WEBVTT NOTE Transcription provided by Deepgram 00:00:00.399 --> 00:00:02.960 Hello, and welcome to the Geek News Central 00:00:02.960 --> 00:00:06.319 podcast sponsored by GoDaddy dot com. I'm your 00:00:06.319 --> 00:00:08.960 host tonight, Ray Cochran. I've got a great 00:00:08.960 --> 00:00:11.360 episode lined up for you folks. Let's get 00:00:11.360 --> 00:00:15.795 it kicked off. Tonight's leading story, GitHub is 00:00:15.795 --> 00:00:18.215 falling over and not in a small way. 00:00:18.515 --> 00:00:20.935 We're gonna talk about why the world's biggest 00:00:20.994 --> 00:00:23.735 code host had its worst month on record, 00:00:23.955 --> 00:00:25.955 what's actually breaking it, and why this is 00:00:25.955 --> 00:00:28.890 the first of many infrastructure stories that we're 00:00:28.970 --> 00:00:31.930 probably gonna tell in the coming years. Also 00:00:31.930 --> 00:00:35.230 on the show, WARP's terminal goes open source. 00:00:35.530 --> 00:00:39.629 Apple admits it underestimated local AI demand. OpenAI 00:00:40.010 --> 00:00:44.035 explains why ChatGPT got obsessed with goblins. The 00:00:44.035 --> 00:00:46.914 first one x home humanoids are about to 00:00:46.914 --> 00:00:51.954 ship. Tesla finally rolling a real semi off 00:00:51.954 --> 00:00:56.214 a real production line. Chinese EVs whose headlights 00:00:56.434 --> 00:00:59.390 project full color movies, and the final GPS 00:00:59.449 --> 00:01:02.250 three satellite is finally in orbit. Let's get 00:01:02.250 --> 00:01:06.009 into it. Tonight's show is brought to you 00:01:06.009 --> 00:01:10.170 by our twenty year partner, GoDaddy. Exclusive deals 00:01:10.170 --> 00:01:13.344 and pricing at geek news central dot com 00:01:13.344 --> 00:01:19.104 forward slash GoDaddy. Well, GitHub's buckling. It's, probably 00:01:19.104 --> 00:01:22.865 its worst reliability month on record. I've got 00:01:22.865 --> 00:01:25.104 this from a few different sources, the GitHub 00:01:25.104 --> 00:01:29.310 blog, The Register. Lots of similar stories going 00:01:29.310 --> 00:01:31.710 on here, but GitHub's having its worst month 00:01:31.710 --> 00:01:35.170 ever. And a few days ago, the CTO, 00:01:35.230 --> 00:01:37.950 Vlad Fedorov, went up on the company blog 00:01:37.950 --> 00:01:41.090 and basically said, we hear you. We're sorry, 00:01:41.150 --> 00:01:45.235 and here's the plan. And, quote, here we 00:01:45.235 --> 00:01:48.195 hear the pain you're experiencing. We read every 00:01:48.195 --> 00:01:50.835 email, social post, support ticket, and we take 00:01:50.835 --> 00:01:52.755 it all to heart. Then the lines that 00:01:52.755 --> 00:01:56.159 I think that actually matter, our priorities are 00:01:56.159 --> 00:01:59.940 clear, availability first, which has been their issue, 00:02:00.320 --> 00:02:04.479 then capacity, then new features. And that's a 00:02:04.479 --> 00:02:07.119 company telling its customers it's it knows the 00:02:07.119 --> 00:02:09.200 wheels are kind of falling off on the 00:02:09.200 --> 00:02:12.305 wagon. And the wheels really are coming off. 00:02:12.305 --> 00:02:14.945 On April twenty third, the merge queue had 00:02:14.945 --> 00:02:18.965 a regression where a squash merge silently produced 00:02:19.105 --> 00:02:22.385 wrong commits. And that that might sound like 00:02:22.385 --> 00:02:25.025 a complicated things for those of you not 00:02:25.025 --> 00:02:29.560 involved with software development. But, basically, like, the 00:02:29.560 --> 00:02:33.480 the chain the order in which the, code 00:02:33.480 --> 00:02:37.000 changes were processed were incorrect. Something got changed 00:02:37.000 --> 00:02:41.215 in such a way that, previously merged work 00:02:41.275 --> 00:02:45.035 got reverted by later merges. So code that 00:02:45.035 --> 00:02:50.255 was fixed on one day, basically got reverted 00:02:50.395 --> 00:02:52.875 by a future change when it should have 00:02:52.875 --> 00:02:55.890 been that future change should have basically been 00:02:55.890 --> 00:03:00.230 tucked into the change list back where it 00:03:00.370 --> 00:03:03.730 diverged from. That's kinda the easiest way for 00:03:03.730 --> 00:03:05.890 me to put it. Just know that they 00:03:05.890 --> 00:03:09.335 had issues with the main thing that drives 00:03:09.335 --> 00:03:13.575 the their repository site. It wasn't, like, a 00:03:13.575 --> 00:03:16.295 slow site. It was straight up corrupting your 00:03:16.295 --> 00:03:19.015 code changes. So this was a pretty big 00:03:19.015 --> 00:03:21.660 deal. It says a six hundred fifty eight 00:03:21.900 --> 00:03:26.140 repositories were reported to be affected, and two 00:03:26.140 --> 00:03:29.760 thousand ninety two pull requests. So that's quite 00:03:30.300 --> 00:03:33.340 a bit. From some of the other articles 00:03:33.340 --> 00:03:35.660 I was looking at, the number definitely seems 00:03:35.660 --> 00:03:40.485 smaller or bigger than what GitHub's reporting. And 00:03:40.485 --> 00:03:44.105 then, additionally, on the twenty seventh, their Elasticsearch 00:03:44.485 --> 00:03:47.945 cluster just went boom from a botnet attack. 00:03:48.165 --> 00:03:52.490 So they've just been having issues and issues 00:03:53.110 --> 00:03:55.850 all month long pretty much. The search across 00:03:56.070 --> 00:03:58.790 pull requests and issues just stopped showing results 00:03:58.790 --> 00:04:02.925 for hours. Then on the twenty eighth, GitHub 00:04:02.925 --> 00:04:05.165 actions went down, and people couldn't get pull 00:04:05.165 --> 00:04:10.945 requests through. So April uptime by outside reconstructions 00:04:11.485 --> 00:04:15.640 is reportedly under eighty five percent. And that's 00:04:15.640 --> 00:04:18.200 actually the number everyone's talking about because GitHub's 00:04:18.200 --> 00:04:20.860 own status page is kinda sidestepping the issue. 00:04:21.080 --> 00:04:24.540 They don't count degraded performance windows as downtime. 