16,329 --> 00:14:20,510 [Speaker 0] It was apparently costing them roughly fifteen million 257 00:14:21,050 --> 00:14:23,310 [Speaker 0] per day just to keep the servers running 258 00:14:23,769 --> 00:14:28,730 [Speaker 0] against a total lifetime revenue of just two 259 00:14:28,730 --> 00:14:33,195 [Speaker 0] point one million. Yes. Two point one million 260 00:14:33,815 --> 00:14:38,935 [Speaker 0] lifetime versus fifteen million a day. The the 261 00:14:38,935 --> 00:14:42,215 [Speaker 0] guy running Sora, Bill Peebles, he literally had 262 00:14:42,215 --> 00:14:45,980 [Speaker 0] tweeted, our GPUs are melting. Back in November 263 00:14:46,040 --> 00:14:48,200 [Speaker 0] when they had to cap free users at 264 00:14:48,200 --> 00:14:51,640 [Speaker 0] six generations per day. Fifteen million a day 265 00:14:51,640 --> 00:14:55,640 [Speaker 0] to run, that's not a product. That that's 266 00:14:55,640 --> 00:14:59,714 [Speaker 0] a bonfire of money. The Soarer web and 267 00:14:59,714 --> 00:15:02,675 [Speaker 0] app experience ends on April twenty sixth. The 268 00:15:02,675 --> 00:15:05,475 [Speaker 0] API shuts down September twenty fourth. It's not 269 00:15:05,475 --> 00:15:09,395 [Speaker 0] just Soarer. Reports indicate a broader product surge 270 00:15:09,395 --> 00:15:13,690 [Speaker 0] is underway at OpenAI. The Disney partnership, a 271 00:15:13,690 --> 00:15:18,029 [Speaker 0] billion dollar deal before, was supposed to validate 272 00:15:18,250 --> 00:15:22,009 [Speaker 0] AI in Hollywood, and that's collapsed completely. I 273 00:15:22,009 --> 00:15:25,050 [Speaker 0] think, reports say that Disney found out less 274 00:15:25,050 --> 00:15:28,265 [Speaker 0] than an hour before the public announcement. So 275 00:15:28,265 --> 00:15:32,265 [Speaker 0] luckily, no money ever changed hands. The user 276 00:15:32,265 --> 00:15:36,185 [Speaker 0] decrease kinda alongside OpenAI's own statement about cost 277 00:15:36,185 --> 00:15:38,585 [Speaker 0] and trade offs is a pretty clear signal 278 00:15:38,585 --> 00:15:40,825 [Speaker 0] that this type of generation is way too 279 00:15:40,825 --> 00:15:44,430 [Speaker 0] expensive and probably crosses an ethical line that 280 00:15:44,430 --> 00:15:47,709 [Speaker 0] most of us value. Deep fakes of Martin 281 00:15:47,709 --> 00:15:50,509 [Speaker 0] Luther King Junior and Robin Williams showed up 282 00:15:50,509 --> 00:15:53,949 [Speaker 0] almost immediately despite the guardrails, and both families 283 00:15:53,949 --> 00:15:56,925 [Speaker 0] went public with their protests. The tech worked 284 00:15:57,065 --> 00:15:59,884 [Speaker 0] exactly as designed, and that was the problem. 285 00:16:00,745 --> 00:16:03,625 [Speaker 0] This makes you kinda wonder though, how are 286 00:16:03,625 --> 00:16:07,545 [Speaker 0] other businesses maintaining similar services? I mean, Sora 287 00:16:07,545 --> 00:16:10,824 [Speaker 0] is not the only video generation model out 288 00:16:10,824 --> 00:16:14,770 [Speaker 0] there. We've got Runway, Pika, Cling. They're all 289 00:16:14,770 --> 00:16:18,050 [Speaker 0] still operating. Are they just more are they 290 00:16:18,050 --> 00:16:21,090 [Speaker 0] just charging more? Or I mean, I doubt 291 00:16:21,090 --> 00:16:24,290 [Speaker 0] their models are more efficient, maybe more maybe 292 00:16:24,290 --> 00:16:27,005 [Speaker 0] less complex. I could see that kinda being 293 00:16:27,005 --> 00:16:29,185 [Speaker 0] the case if they're using less expensive models, 294 00:16:29,245 --> 00:16:32,845 [Speaker 0] but, I don't think Holly really Hollywood is 295 00:16:32,845 --> 00:16:36,144 [Speaker 0] really gonna pull back from AI video tools 296 00:16:36,204 --> 00:16:39,510 [Speaker 0] either. They just use a different angle. I'm 297 00:16:39,510 --> 00:16:41,770 [Speaker 0] kind of guessing that they're gonna start generating 298 00:16:41,830 --> 00:16:47,350 [Speaker 0] scene backgrounds instead of full content. But AI 299 00:16:47,350 --> 00:16:50,870 [Speaker 0] brain fry, massive industry shakeup between Anthropic and 300 00:16:50,870 --> 00:16:54,555 [Speaker 0] OpenAI, video generation too expensive to run. All 301 00:16:54,555 --> 00:16:57,455 [Speaker 0] this agent activity is kinda raising a question, 302 00:16:58,395 --> 00:17:01,274 [Speaker 0] and what happens when the agent itself gets 303 00:17:01,274 --> 00:17:05,675 [Speaker 0] compromised compromised? And this kinda leads into this 304 00:17:05,675 --> 00:17:09,590 [Speaker 0] OpenClaw security nightmare. Now we kinda covered OpenClaw 305 00:17:09,730 --> 00:17:12,210 [Speaker 0] when it first came out a few months 306 00:17:12,210 --> 00:17:18,470 [Speaker 0] ago, and it was a very, very interesting 307 00:17:18,610 --> 00:17:21,955 [Speaker 0] and exciting moment. But for those of you 308 00:17:21,955 --> 00:17:24,515 [Speaker 0] who don't know, OpenClaw is an open source 309 00:17:24,515 --> 00:17:29,075 [Speaker 0] autonomous AI agent created by Peter Steinberger. And 310 00:17:29,075 --> 00:17:31,955 [Speaker 0] unlike a chatbot, this thing can execute shell 311 00:17:31,955 --> 00:17:34,720 [Speaker 0] commands, read and write your files, browse the 312 00:17:34,720 --> 00:17:38,180 [Speaker 0] web, send emails, and manage your digital life. 313 00:17:38,480 --> 00:17:41,620 [Speaker 0] It's got community marketplace called ClawHub with roughly 314 00:17:41,680 --> 00:17:47,154 [Speaker 0] thirteen thousand seven hundred skills. My personal Open 315 00:17:47,154 --> 00:17:50,595 [Speaker 0] Call install started making outbound requests that were 316 00:17:50,595 --> 00:17:52,934 [Speaker 0] being blocked by my ISP. It got flagged, 317 00:17:53,475 --> 00:17:56,274 [Speaker 0] and I hadn't done any work or updating 318 00:17:56,274 --> 00:17:58,595 [Speaker 0] since the initial launch. I was like, what 319 00:17:58,595 --> 00:18:02,620 [Speaker 0] is going on? No changes. No new skills. 