00:04:24.920 --> 00:04:27.880 So their public uptime numbers stay way up 00:04:27.880 --> 00:04:29.975 in the ninety nine percent range, while the 00:04:29.975 --> 00:04:32.855 people actually using the platform are losing hours 00:04:32.855 --> 00:04:35.255 of work a day. The eighty five percent 00:04:35.255 --> 00:04:38.535 figure comes from independent analysts reconstructing what the 00:04:38.535 --> 00:04:40.615 status paid would look like if it were 00:04:40.615 --> 00:04:46.110 honest. So I kinda wanna slow down here 00:04:46.110 --> 00:04:47.470 because I don't think this is really like 00:04:47.470 --> 00:04:49.970 a GitHub is bad story. I haven't personally 00:04:50.190 --> 00:04:53.389 hit on any of these outages, for any 00:04:53.389 --> 00:04:55.069 of my projects, but I think this has 00:04:55.069 --> 00:04:56.669 a lot more to do with those being 00:04:56.669 --> 00:04:59.914 private, not, a lot of branches going from 00:05:00.055 --> 00:05:01.895 my own projects. You know, you generally just 00:05:01.895 --> 00:05:04.775 operate on the main branch when you're working 00:05:04.775 --> 00:05:09.814 on your own stuff. So I I haven't 00:05:09.814 --> 00:05:11.735 personally experienced any of this, but I can 00:05:11.735 --> 00:05:14.470 definitely kinda see the stress that's, you know, 00:05:14.470 --> 00:05:17.270 coming out of this type of issue, seeing, 00:05:17.270 --> 00:05:20.310 like, old code fix fixes that you already 00:05:20.310 --> 00:05:24.730 made getting, like, reverted is just a disaster. 00:05:26.230 --> 00:05:32.764 So, I'm hoping that they can somewhat pull 00:05:32.764 --> 00:05:35.565 it back in. I think, this was a 00:05:35.565 --> 00:05:38.444 big this was a kind of a scaling 00:05:38.444 --> 00:05:40.444 issue from what I understand. So they had 00:05:40.444 --> 00:05:42.444 orig GitHub had originally planned for a ten 00:05:42.444 --> 00:05:46.020 x capacity increase from October twenty twenty five, 00:05:46.320 --> 00:05:49.460 and now they've revised that to thirty x. 00:05:49.920 --> 00:05:53.040 A thirty x scaling target inside eight months 00:05:53.040 --> 00:05:59.595 is is, like, unprecedented. And I think this 00:05:59.595 --> 00:06:01.354 has a lot to do with the amount 00:06:01.354 --> 00:06:04.335 of AI slop that's just being up uploaded 00:06:04.395 --> 00:06:07.275 to the repos, that they get stored on 00:06:07.275 --> 00:06:10.794 GitHub servers. There's so much more code being 00:06:10.794 --> 00:06:13.320 produced now and so much more of it 00:06:13.320 --> 00:06:16.200 that's just being put up on GitHub. So 00:06:16.200 --> 00:06:20.360 I can see some of their struggle with, 00:06:20.360 --> 00:06:23.020 you know, their the server load and consistency 00:06:23.480 --> 00:06:26.200 issues. But at the same time, some of 00:06:26.200 --> 00:06:29.655 their issues, like the GitHub actions going down 00:06:29.655 --> 00:06:32.294 and the merge squash issues are sound like 00:06:32.294 --> 00:06:36.315 issues with using AI or employing AI to 00:06:37.175 --> 00:06:41.675 handle solutions and not properly vetting or reviewing 00:06:41.735 --> 00:06:47.150 the output. So you gotta wonder. But the 00:06:47.150 --> 00:06:49.870 lead defector from GitHub this week is actually 00:06:49.870 --> 00:06:52.370 someone we talked about last week, Mitchell Hashimoto. 00:06:53.630 --> 00:06:56.110 We covered his ghosty terminal landing in Ubuntu 00:06:56.110 --> 00:06:58.645 twenty six zero four. In the same week, 00:06:58.645 --> 00:07:00.965 he's basically pulling off of pulling Ghosty off 00:07:00.965 --> 00:07:03.685 of GitHub entirely, and he's been on GitHub 00:07:03.685 --> 00:07:06.985 since two thousand eight, user twelve ninety nine. 00:07:07.685 --> 00:07:09.945 The the quote that he put out with 00:07:10.164 --> 00:07:14.300 this sort of, planned shift to a new 00:07:14.300 --> 00:07:17.260 platform was I wanna shift software, and it 00:07:17.260 --> 00:07:21.280 doesn't want me to shift software, which is 00:07:21.580 --> 00:07:24.300 a pretty crazy thing to say about a 00:07:24.300 --> 00:07:27.340 system that is meant to store and ship 00:07:27.340 --> 00:07:31.955 your software. So that's kind of the big 00:07:31.955 --> 00:07:36.935 headline right now. Everyone's been talking about these 00:07:37.155 --> 00:07:43.180 GitHub issues and the consistent degradation of, you 00:07:43.180 --> 00:07:47.580 know, system uptime and server availability. And then 00:07:47.580 --> 00:07:50.460 kind of on the opposite side, Warp has 00:07:50.460 --> 00:07:53.660 decided to go open source under AGPL. And 00:07:53.660 --> 00:07:55.260 I I'm not sure if we talked about 00:07:55.260 --> 00:07:58.414 Warp Terminal on the show before, but it's 00:07:58.414 --> 00:08:04.995 a pretty pretty widely known, like, AI driven 00:08:06.735 --> 00:08:12.400 editor tool. They've released under the AGPL license. 00:08:13.020 --> 00:08:17.340 So it's real open source, and they kinda 00:08:17.340 --> 00:08:20.879 did this according to the new stack of 00:08:22.699 --> 00:08:26.139 to stand out amongst their competitors, so to 00:08:26.139 --> 00:08:31.315 speak. They want more people involved in thinking 00:08:31.315 --> 00:08:34.515 that open sourcing it is gonna get more 00:08:34.515 --> 00:08:38.214 people on board with the with the editor, 00:08:38.675 --> 00:08:44.190 and they kind of have shipped this idea 00:08:44.490 --> 00:08:48.170 with the open sourcing of WARP that they 00:08:48.170 --> 00:08:50.250 want the agents to do heavy lifting on 00:08:50.250 --> 00:08:53.550 the code, planning, and testing, and human contributors 00:08:53.770 --> 00:08:59.714 focus on ideas, direction, and verification. And I 00:08:59.714 --> 00:09:02.835 think given, like, the changes that we've been 00:09:02.835 --> 00:09:07.075 seeing with GitHub, we can only hope that 00:09:07.075 --> 00:09:11.075 their service kind of recovers for this type 00:09:11.075 --> 00:09:16.180 of human collaboration, so to speak. And even 00:09:16.180 --> 00:09:19.060 then, I don't know if humans approving is 00:09:19.060 --> 00:09:24.260 really enough by itself. I'd just been back 00:09:24.260 --> 00:09:27.695 and forth on this concept of how involved 00:09:27.915 --> 00:09:30.475 