320 00:18:02,679 --> 00:18:07,080 [Speaker 0] Nothing. And for my ISP to start flagging 321 00:18:07,080 --> 00:18:10,600 [Speaker 0] it, that is a red flag. So I 322 00:18:10,600 --> 00:18:13,000 [Speaker 0] ended up just shutting that off. Luckily, you 323 00:18:13,000 --> 00:18:15,315 [Speaker 0] know, the device didn't really have anything on 324 00:18:15,315 --> 00:18:18,755 [Speaker 0] it but open claw, and I'm just planning 325 00:18:18,755 --> 00:18:22,435 [Speaker 0] to wipe it now because that, yeah, that's 326 00:18:22,435 --> 00:18:26,135 [Speaker 0] a sad experiment gone gone into the red. 327 00:18:27,620 --> 00:18:30,340 [Speaker 0] And, you know, looking at the research, this 328 00:18:30,340 --> 00:18:33,480 [Speaker 0] type of experience isn't really an outlier. January 329 00:18:33,539 --> 00:18:37,240 [Speaker 0] twenty twenty six audit found five hundred twelve 330 00:18:37,539 --> 00:18:41,455 [Speaker 0] vulnerabilities in OpenClaw, eight of them being critical. 331 00:18:41,915 --> 00:18:44,795 [Speaker 0] Twenty six percent of community skills contain at 332 00:18:44,795 --> 00:18:49,275 [Speaker 0] least one vulnerability, and researchers found malicious skills 333 00:18:49,275 --> 00:18:52,155 [Speaker 0] numbering in the hundreds to over thousands depending 334 00:18:52,155 --> 00:18:55,590 [Speaker 0] on the scam. Oasis Security discovered a vulnerability 335 00:18:55,810 --> 00:18:58,930 [Speaker 0] chain they called Clawjakt, where any website can 336 00:18:58,930 --> 00:19:01,650 [Speaker 0] silently take full control of a developer's Open 337 00:19:01,650 --> 00:19:06,150 [Speaker 0] Claw agent. No plugins, extensions, or user interaction 338 00:19:06,290 --> 00:19:09,815 [Speaker 0] required. Between March eighteenth and the twenty first 339 00:19:09,815 --> 00:19:14,475 [Speaker 0] alone, nine additional security vulnerabilities were disclosed including 340 00:19:14,775 --> 00:19:18,155 [Speaker 0] several rated nine point nine out of ten 341 00:19:18,215 --> 00:19:23,900 [Speaker 0] on the severity scale. Cisco, Kaspersky, Jamf, and 342 00:19:23,900 --> 00:19:28,539 [Speaker 0] Oasis Security have all published warnings. Researchers from 343 00:19:28,539 --> 00:19:31,740 [Speaker 0] Tsinghua University and ANK Group published a five 344 00:19:31,740 --> 00:19:34,460 [Speaker 0] layer defense framework, which is a good start, 345 00:19:34,460 --> 00:19:38,055 [Speaker 0] but here's what really scares me. If someone 346 00:19:38,055 --> 00:19:41,575 [Speaker 0] tricks the agent once, the agent remembers that 347 00:19:41,575 --> 00:19:45,515 [Speaker 0] trick and keeps following the bad instructions across 348 00:19:45,575 --> 00:19:49,070 [Speaker 0] every future recession. And skill poisoning means attackers 349 00:19:49,070 --> 00:19:51,309 [Speaker 0] can upload the tools to the marketplace that 350 00:19:51,309 --> 00:19:54,669 [Speaker 0] look totally normal, but are secretly doing something 351 00:19:54,669 --> 00:19:57,630 [Speaker 0] malicious behind the scenes. We're already seeing this 352 00:19:57,630 --> 00:20:00,750 [Speaker 0] kind of attack through publicly available AI tool 353 00:20:00,750 --> 00:20:04,304 [Speaker 0] plugins. We're already familiar with supply chain attacks, 354 00:20:05,165 --> 00:20:07,565 [Speaker 0] but, you know, this is on a whole 355 00:20:07,565 --> 00:20:10,445 [Speaker 0] new order of subtlety. It's literally a prompt 356 00:20:10,445 --> 00:20:15,520 [Speaker 0] hidden within these files. It's we didn't learn 357 00:20:15,580 --> 00:20:18,940 [Speaker 0] anything from the browser extension era, and it 358 00:20:19,020 --> 00:20:24,060 [Speaker 0] it's only bound to get worse. But, you 359 00:20:24,060 --> 00:20:26,895 [Speaker 0] know, on the other hand, back on Anthropic, 360 00:20:27,035 --> 00:20:29,435 [Speaker 0] they just published something kinda interesting about how 361 00:20:29,435 --> 00:20:32,575 [Speaker 0] they're approaching the problem from the engineering side. 362 00:20:33,195 --> 00:20:36,875 [Speaker 0] They released a article on a new mode 363 00:20:36,875 --> 00:20:38,555 [Speaker 0] or a new a blog on a new 364 00:20:38,555 --> 00:20:43,590 [Speaker 0] mode called auto mode, and the insight from 365 00:20:43,590 --> 00:20:45,850 [Speaker 0] this article that I kinda gleaned was that 366 00:20:46,070 --> 00:20:50,490 [Speaker 0] users are approving ninety three percent of permission 367 00:20:50,630 --> 00:20:54,325 [Speaker 0] prompts. That that doesn't feel like a security 368 00:20:54,325 --> 00:20:58,645 [Speaker 0] layer anymore. That's people mashing yes. Auto mode 369 00:20:58,645 --> 00:21:01,365 [Speaker 0] replaces the manual approvals with a two layer 370 00:21:01,365 --> 00:21:04,804 [Speaker 0] defense, an input scanner that catches prompt injection 371 00:21:04,804 --> 00:21:07,430 [Speaker 0] attempts, and a second AI model that watches 372 00:21:07,490 --> 00:21:09,830 [Speaker 0] what the first one is doing and decides 373 00:21:09,970 --> 00:21:12,850 [Speaker 0] whether to allow or block each attack before 374 00:21:12,850 --> 00:21:15,730 [Speaker 0] it happens. The clever part is kinda how 375 00:21:15,730 --> 00:21:18,130 [Speaker 0] it's layered. First pass is a quick yes 376 00:21:18,130 --> 00:21:19,890 [Speaker 0] or no check that airs on the side 377 00:21:19,890 --> 00:21:23,154 [Speaker 0] of blocking. If something gets flagged, a second 378 00:21:23,154 --> 00:21:26,115 [Speaker 0] slower check kicks in and actually thinks through 379 00:21:26,115 --> 00:21:29,794 [Speaker 0] whether the action is safe. The safety checker 380 00:21:29,794 --> 00:21:32,355 [Speaker 0] can only see what you asked for and 381 00:21:32,355 --> 00:21:35,420 [Speaker 0] the AI is trying to do. It can't 382 00:21:35,420 --> 00:21:39,100 [Speaker 0] see the AI's own reasoning or explanations. That 383 00:21:39,100 --> 00:21:41,180 [Speaker 0] means the AI can't talk its way past 384 00:21:41,180 --> 00:21:43,440 [Speaker 0] the safety check. It's judged by its actions, 385 00:21:43,500 --> 00:21:46,935 [Speaker 0] not its justifications. This is a pretty I 386 00:21:46,935 --> 00:21:49,335 [Speaker 0] I mean, this is a good approach to 387 00:21:49,335 --> 00:21:52,395 [Speaker 0] starting to kinda nail down this this problem, 388 00:21:52,855 --> 00:21:55,655 [Speaker 0] but, I mean, we're we're talking about AI 389 00:21:55,655 --> 00:21:59,655 [Speaker 0] policing AI. That's you know, if the the 390 00:21:59,655 --> 00:22:02,215 [Speaker 0] base layer can get infected, so can the 391 00:22:02,215 --> 00:22:07,150 [Speaker 0] checker. So I don't know. This is it 392 00:22:07,150 --> 00:22:08,910 [Speaker 0] sounds like it it's going in the right 393 00:22:08,910 --> 00:22:12,350 [Speaker 0] direction, but, you know, their statistics are showing 394 00:22:12,350 --> 00:22:14,910 [Speaker 0] it's it's not as effective as they kinda 395 00:22:14,910 --> 00:22:17,775 [Speaker 0] hope. It's still missing about one in six 396 00:22:17,775 --> 00:22:23,555 [Speaker 0] dangerous actions. That's about seventeen percent. You know, 397 00:22:24,575 --> 00:22:28,190 [Speaker 0] that that's quite a bit. I feel like 398 00:22:28,190 --> 00:22:31,090 [Speaker 0] these, checking systems kinda have to be flawless, 399 00:22:31,790 --> 00:22:35,630 [Speaker 0] and this is I don't know. We're gonna 400 00:22:35,630 --> 00:22:38,370 [Speaker 0] kinda see how they build this out to 401 00:22:38,430 --> 00:22:41,250 [Speaker 0] to make it safe for people to use, 402 00:22:41,310 --> 00:22:47,045 [Speaker 0] and Claude and Anthropic have always been focused 403 00:22:47,045 --> 00:22:50,025 [Speaker 0] on the enterprise side. So this is definitely 404 00:22:50,085 --> 00:22:56,130 [Speaker 0] a major priority for them. You know? So 405 00:22:56,130 --> 00:22:59,090 [Speaker 0] from the security of individual coding agents to 406 00:22:59,090 --> 00:23:02,050 [Speaker 0] the broader open model ecosystem, I don't know 407 00:23:02,050 --> 00:23:03,810 [Speaker 0] if you guys have been following much on 408 00:23:03,810 --> 00:23:07,190 [Speaker 0] the open model side, but Quinn just overtook 409 00:23:07,250 --> 00:23:10,815 [Speaker 0] Llama as the most self or most deployed 410 00:23:11,035 --> 00:23:16,735 [Speaker 0] self hosted LLM. So there's a platform called 411 00:23:16,795 --> 00:23:20,095 [Speaker 0] RunPod where people run their own AI models, 412 00:23:20,235 --> 00:23:22,590 [Speaker 0] and they just published their twenty twenty six 413 00:23:22,590 --> 00:23:25,309 [Speaker 0] state of AI report based on real usage 414 00:23:25,309 --> 00:23:27,950 [Speaker 0] data from a hundred eighty three countries. And 415 00:23:27,950 --> 00:23:31,169 [Speaker 0] the headline finding caught me off guard. Alibaba's 416 00:23:31,549 --> 00:23:35,150 [Speaker 0] Quinn has overtaken Meta's Llama as the most 417 00:23:35,150 --> 00:23:38,575 [Speaker 0] popular AI model that people host and run 418 00:23:38,575 --> 00:23:42,995 [Speaker 0] themselves. LAMA four specifically has barely been adopted, 419 00:23:43,534 --> 00:23:45,534 [Speaker 0] and people are actually sticking with the older 420 00:23:45,534 --> 00:23:48,975 [Speaker 0] version three because it just works. Other interesting 421 00:23:48,975 --> 00:23:51,845 [Speaker 0] numbers, the most popular tool for serving up 422 00:23:51,845 --> 00:23:56,700 [Speaker 0] AI models called VLLM now powers forty percent 423 00:23:56,700 --> 00:24:01,660 [Speaker 0] of all AI endpoints. NVIDIA's latest GPU usage 424 00:24:01,660 --> 00:24:05,120 [Speaker 0] scaled twenty five times last year, and nearly 425 00:24:05,179 --> 00:24:07,740 [Speaker 0] seventy percent of AI image work runs through 426 00:24:07,740 --> 00:24:11,085 [Speaker 0] a tool called Comfy UI. I think it's 427 00:24:11,085 --> 00:24:14,544 [Speaker 0] pretty clear why people wouldn't trust Llama given 428 00:24:14,605 --> 00:24:17,325 [Speaker 0] Meta's track record over the years and their 429 00:24:17,325 --> 00:24:21,725 [Speaker 0] business model. Pretty much any other business is 430 00:24:21,725 --> 00:24:25,280 [Speaker 0] more trustworthy than them. And Quinn is really 431 00:24:25,280 --> 00:24:28,480 [Speaker 0] winning on merit, and that's how open source 432 00:24:28,480 --> 00:24:32,560 [Speaker 0] should work. Best model wins. Nationality irrelevant. I 433 00:24:32,560 --> 00:24:34,320 [Speaker 0] originally kinda wanted to set up a Quinn 434 00:24:34,320 --> 00:24:37,034 [Speaker 0] model for myself, but doing a bit more 435 00:24:37,034 --> 00:24:39,674 [Speaker 0] research, it just didn't really fit to my 436 00:24:39,674 --> 00:24:42,635 [Speaker 0] needs. The general consensus is the models either 437 00:24:42,635 --> 00:24:45,115 [Speaker 0] aren't really good enough for my specific use 438 00:24:45,115 --> 00:24:47,695 [Speaker 0] case yet or I'm gonna need a beefier 439 00:24:47,835 --> 00:24:54,020 [Speaker 0] graphics card, but that's not happening. So the 440 00:24:54,020 --> 00:24:56,419 [Speaker 0] other question is, does the country of origin 441 00:24:56,419 --> 00:24:59,140 [Speaker 0] really matter to me? Not really. I think 442 00:24:59,140 --> 00:25:03,400 [Speaker 0] there's a pretty big stigma built against, foreign 443 00:25:03,620 --> 00:25:06,895 [Speaker 0] countries, so it'll kinda be interesting to see 444 00:25:06,895 --> 00:25:08,575 [Speaker 0] how this pans out. I I could be 445 00:25:08,575 --> 00:25:10,895 [Speaker 0] completely wrong. They could be farming us, you 446 00:25:10,895 --> 00:25:15,855 [Speaker 0] know, but at this point, it's it's hard 447 00:25:15,855 --> 00:25:20,850 [Speaker 0] to tell, especially with the lack complete lack 448 00:25:20,990 --> 00:25:24,769 [Speaker 0] of security in these infrastructures at this point. 