you should be when actually you're writing the 00:09:30.475 --> 00:09:34.875 code. And it's become more and more clear 00:09:34.875 --> 00:09:38.075 to me that the more hands off I 00:09:38.075 --> 00:09:41.750 get with it, the less intimately I understand 00:09:41.890 --> 00:09:44.370 the system and the less intimately I can 00:09:44.370 --> 00:09:47.089 mentally model it. And I think that's a 00:09:47.089 --> 00:09:50.769 big issue in terms of understanding how you 00:09:50.769 --> 00:09:53.830 want the system to grow, you know, infrastructure 00:09:54.209 --> 00:09:58.905 problems, security issues, all of that. It's pretty 00:09:58.905 --> 00:10:00.345 hard to keep up with if you're not 00:10:00.345 --> 00:10:02.265 even writing any of the code. And, yeah, 00:10:02.265 --> 00:10:04.185 sure, you can have the model review it. 00:10:04.185 --> 00:10:06.365 You can have the model suggest a refactor, 00:10:07.385 --> 00:10:11.805 but there's just something lacking in terms of 00:10:13.920 --> 00:10:20.080 solid foundational structure when you put together a 00:10:20.080 --> 00:10:25.460 project purely through VibeCoding, as it's called. So 00:10:25.920 --> 00:10:29.405 I I don't know where I'm at with 00:10:29.405 --> 00:10:33.725 this still. It's a pretty tricky topic to 00:10:33.725 --> 00:10:37.565 kinda navigate, especially with how great the models 00:10:37.565 --> 00:10:39.805 are becoming, but at the same time, the 00:10:39.805 --> 00:10:45.930 the limitations that are definitely starting to grow 00:10:45.930 --> 00:10:51.930 and and be I identified clearly. So I 00:10:51.930 --> 00:10:56.010 think that these two conflicting stories with one 00:10:56.010 --> 00:11:02.084 terminal leaving, GitHub and another, you know, dev 00:11:02.084 --> 00:11:09.285 tool warp going, open source. So it'd be 00:11:09.285 --> 00:11:12.490 kinda cool if you guys are into open 00:11:12.490 --> 00:11:15.529 source tooling, then definitely check out the the 00:11:15.529 --> 00:11:18.649 new open source stack that Warp's got on 00:11:18.649 --> 00:11:23.130 GitHub. Pretty sweet project to be involved with 00:11:23.130 --> 00:11:27.125 if you wanna get into editor tooling or, 00:11:27.345 --> 00:11:29.985 you know, agentic workflows. I think that's pretty 00:11:29.985 --> 00:11:32.145 sweet. You can definitely make a fork of 00:11:32.145 --> 00:11:35.904 it and create your own fun tool. You 00:11:35.904 --> 00:11:40.459 know? So something to look into. But let's 00:11:40.459 --> 00:11:42.779 move on. I think I got one more 00:11:42.779 --> 00:11:45.100 in terms of, like, the AI tooling boom, 00:11:45.100 --> 00:11:49.180 but Apple just told Wall Street that the 00:11:49.180 --> 00:11:53.519 local AI demand caught them completely flat footed. 00:11:54.675 --> 00:11:56.595 And I'm sure that you guys saw the 00:11:56.595 --> 00:12:00.195 boom that came with the OpenClaw drama and 00:12:00.195 --> 00:12:02.615 how all the Mac Minis and Mac Studios 00:12:02.675 --> 00:12:07.050 were just getting bought up for these AI 00:12:07.050 --> 00:12:13.210 driven workflows or, you know, systems, and I 00:12:13.210 --> 00:12:15.290 I didn't expect this type of thing. I 00:12:15.290 --> 00:12:17.930 didn't think that Mac minis and studios would 00:12:17.930 --> 00:12:21.050 be the go to device for this type 00:12:21.050 --> 00:12:25.154 of work, but Tim Cook basically said both 00:12:25.154 --> 00:12:28.115 of these are amazing platforms for AI and 00:12:28.115 --> 00:12:30.834 agentic tooling, and the customer recognition of that 00:12:30.834 --> 00:12:33.175 is happening faster than what we had predicted. 00:12:34.470 --> 00:12:37.190 So they just didn't expect all these people 00:12:37.190 --> 00:12:41.050 to wanna run, AI workloads on their machines, 00:12:41.190 --> 00:12:44.410 and I find that a little bit shocking. 00:12:45.190 --> 00:12:49.685 I somewhat consider Apple devices approachable Linux systems 00:12:49.745 --> 00:12:51.985 even though they're not. But at the same 00:12:51.985 --> 00:12:55.185 time, they kinda are. But outside of that, 00:12:55.185 --> 00:12:59.045 you know, Apple had done some pretty interesting 00:12:59.265 --> 00:13:02.740 supply moves. They discontinued the five twelve gigabyte 00:13:02.800 --> 00:13:06.880 Mac Studio upgrade option in early March. They 00:13:06.880 --> 00:13:09.680 raised the two hundred fifty six gigabyte price 00:13:09.680 --> 00:13:12.399 by four hundred dollars. You know, as of 00:13:12.399 --> 00:13:14.640 April eleventh, the Mac mini at thirty two 00:13:14.640 --> 00:13:16.725 and sixty four gigs and the Mac Studio 00:13:16.725 --> 00:13:18.644 at one twenty eight and two fifty six 00:13:18.644 --> 00:13:22.665 gigs all went currently unavailable on the online 00:13:22.725 --> 00:13:27.445 store. And, you know, the estimated delivery times 00:13:27.445 --> 00:13:30.690 for some of the upgraded models are four 00:13:30.690 --> 00:13:33.210 to five months, and Tim Cook is basically 00:13:33.210 --> 00:13:36.110 saying that the bottleneck is the advanced TSMC 00:13:36.810 --> 00:13:42.650 nodes for the SOCs, not memory. Sorry. I'm 00:13:42.650 --> 00:13:46.030 not too hardware in-depth. So if you guys 00:13:47.345 --> 00:13:50.404 know what a TSMC node or the SOC 00:13:50.464 --> 00:13:54.705 is, hit us with a comment. But I 00:13:54.705 --> 00:13:57.125 just wonder. I don't think the original target 00:13:57.265 --> 00:14:01.390 consumer for hardware like Mac minis was, like, 00:14:01.390 --> 00:14:03.970 thought to be the, like, local AI crowd. 00:14:04.830 --> 00:14:06.990 I had kinda considered Mac minis to be, 00:14:06.990 --> 00:14:10.750 like, the cheap entry point into Apple's tech 00:14:10.750 --> 00:14:15.824 stack, and this was obviously before the Apple 00:14:15.824 --> 00:14:21.764 Neo got released, but these are fantastic machines. 