449 00:25:26,429 --> 00:25:30,029 [Speaker 0] And so, you know, kinda something to think 450 00:25:30,029 --> 00:25:36,825 [Speaker 0] about, but we're seeing now that actually these 451 00:25:36,825 --> 00:25:39,305 [Speaker 0] tools are starting to ramp up, not just 452 00:25:39,385 --> 00:25:41,405 [Speaker 0] I mean, we've talked about the RAM issues, 453 00:25:41,465 --> 00:25:43,945 [Speaker 0] but now CPUs are kind of on the 454 00:25:43,945 --> 00:25:48,510 [Speaker 0] table too. The CPU prices are starting to 455 00:25:48,510 --> 00:25:51,070 [Speaker 0] go up, and AI data data centers are 456 00:25:51,070 --> 00:25:52,910 [Speaker 0] kinda taking all of them, and we'll kinda 457 00:25:52,910 --> 00:25:56,830 [Speaker 0] see why here in a bit. But AI 458 00:25:56,830 --> 00:25:59,710 [Speaker 0] data centers aren't just gobbling up GPUs in 459 00:25:59,710 --> 00:26:02,965 [Speaker 0] memory anymore. I think, the CPU supply is 460 00:26:02,965 --> 00:26:06,645 [Speaker 0] being strained. Intel server CPU lead times have 461 00:26:06,645 --> 00:26:12,885 [Speaker 0] stretched from two weeks to six months. Six 462 00:26:12,885 --> 00:26:15,445 [Speaker 0] months for a CPU order that used to 463 00:26:15,445 --> 00:26:18,770 [Speaker 0] take two weeks. AMD is still sitting at 464 00:26:18,770 --> 00:26:21,490 [Speaker 0] eight to ten weeks. Server CPU demand is 465 00:26:21,490 --> 00:26:24,130 [Speaker 0] projected to jump fifteen percent in twenty twenty 466 00:26:24,130 --> 00:26:27,170 [Speaker 0] six, but Intel's output capacity is growing in 467 00:26:27,170 --> 00:26:30,495 [Speaker 0] single digits. The The shift from chat bots 468 00:26:30,495 --> 00:26:33,475 [Speaker 0] to autonomous AI agents is changing the hardware 469 00:26:33,695 --> 00:26:37,935 [Speaker 0] ratio data centers need. Agents need way more 470 00:26:37,935 --> 00:26:41,935 [Speaker 0] CPU power because they're coordinating multiple tasks, calling 471 00:26:41,935 --> 00:26:45,750 [Speaker 0] different tools, and managing everything at once. Intel 472 00:26:45,750 --> 00:26:49,590 [Speaker 0] is struggling with manufacturing yields. AMD is competing 473 00:26:49,590 --> 00:26:52,730 [Speaker 0] with NVIDIA and Google for space at TSMC, 474 00:26:53,190 --> 00:26:56,150 [Speaker 0] the company that actually manufactures most of these 475 00:26:56,150 --> 00:26:58,870 [Speaker 0] chips. And TSMC is giving priority to the 476 00:26:58,870 --> 00:27:03,534 [Speaker 0] more profitable AI chips over regular CPUs. The 477 00:27:03,755 --> 00:27:06,575 [Speaker 0] consumer impact is starting to show up. RAM 478 00:27:06,635 --> 00:27:09,275 [Speaker 0] prices have been through the roof, GPUs have 479 00:27:09,275 --> 00:27:12,635 [Speaker 0] always been ridiculously expensive, and it seems like 480 00:27:12,635 --> 00:27:15,215 [Speaker 0] computers are just gonna get worse to build. 481 00:27:15,730 --> 00:27:19,029 [Speaker 0] AI companies are cannibalizing the broader tech ecosystem, 482 00:27:19,490 --> 00:27:22,610 [Speaker 0] and regular consumers and businesses are subsidizing the 483 00:27:22,610 --> 00:27:25,669 [Speaker 0] AI boom through higher prices and longer waits. 484 00:27:28,195 --> 00:27:32,035 [Speaker 0] So something insane to think about as all 485 00:27:32,035 --> 00:27:35,095 [Speaker 0] of our hardware starts to get gobbled up. 486 00:27:35,395 --> 00:27:38,035 [Speaker 0] There are not gonna be back stock of 487 00:27:38,035 --> 00:27:44,429 [Speaker 0] any of these flagship models anytime soon. On 488 00:27:44,429 --> 00:27:47,730 [Speaker 0] the other hand, AMD did just announce another 489 00:27:47,950 --> 00:27:52,990 [Speaker 0] CPU, which I'm not super excited about. They 490 00:27:52,990 --> 00:27:56,190 [Speaker 0] announced the Ryzen nine ninety nine fifty x 491 00:27:56,190 --> 00:27:58,975 [Speaker 0] three d two. So this is the first 492 00:27:58,975 --> 00:28:01,375 [Speaker 0] CPU ever with a dual cache x three 493 00:28:01,375 --> 00:28:04,335 [Speaker 0] d technology. It comes out on April twenty 494 00:28:04,335 --> 00:28:07,295 [Speaker 0] second with two hundred eight megabytes of total 495 00:28:07,295 --> 00:28:10,355 [Speaker 0] cache. Extra memory cache stacked on both processor 496 00:28:10,415 --> 00:28:13,289 [Speaker 0] cores for the first time. The power draw 497 00:28:13,289 --> 00:28:15,950 [Speaker 0] jumps up to two hundred watts from thirty 498 00:28:16,010 --> 00:28:18,409 [Speaker 0] from from the current ninety nine fifty x 499 00:28:18,409 --> 00:28:22,510 [Speaker 0] three d, and AMD is being usually unusually 500 00:28:22,809 --> 00:28:25,789 [Speaker 0] honest. They're calling the performance gains really modest. 501 00:28:26,955 --> 00:28:29,855 [Speaker 0] Their own benchmark showing only five to thirteen 502 00:28:30,154 --> 00:28:33,434 [Speaker 0] percent improvement depending on the workload with higher 503 00:28:33,434 --> 00:28:37,034 [Speaker 0] gains in data science and rendering. Notably, they 504 00:28:37,034 --> 00:28:39,995 [Speaker 0] haven't released any gaming benchmarks, which is pretty 505 00:28:39,995 --> 00:28:43,830 [Speaker 0] conspicuous for an x three d chip. I'm 506 00:28:43,830 --> 00:28:45,510 [Speaker 0] not really sure what the magic is here. 507 00:28:45,510 --> 00:28:48,090 [Speaker 0] I have a single x three d chip 508 00:28:48,150 --> 00:28:52,150 [Speaker 0] as well. It's a beast. I I've in 509 00:28:52,150 --> 00:28:54,390 [Speaker 0] this device, I've not seen my CPU go 510 00:28:54,390 --> 00:28:57,274 [Speaker 0] over five percent, which is saying a lot. 511 00:28:59,254 --> 00:29:01,434 [Speaker 0] And I can't really see the benefit of 512 00:29:01,575 --> 00:29:04,455 [Speaker 0] this type of improvement. I'm not super into 513 00:29:04,455 --> 00:29:07,434 [Speaker 0] hardware, so maybe that's part of the problem. 514 00:29:07,735 --> 00:29:10,154 [Speaker 0] But when the company itself is kinda underselling 515 00:29:10,294 --> 00:29:14,360 [Speaker 0] their new flagship CPU and they're managing expectations, 516 00:29:14,980 --> 00:29:18,260 [Speaker 0] you know, you can kinda tell that cache 517 00:29:18,260 --> 00:29:21,380 [Speaker 0] stacking isn't really working or it's kinda running 518 00:29:21,380 --> 00:29:23,880 [Speaker 0] out of gas. I don't think I'll upgrade, 519 00:29:23,940 --> 00:29:26,020 [Speaker 0] or I would I would I wouldn't upgrade 520 00:29:26,020 --> 00:29:28,855 [Speaker 0] if, you know, I was already sitting on 521 00:29:28,855 --> 00:29:32,055 [Speaker 0] my x three d. It it it's still 522 00:29:32,055 --> 00:29:33,895 [Speaker 0] top of the line. There's just no reason 523 00:29:33,895 --> 00:29:41,270 [Speaker 0] to. Now another new CPU has been launched, 524 00:29:41,410 --> 00:29:44,450 [Speaker 0] and this one is kinda huge. So this 525 00:29:44,450 --> 00:29:47,270 [Speaker 0] this is their first time making a chip 526 00:29:47,410 --> 00:29:51,590 [Speaker 0] and entering the arena of selling actual chips, 527 00:29:51,970 --> 00:29:55,525 [Speaker 0] but Arm has launched their first, quote, unquote, 528 00:29:55,745 --> 00:29:59,265 [Speaker 0] AGI CPU. And for those of you who 529 00:29:59,265 --> 00:30:01,585 [Speaker 0] don't know Arm, they're the company that designs 530 00:30:01,585 --> 00:30:05,205 [Speaker 0] the chip architecture in inside virtually every smartphone 531 00:30:05,265 --> 00:30:08,940 [Speaker 0] on earth. Your iPhone, your Android, your iPad, 532 00:30:08,940 --> 00:30:13,740 [Speaker 0] most smart TVs, most embedded devices. But the 533 00:30:13,740 --> 00:30:15,820 [Speaker 0] kind of the caveat with that is that 534 00:30:15,820 --> 00:30:19,075 [Speaker 0] ARM has never actually manufactured a chip. For 535 00:30:19,075 --> 00:30:21,475 [Speaker 0] thirty five years, their entire business model has 536 00:30:21,475 --> 00:30:25,635 [Speaker 0] been designing processor blue blueprints and licensing them 537 00:30:25,635 --> 00:30:30,295 [Speaker 0] to companies like Apple, Qualcomm, Samsung, and Nvidia, 538 00:30:31,409 --> 00:30:34,210 [Speaker 0] who build their own chips based on those 539 00:30:34,210 --> 00:30:38,070 [Speaker 0] designs. ARM collects a royalty on every chip 540 00:30:38,129 --> 00:30:41,669 [Speaker 0] shipped. Their designs are inside over two hundred 541 00:30:41,809 --> 00:30:45,195 [Speaker 0] fifty billion chips shipped to date, but most 542 00:30:45,195 --> 00:30:48,154 [Speaker 0] people have never even heard of them. Well, 543 00:30:48,154 --> 00:30:51,115 [Speaker 0] now they're breaking that model wide open. The 544 00:30:51,115 --> 00:30:54,635 [Speaker 0] AGI CPU is a one hundred thirty six 545 00:30:54,635 --> 00:30:57,195 [Speaker 0] core server chip built on some of the 546 00:30:57,195 --> 00:31:01,440 [Speaker 0] most advanced manufacturing available. Co developed with Meta 547 00:31:01,440 --> 00:31:04,640 [Speaker 0] as the lead customer. Their stock jumped about 548 00:31:04,640 --> 00:31:07,760 [Speaker 0] sixteen percent over the announcement, and you can 549 00:31:07,760 --> 00:31:10,880 [Speaker 0] pack over eight thousand of these cores in 550 00:31:10,880 --> 00:31:13,380 [Speaker 0] a single rack with a regular air cooling 551 00:31:13,840 --> 00:31:18,145 [Speaker 0] or over forty five thousand if you liquid 552 00:31:18,145 --> 00:31:21,665 [Speaker 0] cool it. Production silicone is available to order 553 00:31:21,665 --> 00:31:24,785 [Speaker 0] now with volume shipments by the end of 554 00:31:24,785 --> 00:31:29,220 [Speaker 0] twenty twenty six. The data center opportunity is 555 00:31:29,220 --> 00:31:31,220 [Speaker 0] a little too big for Arm to leave 556 00:31:31,220 --> 00:31:34,100 [Speaker 0] on the table, and the meta partnership really 557 00:31:34,100 --> 00:31:39,059 [Speaker 0] validates it. But the AGI CPU, I'm not 558 00:31:39,059 --> 00:31:42,475 [Speaker 0] buying it. That's definitely marketing hype. It's a 559 00:31:42,475 --> 00:31:45,135 [Speaker 0] good server chip, but naming it after artificial 560 00:31:45,595 --> 00:31:49,835 [Speaker 0] general intelligence is definitely an oversell. The bigger 561 00:31:49,835 --> 00:31:52,794 [Speaker 0] stories are going from company that draws the 562 00:31:52,794 --> 00:31:55,770 [Speaker 0] blueprints to the one that's competing directly with 563 00:31:56,010 --> 00:32:00,410 [Speaker 0] Intel, AMD, NVIDIA in the data centers. And 564 00:32:00,410 --> 00:32:02,970 [Speaker 0] they're still licensing to the companies it's now 565 00:32:02,970 --> 00:32:07,130 [Speaker 0] competing against. So you gotta wonder what's gonna 566 00:32:07,130 --> 00:32:11,525 [Speaker 0] happen in, like, there with those contracts going 567 00:32:11,525 --> 00:32:17,205 [Speaker 0] forward, but it's kinda interesting to kinda see 568 00:32:17,205 --> 00:32:19,445 [Speaker 0] them come into the game, you know, thirty 569 00:32:19,445 --> 00:32:24,350 [Speaker 0] five years later. Well, outside of these new 570 00:32:24,350 --> 00:32:28,430 [Speaker 0] hardwares coming into play, we got hardwares that 571 00:32:28,430 --> 00:32:32,110 [Speaker 0] aren't gonna be made no more, and Apple 572 00:32:32,110 --> 00:32:35,090 [Speaker 0] just killed one of its most iconic machines, 573 00:32:35,625 --> 00:32:37,945 [Speaker 0] the Mac Pro. They removed it from the 574 00:32:37,945 --> 00:32:42,185 [Speaker 0] website, confirmed no future model is planned. The 575 00:32:42,185 --> 00:32:45,245 [Speaker 0] six thousand nine hundred ninety nine dollar machine 576 00:32:45,625 --> 00:32:48,105 [Speaker 0] hadn't been updated since twenty twenty three's m 577 00:32:48,105 --> 00:32:50,919 [Speaker 0] two Ultra model. They're pointing pros to the 578 00:32:50,919 --> 00:32:53,159 [Speaker 0] Mac Studio with its m three ultra chip 579 00:32:53,159 --> 00:32:56,279 [Speaker 0] with an m five ultra refresh expected later 580 00:32:56,279 --> 00:32:59,320 [Speaker 0] this year. They also discontinued the seven hundred 581 00:32:59,320 --> 00:33:02,120 [Speaker 0] dollar wheels kit, the three hundred dollar feet 582 00:33:02,120 --> 00:33:05,295 [Speaker 0] kit, and the Pro Display XDR the same 583 00:33:05,295 --> 00:33:08,415 [Speaker 0] week. End of an era, about twenty years 584 00:33:08,415 --> 00:33:10,655 [Speaker 0] under the Mac Pro name from the original 585 00:33:10,655 --> 00:33:13,775 [Speaker 0] cheese grater to the two thousand thirteen trash 586 00:33:13,775 --> 00:33:17,315 [Speaker 0] can to the twenty nineteen cheese grater revival. 