00:14:21.824 --> 00:14:24.625 They have great performance, and you don't get 00:14:24.625 --> 00:14:27.425 a lot of compatibility sacrifices when you work 00:14:27.425 --> 00:14:31.670 on Apple's OS. And buyers just like those 00:14:31.890 --> 00:14:33.810 types of systems that you can just open 00:14:33.810 --> 00:14:36.070 up and it gets working. That's why Microsoft 00:14:36.130 --> 00:14:38.390 and Apple have been successful for so long. 00:14:39.010 --> 00:14:42.790 You know, they have integrated stacks that work 00:14:43.095 --> 00:14:48.714 on most systems, right, if not all. But 00:14:50.055 --> 00:14:53.675 it's interesting to now see the the OpenClaw 00:14:53.894 --> 00:14:56.295 wave taking over so hard and all these 00:14:56.295 --> 00:15:01.500 people buying up all this hardware to be 00:15:01.500 --> 00:15:05.120 able to use, you know, systems like OpenClaw 00:15:05.420 --> 00:15:08.720 for, you know, their own daily automation and, 00:15:09.340 --> 00:15:12.800 you know, long term memory and stored context. 00:15:13.745 --> 00:15:16.065 And I don't know. I'm still pretty much 00:15:16.065 --> 00:15:18.305 on the fence with the OpenClaw stuff, especially 00:15:18.305 --> 00:15:20.625 with that scare I had talked about a 00:15:20.625 --> 00:15:25.185 few weeks back where my own, OpenClaw install 00:15:25.185 --> 00:15:28.700 on my little mini PC started making weird 00:15:29.000 --> 00:15:34.840 requests. But that kind of tooling alongside, like, 00:15:34.840 --> 00:15:38.460 the constant stream of prompt injections and leaks 00:15:38.840 --> 00:15:43.505 and just security issues galore. I I think 00:15:43.505 --> 00:15:45.425 I'll wait for the ecosystem to get a 00:15:45.425 --> 00:15:47.765 little further ahead before I really lean in. 00:15:48.625 --> 00:15:51.105 I'm not really in a rush to start 00:15:51.105 --> 00:15:53.745 working through an open claw system, especially because 00:15:53.745 --> 00:15:58.519 someone's gonna deliver some sort of integrated system 00:15:58.519 --> 00:16:03.320 that works much more clearly than something like 00:16:03.320 --> 00:16:06.459 OpenClaw does, you know, just UI bound, and 00:16:06.760 --> 00:16:10.860 that's really easy to get working with. So 00:16:12.575 --> 00:16:16.895 it'll be kinda interesting. I wonder how they'll 00:16:16.895 --> 00:16:19.935 kind of adapt to this shortage or, yeah, 00:16:19.935 --> 00:16:23.715 bottleneck in terms of their device output. But 00:16:25.730 --> 00:16:29.970 I think pretty much all the companies are 00:16:29.970 --> 00:16:33.990 competing for hardware right now and more silicon. 00:16:34.130 --> 00:16:39.090 So we'll see we'll see how this kinda 00:16:39.090 --> 00:16:43.625 pans out or whether device, ship times ever 00:16:43.625 --> 00:16:46.205 come back to what it was, especially with 00:16:46.425 --> 00:16:51.145 the availability of tools like Open Claw. But 00:16:51.145 --> 00:16:54.009 let's move on. I don't know if you 00:16:54.009 --> 00:16:56.170 guys caught on to this weird trend that 00:16:56.170 --> 00:16:57.850 had been going around for a little bit, 00:16:57.850 --> 00:17:00.269 but OpenAI just published one of the most 00:17:00.810 --> 00:17:04.730 interesting explainers of how their models behave or, 00:17:04.730 --> 00:17:08.385 like, was trained. So if you use ChatGPT 00:17:08.845 --> 00:17:11.805 after the GPT five point one launch, you 00:17:11.805 --> 00:17:14.525 might have noticed it kept calling things or 00:17:14.525 --> 00:17:19.505 referring to goblins. Goblins were everywhere. Goblin metaphors. 00:17:20.340 --> 00:17:23.700 The model the models just couldn't stop. And 00:17:23.700 --> 00:17:29.620 OpenAI published some goblin usage numbers, and the 00:17:29.620 --> 00:17:32.835 usage in chat tbt responses jumped roughly a 00:17:33.075 --> 00:17:36.995 hundred seventy five percent after GPD five point 00:17:36.995 --> 00:17:41.235 one launched. Gremlin, as a word, rose fifty 00:17:41.235 --> 00:17:45.875 two percent. And how they kinda traced it 00:17:45.875 --> 00:17:48.620 down and tracked it back was was really 00:17:48.620 --> 00:17:51.500 interesting to read about, and they had kinda 00:17:51.500 --> 00:17:54.320 tracked it back to the optional nerdy personality. 00:17:55.740 --> 00:17:58.620 The nerdy personality was about two point five 00:17:58.620 --> 00:18:01.605 percent of all ChatGPT responses, but it was 00:18:01.605 --> 00:18:05.365 responsible for sixty six point seven of all 00:18:05.365 --> 00:18:08.885 the goblin mentions. Two thirds of the bias 00:18:08.885 --> 00:18:12.485 came from one person's one personality preset that 00:18:12.485 --> 00:18:14.585 was less than three percent of the traffic. 00:18:16.850 --> 00:18:19.650 Just think about that. Two thirds of the 00:18:19.650 --> 00:18:23.490 bias came from one personality preset that was 00:18:23.490 --> 00:18:28.150 less than three percent of traffic. The cause 00:18:28.290 --> 00:18:31.044 in Open AI's own words is that the 00:18:31.044 --> 00:18:35.065 reward signal during the personality training quietly favored 00:18:35.125 --> 00:18:38.804 creature metaphors, and then through later supervised fine 00:18:38.804 --> 00:18:41.445 tuning, that preference leaked out of the nerdy 00:18:41.445 --> 00:18:44.230 mode into the rest of the model. And 00:18:44.230 --> 00:18:48.070 here's a key quote. Reinforcement learning does not 00:18:48.070 --> 00:18:51.029 guarantee that learned behavior stay neatly scoped to 00:18:51.029 --> 00:18:55.190 the condition that produced them. And that is 00:18:55.190 --> 00:18:58.865 a sentence that should probably be tattooed on 00:18:58.865 --> 00:19:02.304 the inside of every AI engineer's eyelids. They 00:19:02.304 --> 00:19:04.945 ended up retiring the nerdy personality in March 00:19:04.945 --> 00:19:07.525 after the launch of GPT five point four, 00:19:07.745 --> 00:19:12.565 which also removed the goblin affine reward signal, 00:19:13.110 --> 00:19:16.710 filtered creature words out of training data. And 00:19:16.710 --> 00:19:19.510 then when they tested GPT five point five 00:19:19.510 --> 00:19:22.950 in codex, the goblin tick came back, and 00:19:22.950 --> 00:19:25.750 they had to write an explicit system prompt 00:19:25.750 --> 00:19:29.155 rule, never talk about goblins, goblins, gremlins, raccoons, 00:19:29.375 --> 00:19:34.895 trolls, ogres, or pigeons. And I I think 00:19:34.895 --> 00:19:39.455 this is an a a great overarching signal 00:19:39.455 --> 00:19:43.030 into how LLMs operate, and it's sort of 00:19:43.030 --> 00:19:46.650 the beauty and the disaster of pattern matching. 00:19:46.950 --> 00:19:49.910 The Goblin reward is a really fantastic example 00:19:49.910 --> 00:19:52.870 of how, you know, problematic training patterns can 00:19:52.870 --> 00:19:55.995 emerge and how much research actually goes into 00:19:55.995 --> 00:19:59.115 discovering and managing these types of issues when 00:19:59.115 --> 00:20:02.794 you start to see model output. I kinda 00:20:02.794 --> 00:20:04.875 would think that it's not as easy as, 00:20:04.875 --> 00:20:09.250 you know, editing lines of code as it's 00:20:09.250 --> 00:20:11.170 it's much more of a black box. You 00:20:11.170 --> 00:20:13.490 know, the output is a result of a 00:20:13.490 --> 00:20:17.670 pattern abstraction, not not a static function with 00:20:17.730 --> 00:20:21.775 consistent inputs and outputs. The LLMs don't have 00:20:21.775 --> 00:20:24.195 consistent outputs. So when you find a problem, 00:20:24.415 --> 00:20:28.995 you can't directly fix it. You can only 00:20:29.055 --> 00:20:32.675 somewhat patch it or retrain the models. And 00:20:32.940 --> 00:20:35.259 the more patches that you have to do, 00:20:35.259 --> 00:20:37.019 you know, those just start to add up. 00:20:37.019 --> 00:20:39.019 You just think about it. They literally had 00:20:39.019 --> 00:20:41.259 to add a prompt rule that says never 00:20:41.259 --> 00:20:45.039 talk about goblins, gremlins, raccoons, trolls, ogres, pigeons. 00:20:45.695 --> 00:20:48.415 And then what happens five versions from now 00:20:48.415 --> 00:20:51.775 when there's, you know, something else that that 00:20:51.775 --> 00:20:55.795 they consider a tick? Will that just be 00:20:56.015 --> 00:20:59.700 prompted out as well, or will they have 00:20:59.700 --> 00:21:04.340 to revisit the infrastructure of their training and 00:21:04.340 --> 00:21:06.900 how they're managing their data? I wonder. I 00:21:06.900 --> 00:21:11.275 don't know. I wonder. But I think it's 00:21:11.275 --> 00:21:13.275 kinda interesting to see that, you know, that 00:21:13.275 --> 00:21:18.635 nerdy personality really leak into the outputs with 00:21:18.635 --> 00:21:24.780 those goblins and gremlins, but I think that's 00:21:24.780 --> 00:21:28.700 cool. Yeah. Great, blog by OpenAI. Definitely check 00:21:28.700 --> 00:21:31.420 that out. Let's talk about GoDaddy for a 00:21:31.420 --> 00:21:34.860 second. GoDaddy has been sponsoring this show for 00:21:34.860 --> 00:21:37.179 over twenty years. Just let that sink in 00:21:37.179 --> 00:21:40.445 for a second. Twenty years. And this is 00:21:40.445 --> 00:21:43.025 before most people knew what a podcast was. 00:21:43.085 --> 00:21:46.125 GoDaddy was ready to back this show. I 00:21:46.125 --> 00:21:49.005 use GoDaddy for my domains, my hosting, my 00:21:49.005 --> 00:21:51.970 email, the whole stack. I've tried other platforms 00:21:51.970 --> 00:21:53.890 over the years, and I keep coming back 00:21:53.890 --> 00:21:55.730 because it works. And it's a the support 00:21:55.730 --> 00:21:57.970 is there when I need it. If you 00:21:57.970 --> 00:22:00.929 need a domain, hosting, or a website, head 00:22:00.929 --> 00:22:03.170 to geek new central dot com forward slash 00:22:03.170 --> 00:22:06.455 GoDaddy. You'll get exclusive pricing and every purchase 00:22:06.455 --> 00:22:09.434 directly supports this show. That's geek new central 00:22:09.495 --> 00:22:14.475 dot com forward slash GoDaddy. Here's, an interesting 00:22:15.174 --> 00:22:18.740 new report from Bloomberg. A company called OneX 00:22:19.040 --> 00:22:23.280 has just opened a US humanoid factory in, 00:22:23.280 --> 00:22:28.640 I think, Hayward, California. So it's a fifty 00:22:28.640 --> 00:22:33.674 eight thousand square foot humanoid robot factory. One 00:22:33.674 --> 00:22:37.355 x is a Norway founded company, but it's 00:22:37.355 --> 00:22:40.235 OpenAI backed, so they're getting some money from 00:22:40.235 --> 00:22:44.174 OpenAI. But they're calling this America's first vertically 00:22:44.394 --> 00:22:51.740 integrated humanoid robot factory. So I guess we're 00:22:51.800 --> 00:22:56.280 starting to compete against China. We've been talking 00:22:56.280 --> 00:23:00.300 about those walkers with Walker x twos for 00:23:01.365 --> 00:23:04.805 months now. You know, I've covered several several 00:23:04.805 --> 00:23:08.425 stories on, you know, China and their humanoid 00:23:08.645 --> 00:23:14.965 robot projects, and somewhat exciting to see this 00:23:14.965 --> 00:23:17.890 finally making it to the US. A little 00:23:17.890 --> 00:23:23.970 bit terrifying as well, but their goal or 00:23:23.970 --> 00:23:28.450 the capacity of this new facility is ten 00:23:28.450 --> 00:23:33.804 thousand NEO home humanoids in year one, and 00:23:33.804 --> 00:23:36.205 they're also building a much bigger two hundred 00:23:36.205 --> 00:23:38.684 thirty one thousand square foot facility in San 00:23:38.684 --> 00:23:40.924 Carlos. And their goal by the end of 00:23:40.924 --> 00:23:43.265 twenty twenty seven is a hundred k robots. 00:23:44.570 --> 00:23:46.890 Now their their robot, I guess, is called 00:23:46.890 --> 00:23:51.930 Neo, is, there's two price points. So one 00:23:51.930 --> 00:23:54.590 is straight twenty k. You can buy it, 00:23:55.050 --> 00:23:57.725 for a one time deal, or you can 00:23:57.725 --> 00:24:00.045 pay a four ninety nine, and I mean 00:24:00.045 --> 00:24:04.065 four hundred and ninety nine dollars a month 00:24:04.925 --> 00:24:10.740 to lease it. And the crazy part and 00:24:10.740 --> 00:24:12.899 I didn't expect this at all. The first 00:24:12.899 --> 00:24:17.299 ten thousand unit allocation has reportedly sold out 00:24:17.299 --> 00:24:20.580 in five days when preorders opened back in 00:24:20.580 --> 00:24:26.055 October. So each Neo runs on NVIDIA's Jetson 00:24:26.115 --> 00:24:31.315 Thor onboard. Wow. So they they sold ten 00:24:31.315 --> 00:24:34.915 thousand units already, and they haven't even they're 00:24:34.915 --> 00:24:37.335 not even they haven't even started yet. Right? 