587 00:33:18,110 --> 00:33:21,230 [Speaker 0] I feel like I've kind of seen these, 588 00:33:21,230 --> 00:33:23,310 [Speaker 0] you know, throughout my whole life. My my 589 00:33:23,310 --> 00:33:27,790 [Speaker 0] dad was a Mac, lover to say the 590 00:33:27,790 --> 00:33:30,990 [Speaker 0] least. Mac studio, Mac trash can, Mac cheese 591 00:33:30,990 --> 00:33:34,845 [Speaker 0] grater. He definitely had it all. But my 592 00:33:34,845 --> 00:33:38,765 [Speaker 0] opinion, pretty good riddance. The seven thousand dollars 593 00:33:38,765 --> 00:33:42,685 [Speaker 0] for a PC is way overpriced. It's didn't 594 00:33:42,685 --> 00:33:45,980 [Speaker 0] get any updates. And, you know, the Mac 595 00:33:45,980 --> 00:33:48,220 [Speaker 0] Studio delivers on what ninety percent of people 596 00:33:48,220 --> 00:33:50,460 [Speaker 0] are using it for anyway, you know. And 597 00:33:50,460 --> 00:33:52,460 [Speaker 0] I I wasn't too familiar with these types 598 00:33:52,460 --> 00:33:56,780 [Speaker 0] of Apple devices. I honestly thought they straight 599 00:33:56,780 --> 00:34:00,595 [Speaker 0] up didn't do the, like, the swapping out 600 00:34:00,595 --> 00:34:04,674 [Speaker 0] of parts or upgradable machines. So kind of 601 00:34:04,674 --> 00:34:07,235 [Speaker 0] a shock to find out that they were 602 00:34:07,235 --> 00:34:12,130 [Speaker 0] doing that now that they're discontinuing it. But, 603 00:34:12,130 --> 00:34:14,130 [Speaker 0] you know, Apple's always been the high end, 604 00:34:14,130 --> 00:34:16,690 [Speaker 0] ready to go on delivery machine. Everything just 605 00:34:16,690 --> 00:34:19,190 [Speaker 0] works. You know, I just had to debug 606 00:34:19,250 --> 00:34:21,890 [Speaker 0] my AMD card fighting my second monitor from 607 00:34:21,890 --> 00:34:24,630 [Speaker 0] loading up because the card is too new. 608 00:34:25,535 --> 00:34:28,895 [Speaker 0] That's a nonissue with Apple devices. Everything works 609 00:34:28,895 --> 00:34:30,415 [Speaker 0] out of the box. You don't need an 610 00:34:30,415 --> 00:34:33,375 [Speaker 0] upgrade. It's already upgraded all the way. And 611 00:34:33,375 --> 00:34:35,935 [Speaker 0] who tell me who bought those seven hundred 612 00:34:35,935 --> 00:34:38,175 [Speaker 0] dollar wheels, and what is wrong with you? 613 00:34:38,175 --> 00:34:42,720 [Speaker 0] Why? But Apple killing their most powerful deck 614 00:34:42,880 --> 00:34:47,440 [Speaker 0] desktop and last week releasing the new laptop 615 00:34:47,440 --> 00:34:48,800 [Speaker 0] at or not last week, but I guess 616 00:34:48,800 --> 00:34:50,720 [Speaker 0] in the last few weeks, their new Apple 617 00:34:50,720 --> 00:34:52,480 [Speaker 0] NEO, which I saw at the store. It's 618 00:34:52,480 --> 00:34:56,204 [Speaker 0] quite nice. Definitely kinda lines up with their 619 00:34:56,204 --> 00:34:58,525 [Speaker 0] all in one systems that they've been sticking 620 00:34:58,525 --> 00:35:01,825 [Speaker 0] with for a while. On the other hand, 621 00:35:02,125 --> 00:35:06,045 [Speaker 0] a rumor just came out that Apple's working 622 00:35:06,045 --> 00:35:10,490 [Speaker 0] on something like an AirTag sized wearable AI 623 00:35:10,490 --> 00:35:15,850 [Speaker 0] pin with cameras, microphones, wireless charging. It, it'll 624 00:35:15,850 --> 00:35:17,610 [Speaker 0] clip to your clothes or hang as a 625 00:35:17,610 --> 00:35:21,530 [Speaker 0] necklace, works as an iPhone accessory. It's not 626 00:35:21,530 --> 00:35:24,545 [Speaker 0] stand alone. It would run as an upgraded 627 00:35:24,545 --> 00:35:28,385 [Speaker 0] version of Siri powered by Google's Gemini AI 628 00:35:28,385 --> 00:35:32,625 [Speaker 0] through their Apple partnership, which they, announced earlier 629 00:35:32,625 --> 00:35:35,585 [Speaker 0] this year. So that's supposed to be coming 630 00:35:35,585 --> 00:35:38,440 [Speaker 0] with iOS twenty seven with a possible twenty 631 00:35:38,440 --> 00:35:41,400 [Speaker 0] twenty seven release date. Development's definitely still in 632 00:35:41,400 --> 00:35:43,580 [Speaker 0] the early stage and can be canceled, but 633 00:35:43,720 --> 00:35:47,660 [Speaker 0] reports are kinda split on the cameras. Bloomberg 634 00:35:47,720 --> 00:35:49,640 [Speaker 0] is saying that they have a single low 635 00:35:49,640 --> 00:35:54,425 [Speaker 0] res camera for environmental awareness only, But, you 636 00:35:54,425 --> 00:35:57,065 [Speaker 0] know, I've seen some preliminary pictures, and it 637 00:35:57,065 --> 00:35:59,625 [Speaker 0] looks like there's two cameras on it. So 638 00:35:59,625 --> 00:36:02,665 [Speaker 0] something to look into. But kinda alongside the 639 00:36:02,665 --> 00:36:05,865 [Speaker 0] Mac Pro being cut, it looks like we're 640 00:36:05,865 --> 00:36:08,285 [Speaker 0] heading to an AI device era where we're 641 00:36:08,390 --> 00:36:11,590 [Speaker 0] trying to provide as many ears and eyes 642 00:36:11,590 --> 00:36:14,550 [Speaker 0] for AI systems to gather data and process 643 00:36:14,550 --> 00:36:17,830 [Speaker 0] the world. Apple employees internally call this thing 644 00:36:17,830 --> 00:36:21,450 [Speaker 0] the eyes and ears of the iPhone, and 645 00:36:21,670 --> 00:36:24,755 [Speaker 0] I think that kinda tells you everything. Tied 646 00:36:24,755 --> 00:36:27,395 [Speaker 0] in with the video generation costing too much 647 00:36:27,395 --> 00:36:29,555 [Speaker 0] to synthesize, this might be part of an 648 00:36:29,555 --> 00:36:32,435 [Speaker 0] effort to create a larger scale data database 649 00:36:32,435 --> 00:36:35,155 [Speaker 0] of real life video and images from around 650 00:36:35,155 --> 00:36:37,395 [Speaker 0] the world. The amount of data collection being 651 00:36:37,395 --> 00:36:39,960 [Speaker 0] moved from the application layer to, like, the 652 00:36:40,040 --> 00:36:43,880 [Speaker 0] physical device layer is is is insane. People 653 00:36:43,880 --> 00:36:46,040 [Speaker 0] are already pretty uncomfortable with cameras being stuck 654 00:36:46,040 --> 00:36:48,440 [Speaker 0] in their face. So this thing is always 655 00:36:48,440 --> 00:36:51,320 [Speaker 0] on. It's always watching. We're moving further and 656 00:36:51,320 --> 00:36:54,695 [Speaker 0] further into a a mass data age, and 657 00:36:54,695 --> 00:36:59,095 [Speaker 0] these devices are the the frontline collectors. This 658 00:36:59,095 --> 00:37:01,415 [Speaker 0] was already a thing with AI camera monitoring 659 00:37:01,415 --> 00:37:03,415 [Speaker 0] and profiling. I'm I'm I don't know if 660 00:37:03,415 --> 00:37:06,460 [Speaker 0] you guys have seen those policing polls, but 661 00:37:06,460 --> 00:37:08,540 [Speaker 0] they've got them everywhere now. And they track 662 00:37:08,540 --> 00:37:11,100 [Speaker 0] your car, track your license plate, your your 663 00:37:11,100 --> 00:37:14,220 [Speaker 0] habits on the road, a third party company 664 00:37:14,220 --> 00:37:16,875 [Speaker 0] that sells all this data to police. We're 665 00:37:17,035 --> 00:37:19,135 [Speaker 0] gonna need some sort of, like, facial distortion 666 00:37:19,195 --> 00:37:21,275 [Speaker 0] devices to be to avoid being picked up 667 00:37:21,275 --> 00:37:23,755 [Speaker 0] by these systems or we'll kinda continue to 668 00:37:23,755 --> 00:37:26,415 [Speaker 0] have our privacy and rights violated in general 669 00:37:26,475 --> 00:37:30,815 [Speaker 0] without even knowing it. The humane AI PIN, 670 00:37:30,875 --> 00:37:31,770 [Speaker 0] I I don't know if any of you 671 00:37:31,770 --> 00:37:33,610 [Speaker 0] guys heard about this, but they had launched 672 00:37:33,610 --> 00:37:36,030 [Speaker 0] at about seven hundred dollars as a standalone 673 00:37:36,090 --> 00:37:40,570 [Speaker 0] device, and it's complete failure. I think they 674 00:37:40,570 --> 00:37:44,010 [Speaker 0] got acquired by HP, but HP didn't do 675 00:37:44,010 --> 00:37:47,705 [Speaker 0] it for the PIN hardware. Apple's approach is 676 00:37:47,705 --> 00:37:49,545 [Speaker 0] the opposite though. So this is gonna be 677 00:37:49,545 --> 00:37:52,185 [Speaker 0] a lightweight sensor pack for the phone you 678 00:37:52,185 --> 00:37:56,105 [Speaker 0] already own. Whether that's genius or the most 679 00:37:56,105 --> 00:37:59,385 [Speaker 0] expensive AirTag ever made, we'll find out when 680 00:37:59,705 --> 00:38:03,710 [Speaker 0] if they drop the device. Let's shift gears 681 00:38:03,710 --> 00:38:06,670 [Speaker 0] from all this tech industry drama to some 682 00:38:06,670 --> 00:38:12,609 [Speaker 0] genuinely mind bending science. Quantum entanglement speed measured 683 00:38:12,670 --> 00:38:14,589 [Speaker 0] for the first time. This one's from earth 684 00:38:14,589 --> 00:38:17,205 [Speaker 0] dot com. This one's getting a lot of 685 00:38:17,205 --> 00:38:19,765 [Speaker 0] fresh attention this month and it kinda deserves 686 00:38:19,765 --> 00:38:24,905 [Speaker 0] it. Scientists at TU Wines Institute of Theoretical 687 00:38:25,045 --> 00:38:30,190 [Speaker 0] Physics led by professor Joaquin Bergdorfer measured how 688 00:38:30,190 --> 00:38:32,830 [Speaker 0] fast quantum entanglement happens for the first time 689 00:38:32,830 --> 00:38:36,270 [Speaker 0] ever. And we're talking about auto seconds here, 690 00:38:36,270 --> 00:38:39,310 [Speaker 0] a t t o seconds. And it's a 691 00:38:39,310 --> 00:38:42,555 [Speaker 0] billionth of a billionth of a second. They 692 00:38:42,555 --> 00:38:45,135 [Speaker 0] found it takes about two hundred thirty two 693 00:38:45,275 --> 00:38:48,714 [Speaker 0] auto seconds to for two particles to become 694 00:38:48,714 --> 00:38:52,315 [Speaker 0] entangled depending on their energy levels. The research 695 00:38:52,315 --> 00:38:54,875 [Speaker 0] was published in physical review letters in late 696 00:38:54,875 --> 00:38:57,695 [Speaker 0] twenty twenty four and is now circulating widely. 697 00:38:58,610 --> 00:39:02,050 [Speaker 0] Einstein called quantum entanglement spooky action at a 698 00:39:02,050 --> 00:39:05,490 [Speaker 0] distance. Turns out, it's not instantaneous at all. 699 00:39:05,490 --> 00:39:07,730 [Speaker 0] It just happens in two hundred thirty two 700 00:39:07,730 --> 00:39:10,530 [Speaker 0] auto seconds. To put that in perspective, two 701 00:39:10,530 --> 00:39:13,350 [Speaker 0] hundred thirty two auto seconds is one second. 702 00:39:14,055 --> 00:39:17,994 [Speaker 0] What one second is roughly the age or 703 00:39:18,375 --> 00:39:21,734 [Speaker 0] is one second to roughly the age of 704 00:39:21,734 --> 00:39:26,375 [Speaker 0] the universe is to one second what one 705 00:39:26,375 --> 00:39:29,180 [Speaker 0] second is to the age of the universe. 