00:24:38.610 --> 00:24:40.610 I guess they just finished the factory, but 00:24:40.610 --> 00:24:42.929 still, like, they haven't even started really building 00:24:42.929 --> 00:24:47.809 the robots. But, you know, a few months 00:24:47.809 --> 00:24:51.170 back, we talked about how the advent of 00:24:51.170 --> 00:24:54.470 smart homes is probably coming back, and I 00:24:54.664 --> 00:24:57.965 feel like this device might be one of 00:24:58.424 --> 00:25:01.544 the main gadgets featured in a smart home 00:25:01.544 --> 00:25:05.865 system. You know? It would be something like 00:25:05.865 --> 00:25:07.945 the manager, and I think that'd be kinda 00:25:07.945 --> 00:25:11.669 cool to see I don't know. Maybe just 00:25:11.669 --> 00:25:12.870 to see. I'm not sure if I would 00:25:12.870 --> 00:25:15.429 wanna live that or experience it. I don't 00:25:15.429 --> 00:25:19.049 think I'd really want this type of robot, 00:25:19.110 --> 00:25:21.110 nor would I be willing to pay either 00:25:21.110 --> 00:25:24.655 twenty k or five hundred a month. I'm 00:25:24.655 --> 00:25:27.935 not sure what kind of problem this really 00:25:27.935 --> 00:25:31.295 solves for the everyday person. I do think 00:25:31.295 --> 00:25:36.675 it's cool, but I I just don't know 00:25:38.010 --> 00:25:40.830 how truly effective something like this will be. 00:25:41.130 --> 00:25:44.730 And, you know, you gotta wonder because, you 00:25:44.730 --> 00:25:48.190 know, things like or devices like the iPhone 00:25:48.330 --> 00:25:52.875 really revolutionized, you know, the foam. I don't 00:25:52.875 --> 00:25:57.755 see this robot revolutionizing the home or home 00:25:57.755 --> 00:26:02.715 dynamics. You know? In fact, I there's probably 00:26:02.715 --> 00:26:08.520 an argument that that this type of, you 00:26:08.520 --> 00:26:11.660 know, dependency or change might be actually worse 00:26:12.200 --> 00:26:15.020 for homes. And now I could definitely see 00:26:15.080 --> 00:26:18.220 a place for, you know, elderly or disabled 00:26:18.280 --> 00:26:22.245 individuals to find real help with a device 00:26:22.245 --> 00:26:27.044 like this. But I don't know. With with 00:26:27.044 --> 00:26:30.325 how much people have kinda gone crazy over 00:26:30.325 --> 00:26:34.005 the LLM movement and, you know, people even 00:26:34.005 --> 00:26:36.840 going as far as you know, falling in 00:26:36.840 --> 00:26:40.680 love with, you know, the LLMs for one, 00:26:40.680 --> 00:26:43.720 and two, getting into this idea that you 00:26:43.720 --> 00:26:49.615 can, quote, unquote, awaken the LLM and have 00:26:49.615 --> 00:26:53.414 it, you know, be on your side, quote, 00:26:53.414 --> 00:26:57.495 unquote. It it it is mildly terrifying to 00:26:57.495 --> 00:26:59.995 see people really lean into it that hard. 00:27:01.815 --> 00:27:07.600 I find it terrifying. But I don't think 00:27:07.600 --> 00:27:11.200 I'd let this one x robot drive my 00:27:11.200 --> 00:27:14.320 car, but Tesla just rolled out the first 00:27:14.320 --> 00:27:17.380 real semi truck off a real production line. 00:27:17.995 --> 00:27:20.154 And this is something they've been talking about, 00:27:20.154 --> 00:27:22.895 I guess, announced way back since twenty seventeen. 00:27:26.154 --> 00:27:29.195 And only now, nine years later, we're starting 00:27:29.195 --> 00:27:32.235 to see some real delivery. And before I 00:27:32.235 --> 00:27:34.700 really get into the story, you know, I 00:27:34.700 --> 00:27:36.780 wanted to start this one with something I 00:27:36.780 --> 00:27:39.760 saw recently. And, I was driving to work 00:27:39.820 --> 00:27:45.915 this, past, what, Sunday, and I was there 00:27:45.915 --> 00:27:48.315 were traffic was horrible. Two of the three 00:27:48.315 --> 00:27:51.355 lanes were closed. We were hardly moving. You 00:27:51.355 --> 00:27:53.695 know, a ten minute trip had turned into 00:27:53.995 --> 00:27:59.840 an hour. So I'd say nobody sitting there 00:27:59.840 --> 00:28:03.440 was happy, but I'm sure that everyone was 00:28:03.440 --> 00:28:06.000 happy to not be the guy that was 00:28:06.000 --> 00:28:08.799 driving the semi truck because the semi truck 00:28:08.799 --> 00:28:11.895 that had been stopped, the entire cabin had 00:28:11.895 --> 00:28:17.575 been removed from the vehicle. They had the 00:28:17.575 --> 00:28:21.095 entire cabin had been removed, and just the 00:28:21.095 --> 00:28:23.655 front engine was sitting there on wheels. And 00:28:23.655 --> 00:28:26.020 I I was I was baffled. I'd never 00:28:26.020 --> 00:28:27.940 seen something like that. I'm not sure if 00:28:27.940 --> 00:28:30.260 this is common or, you know, those just 00:28:30.260 --> 00:28:34.420 come off, but it seemed like a pretty 00:28:34.420 --> 00:28:36.820 serious situation when they were taking that off 00:28:36.820 --> 00:28:38.580 in the middle of the freeway and trying 00:28:38.580 --> 00:28:43.235 to tow the tow the semi away. So 00:28:44.495 --> 00:28:51.054 that that was crazy. But I think kind 00:28:51.054 --> 00:28:52.654 of what I'm getting at with this is, 00:28:52.654 --> 00:28:55.919 you know, these are massive machines. They they're 00:28:55.919 --> 00:29:00.320 huge. And when they fail, they fail in 00:29:00.320 --> 00:29:04.559 ways that are completely different or there are 00:29:04.559 --> 00:29:07.760 completely different failure modes than what you see 00:29:07.760 --> 00:29:13.115 with regular cars. And to kinda see Tesla 00:29:13.115 --> 00:29:15.675 rolling its first semi off of its, you 00:29:15.675 --> 00:29:21.515 know, factory, you you gotta wonder how this 00:29:21.515 --> 00:29:24.730 is gonna work. You know? We Tesla already 00:29:24.730 --> 00:29:27.130 has a reputation for being, you know, self 00:29:27.130 --> 00:29:31.210 driving vehicle, one of the, you know, the 00:29:31.210 --> 00:29:33.770 better options in terms of electrical cars in 00:29:33.770 --> 00:29:39.684 the United States. So just interesting to kinda 00:29:39.684 --> 00:29:41.924 see their market share continue to grow. You 00:29:41.924 --> 00:29:45.225 know? SpaceX is also up there in terms 00:29:47.125 --> 00:29:50.184 of the amount of, you know, NASA contracts 00:29:50.245 --> 00:29:52.410 that they get to pick up. I believe 00:29:52.410 --> 00:29:54.650 that they do still get the majority of 00:29:54.650 --> 00:29:59.290 it right now. But I think the factory 00:29:59.290 --> 00:30:02.730 is targeting to hit about fifty thousand trucks 00:30:02.730 --> 00:30:05.870 a year, and they're saying the long range 00:30:06.405 --> 00:30:09.785 version, starting at two hundred ninety thousand dollars, 00:30:10.165 --> 00:30:14.405 three twenty five mile three hundred twenty five 00:30:14.405 --> 00:30:19.125 miles? Okay. No. No. No. No. So the 00:30:19.125 --> 00:30:21.760 long range is five hundred miles, and then 00:30:21.760 --> 00:30:24.000 the standard range is three hundred twenty five 00:30:24.000 --> 00:30:26.640 miles at two hundred sixty thousand. So that 00:30:26.640 --> 00:30:31.460 extra thirty k gets you that extra, what, 00:30:31.600 --> 00:30:36.395 one seventy five miles. And not only that, 00:30:36.395 --> 00:30:38.235 the Long Range gets the one point two 00:30:38.235 --> 00:30:42.395 megawatt mega charger that restores sixty percent of 00:30:42.395 --> 00:30:47.055 range in about thirty minutes. The standard range 00:30:47.460 --> 00:30:51.539 version is MCS compatible, but doesn't hit that 00:30:51.539 --> 00:30:54.260 same peak of sixty percent range in thirty 00:30:54.260 --> 00:30:59.620 minutes. Yeah. So I think fifty thousand is 00:30:59.620 --> 00:31:02.020 a pretty big number of trucks to attempt 00:31:02.020 --> 00:31:06.615 to deliver. Reportedly, analysts think five to fifteen 00:31:06.674 --> 00:31:09.654 thousand delivered trucks is a lot more realistic 00:31:09.794 --> 00:31:13.235 in twenty twenty six. And I don't know. 00:31:13.235 --> 00:31:14.914 I'm just I'm not really sure how to 00:31:14.914 --> 00:31:18.570 respond to this type of release, so to 00:31:18.570 --> 00:31:22.410 speak. I think electric semi truck is cool, 00:31:22.410 --> 00:31:26.430 I guess, but I'm curious what the actual 00:31:26.650 --> 00:31:30.545 benefits over a classic semi is. From my 00:31:30.545 --> 00:31:34.385 understanding, you know, the electric car argument is 00:31:34.385 --> 00:31:38.485 not as straightforward or simple as electric vehicles 00:31:38.545 --> 00:31:42.245 use less fuel. You know, the electricity itself 00:31:42.705 --> 00:31:46.190 is often derived from coal or natural gases, 00:31:46.570 --> 00:31:49.370 and the batteries contain rare earth minerals that 00:31:49.370 --> 00:31:51.290 get mined in places with, you know, some 00:31:51.290 --> 00:31:56.115 serious ethical concerns behind the workers. So I 00:31:56.115 --> 00:31:57.875 don't know if this is actually a net 00:31:57.875 --> 00:32:01.315 win for the planet, but good good for 00:32:01.315 --> 00:32:03.715 Tesla, I guess, in terms of, you know, 00:32:03.715 --> 00:32:09.410 another new product released. I I do find 00:32:09.410 --> 00:32:13.730 it shocking that their two different models are 00:32:13.730 --> 00:32:17.410 so significantly different for only thirty thousand dollar 00:32:17.410 --> 00:32:19.250 difference. I don't I don't see why you 00:32:19.250 --> 00:32:23.715 would buy the standard range, especially being that 00:32:23.715 --> 00:32:26.275 it doesn't come with the megawatt megacharger, one 00:32:26.275 --> 00:32:29.875 point two megawatt megacharger. You know? That that 00:32:29.875 --> 00:32:33.635 doesn't seem to make sense, especially if, you 00:32:33.635 --> 00:32:37.155 know, it's only thirty k. So I I 00:32:37.155 --> 00:32:42.059 don't know. But definitely pretty cool, release. So, 00:32:42.059 --> 00:32:44.399 again, they kinda unveiled it in twenty seventeen. 00:32:44.860 --> 00:32:47.980 They had promised, the first semi in twenty 00:32:47.980 --> 00:32:51.100 nineteen, then twenty twenty, then twenty one, then 00:32:51.100 --> 00:32:55.425 twenty two. And finally, I think PepsiCo got 00:32:55.425 --> 00:32:58.065 some in late twenty twenty two, but now 00:32:58.065 --> 00:33:01.905 this is the real production line, today, nine 00:33:01.905 --> 00:33:05.549 years after the reveal. So we'll see how 00:33:05.549 --> 00:33:07.150 it goes. I wonder if we'll start seeing 00:33:07.150 --> 00:33:12.669 some Tesla Tesla semi trucks around. That's, probably 00:33:12.669 --> 00:33:16.029 a great win for them. Now this next 00:33:16.029 --> 00:33:18.835 one, this one's cool. It's from the Chinese 00:33:18.835 --> 00:33:22.355 EV market, and, you know, this also kinda 00:33:22.355 --> 00:33:24.434 goes back to my, I don't know where 00:33:24.434 --> 00:33:26.035 this is going or I don't know if 00:33:26.035 --> 00:33:28.775 this, you know, is actually beneficial or provides 00:33:28.835 --> 00:33:32.295 any real, you know, advancement to humankind, but 00:33:32.480 --> 00:33:38.180 it's cool. So just to preface, we have, 00:33:38.240 --> 00:33:41.780 I guess, headlight laws here that really regulate, 00:33:43.120 --> 00:33:45.200 you know, how far or how bright our 00:33:45.200 --> 00:33:50.935 headlights can be, and our regulations haven't really 00:33:50.935 --> 00:33:52.615 caught up to work with some of the 00:33:52.615 --> 00:33:57.755 newer technology. But Huawei's x pixel headlight system, 00:33:58.055 --> 00:34:01.350 which was deployed across a few different cars, 00:34:01.350 --> 00:34:04.230 can project full color movies up to about 00:34:04.230 --> 00:34:06.470 a hundred inch image in front of a 00:34:06.470 --> 00:34:10.230 car. And this is on top of all 00:34:10.230 --> 00:34:13.905 the safety stuff. So adaptive driving beam, lane 00:34:13.905 --> 00:34:19.825 changing path projection, pedestrian crossing, direction signaling. Wow. 00:34:19.825 --> 00:34:23.825 What? This so Xpixels been around about three 00:34:23.825 --> 00:34:25.985 years, and