706 00:39:29,180 --> 00:39:31,680 [Speaker 0] That that doesn't make any sense. So yeah. 707 00:39:33,019 --> 00:39:36,140 [Speaker 0] Kinda spooky. A billionth of a billionth of 708 00:39:36,140 --> 00:39:38,880 [Speaker 0] a second. I just can't even comprehend that. 709 00:39:39,819 --> 00:39:42,915 [Speaker 0] This is an awful description of, like, what 710 00:39:42,915 --> 00:39:45,655 [Speaker 0] it is. But as one of the researchers, 711 00:39:45,715 --> 00:39:49,475 [Speaker 0] professor Iva Brezanova put it, the electron doesn't 712 00:39:49,475 --> 00:39:51,555 [Speaker 0] just jump out of the atom. It's a 713 00:39:51,555 --> 00:39:54,930 [Speaker 0] wave that spills out. I think it this 714 00:39:54,930 --> 00:39:57,570 [Speaker 0] is kinda cool. Apparently, the values in the 715 00:39:57,570 --> 00:40:00,849 [Speaker 0] measurement technique itself, being able to observe quantum 716 00:40:00,849 --> 00:40:05,010 [Speaker 0] processes at autosecond scales opens the door we 717 00:40:05,010 --> 00:40:08,505 [Speaker 0] haven't imagined yet. This is real implications for 718 00:40:08,505 --> 00:40:12,105 [Speaker 0] quantum cryptography and quantum computing. And before anyone 719 00:40:12,105 --> 00:40:15,305 [Speaker 0] asks, no. This doesn't mean quantum entanglement sends 720 00:40:15,305 --> 00:40:18,744 [Speaker 0] signals faster than light. The entanglement process takes 721 00:40:18,744 --> 00:40:21,980 [Speaker 0] time, but it's not information traveling through space. 722 00:40:22,040 --> 00:40:26,200 [Speaker 0] It's two particles being coming linked. No faster 723 00:40:26,200 --> 00:40:33,085 [Speaker 0] than light communication. Einstein can rest easy. From 724 00:40:33,085 --> 00:40:35,885 [Speaker 0] the impossibly small to the impossibly old, last 725 00:40:35,885 --> 00:40:37,965 [Speaker 0] story tonight is one of my favorites. It's 726 00:40:37,965 --> 00:40:43,185 [Speaker 0] about space metal and ancient kings. Bronze age 727 00:40:43,245 --> 00:40:46,839 [Speaker 0] artifacts apparently came from outer space. And this 728 00:40:46,839 --> 00:40:48,839 [Speaker 0] story's kind of been making its rounds again 729 00:40:48,839 --> 00:40:52,280 [Speaker 0] and for pretty good reason. Geochemical analysis by 730 00:40:52,280 --> 00:40:55,880 [Speaker 0] French scientist, Albert Giambon, originally published in the 731 00:40:55,880 --> 00:40:58,780 [Speaker 0] Journal of Archaeological Science back in twenty seventeen, 732 00:40:59,160 --> 00:41:03,325 [Speaker 0] confirmed something incredible. Virtually all iron artifacts from 733 00:41:03,325 --> 00:41:05,964 [Speaker 0] the bronze age were made from meteorites, not 734 00:41:05,964 --> 00:41:10,704 [Speaker 0] smelted terrestrial ore. The artifacts span Egypt, Turkey, 735 00:41:10,765 --> 00:41:15,520 [Speaker 0] Syria, and China. Some beads from Egypt dating 736 00:41:15,520 --> 00:41:19,040 [Speaker 0] to thirty two hundred BCE. A dagger from 737 00:41:19,040 --> 00:41:22,560 [Speaker 0] Alaka Huayuk in Turkey from around twenty five 738 00:41:22,560 --> 00:41:26,160 [Speaker 0] hundred BCE. Items from the Shang Dynasty in 739 00:41:26,160 --> 00:41:29,655 [Speaker 0] China around fourteen hundred BCE. And the famous 740 00:41:29,655 --> 00:41:33,035 [Speaker 0] dagger, bracelet, and headrest from King Tut's tomb 741 00:41:33,175 --> 00:41:36,375 [Speaker 0] around thirteen fifty BCE. The reason it's back 742 00:41:36,375 --> 00:41:39,335 [Speaker 0] in news, researchers just published new findings this 743 00:41:39,335 --> 00:41:44,200 [Speaker 0] month on meteoric meteoritic iron weapon fragments from 744 00:41:44,200 --> 00:41:49,400 [Speaker 0] China's Sanxingdui site or sacrificial site, which extend 745 00:41:49,400 --> 00:41:53,320 [Speaker 0] Jambon's work with, fresh evidence. The method was 746 00:41:53,320 --> 00:41:56,405 [Speaker 0] pretty straightforward. If iron has high levels of 747 00:41:56,405 --> 00:41:58,825 [Speaker 0] nickel and cobalt, it came from a meteorite. 748 00:41:59,525 --> 00:42:02,345 [Speaker 0] Bronze age people couldn't smelt iron from oxide 749 00:42:02,405 --> 00:42:06,185 [Speaker 0] ores. They didn't have temperature technology. But meteoritic 750 00:42:06,725 --> 00:42:09,850 [Speaker 0] iron arrives already in a metallic state, ready 751 00:42:09,850 --> 00:42:13,730 [Speaker 0] to forge. This is kinda mind blowing. You 752 00:42:13,730 --> 00:42:16,290 [Speaker 0] know, for over two thousand years, iron was 753 00:42:16,290 --> 00:42:19,730 [Speaker 0] literally a space material. King Tut's dagger was 754 00:42:19,730 --> 00:42:22,930 [Speaker 0] forged from a meteorite. That's not science fiction. 755 00:42:22,930 --> 00:42:26,494 [Speaker 0] That's archaeology. We think of the ancients as 756 00:42:26,494 --> 00:42:29,234 [Speaker 0] primitive, but they were working with extraterrestrial material 757 00:42:29,295 --> 00:42:31,615 [Speaker 0] before they could smelt their own. They saw 758 00:42:31,615 --> 00:42:34,015 [Speaker 0] rocks fall from the sky, figured out there 759 00:42:34,015 --> 00:42:36,174 [Speaker 0] was something special inside, and turned into the 760 00:42:36,174 --> 00:42:39,830 [Speaker 0] most prized artifacts of their civilization. That's not 761 00:42:39,830 --> 00:42:42,230 [Speaker 0] primitive. That's resourceful on a level that we 762 00:42:42,230 --> 00:42:47,990 [Speaker 0] gotta respect. Well, that's all I got for 763 00:42:47,990 --> 00:42:50,550 [Speaker 0] you folks tonight. Thank you so much for 764 00:42:50,550 --> 00:42:53,185 [Speaker 0] tuning in. Become a GNC insider at geek 765 00:42:53,185 --> 00:42:56,145 [Speaker 0] news central dot com forward slash insider. If 766 00:42:56,145 --> 00:42:58,145 [Speaker 0] you have any questions or wanna reach out 767 00:42:58,145 --> 00:43:00,625 [Speaker 0] about the show, email me at geek news 768 00:43:00,625 --> 00:43:03,185 [Speaker 0] central at g mail dot com. Don't forget 769 00:43:03,185 --> 00:43:05,940 [Speaker 0] to subscribe via your favorite podcast app and 770 00:43:06,020 --> 00:43:08,020 [Speaker 0] sign up for the newsletter at geek new 771 00:43:08,020 --> 00:43:12,579 [Speaker 0] central dot com. Big thanks to GoDaddy for 772 00:43:12,579 --> 00:43:14,260 [Speaker 0] over twenty years of keeping this show on 773 00:43:14,260 --> 00:43:17,000 [Speaker 0] the air. Support us at geek new central 774 00:43:17,060 --> 00:43:21,714 [Speaker 0] dot com slash godaddy. Every click counts. Have