the new step is the full 00:34:25.985 --> 00:34:29.600 color version that just debuted on the Aito 00:34:29.660 --> 00:34:32.860 m nine. I think it's just cool. You 00:34:32.860 --> 00:34:36.960 know, we we use our cars for practically 00:34:37.020 --> 00:34:39.020 everything. It's our second home for those of 00:34:39.020 --> 00:34:42.444 us that have one, and I think it's 00:34:42.444 --> 00:34:44.444 cool to kinda see your car becoming something 00:34:44.444 --> 00:34:46.525 like a Swiss army knife of, you know, 00:34:46.525 --> 00:34:49.885 personal devices that everyone just loves. You can 00:34:49.885 --> 00:34:51.565 sleep in it, travel in it, eat in 00:34:51.565 --> 00:34:53.645 it, listen to music, and now you can 00:34:53.645 --> 00:34:55.325 park in front of a wall and watch 00:34:55.325 --> 00:34:58.180 a movie projector from your headlights. It's just 00:34:58.180 --> 00:35:02.920 awesome. You know? And the underlying pixel addressable 00:35:03.060 --> 00:35:05.860 headlight platform is the actual safety upgrade hiding 00:35:05.860 --> 00:35:09.380 inside the gimmick. The same hardware that projects 00:35:09.380 --> 00:35:12.820 movies is the same hardware that adaptively dims 00:35:12.820 --> 00:35:16.365 for oncoming traffic and lights a pedestrian safe 00:35:16.365 --> 00:35:21.645 path across the street. That's just cool. So 00:35:21.645 --> 00:35:26.625 in the US, we've only approved adaptive driving 00:35:26.685 --> 00:35:32.110 beam in February twenty twenty two. So okay. 00:35:32.110 --> 00:35:35.550 But the pixel addressable projection systems aren't covered 00:35:35.550 --> 00:35:38.830 by our current regulations. So even if they 00:35:38.830 --> 00:35:42.190 Aydoc sold cars in the US, the headlights 00:35:42.190 --> 00:35:44.110 would have to be downgraded to be street 00:35:44.110 --> 00:35:50.895 legal. So, if you do plan on getting 00:35:50.895 --> 00:35:53.295 this car, don't don't import it to the 00:35:53.295 --> 00:35:58.050 US if you even can. But, you know, 00:35:58.050 --> 00:36:00.950 I think we've seen a lot of groundbreaking 00:36:01.090 --> 00:36:05.250 technology coming out of China this year, and 00:36:05.250 --> 00:36:07.890 they're definitely setting a strong example of, you 00:36:07.890 --> 00:36:12.055 know, forward movement for tech stacks. I think 00:36:12.055 --> 00:36:17.335 that's cool. This is a really interesting story 00:36:17.335 --> 00:36:19.495 that, you know, I thought a lot of 00:36:19.495 --> 00:36:26.359 people probably wouldn't actually consciously notice. So SpaceX 00:36:26.500 --> 00:36:29.880 just launched GPS three s v dash ten, 00:36:30.099 --> 00:36:33.220 the tenth and final GPS three satellite on 00:36:33.220 --> 00:36:36.420 a Falcon nine from Cape Canaveral on April 00:36:36.420 --> 00:36:40.835 twenty first. So GPS three gives you signals 00:36:40.975 --> 00:36:44.895 three times more accurate and eight times more 00:36:44.895 --> 00:36:49.375 resistant to jamming than the previous constellation. The 00:36:49.375 --> 00:36:52.035 l one c signal that interoperates with Galileo, 00:36:52.660 --> 00:36:57.400 BeiDou, IRNSS, and QZSS, and m code military 00:36:57.540 --> 00:37:02.500 encryption. So this just got silently delivered on 00:37:02.500 --> 00:37:05.540 the twenty first. It's probably already working on 00:37:05.540 --> 00:37:08.175 your phone. So if you noticed that your 00:37:08.175 --> 00:37:11.795 GPS has had a significant improvement in performance, 00:37:12.095 --> 00:37:17.955 that's probably why. But launches for the next 00:37:18.335 --> 00:37:22.720 version, GPS I I I f. So, now 00:37:22.720 --> 00:37:25.380 I'm wondering if it's GPS I I I, 00:37:25.760 --> 00:37:28.980 but that looks like three. Okay. Well, GPS 00:37:29.039 --> 00:37:32.079 I I I f, is slated to start 00:37:32.079 --> 00:37:34.805 launching in twenty twenty seven, and up to 00:37:34.805 --> 00:37:37.204 twenty two satellites are going up through about 00:37:37.204 --> 00:37:41.924 twenty thirty seven. And the IIIF system adds 00:37:41.924 --> 00:37:46.805 laser inter satellite links and optical reflectors for 00:37:46.805 --> 00:37:52.820 centimeter level satellite tracking. Wow. And like I 00:37:52.820 --> 00:37:54.820 said, this is a pretty quiet upgrade that 00:37:54.820 --> 00:37:56.340 not a lot of people are gonna directly 00:37:56.340 --> 00:37:58.420 notice, and I think that's why it's so 00:37:58.420 --> 00:38:02.500 amazing. I love these, you know, systems that 00:38:02.500 --> 00:38:06.345 drive our society and, you know, drive everyday 00:38:06.485 --> 00:38:09.525 activities without people actually realizing it. You know? 00:38:09.525 --> 00:38:11.125 A lot of us have come to really 00:38:11.125 --> 00:38:16.025 mod or rely on modern GPS systems, and 00:38:18.300 --> 00:38:19.980 I I don't know what kind of chaos 00:38:19.980 --> 00:38:23.260 would ensue if GPS went even down for 00:38:23.260 --> 00:38:25.580 a day. You know? I think people will 00:38:25.580 --> 00:38:29.020 forget how to drive. So I think that's 00:38:29.020 --> 00:38:33.465 really cool. Definitely look more into some of 00:38:33.465 --> 00:38:38.505 the new modernizations that they've provided with, this 00:38:38.505 --> 00:38:43.640 new GPS three satellite. So but, yeah, what 00:38:43.640 --> 00:38:44.920 do you guys think? I I feel like 00:38:44.920 --> 00:38:47.580 it's genuinely cool to kinda see these silent 00:38:47.640 --> 00:38:51.000 projects that influence, you know, basically, like, global 00:38:51.000 --> 00:38:55.525 economics and world infrastructure and being, like, one 00:38:55.525 --> 00:38:59.125 of the mostly or most widely dependent on 00:38:59.125 --> 00:39:02.485 technologies in the world. We don't really talk 00:39:02.485 --> 00:39:04.405 about GPS that much. It's not it's not 00:39:04.405 --> 00:39:06.165 really, you know, put on a pedestal. It's 00:39:06.165 --> 00:39:10.860 just a given. So I think that's cool. 00:39:11.720 --> 00:39:15.260 We'll start to see more and more data 00:39:15.400 --> 00:39:17.480 coming back, I feel like, from, you know, 00:39:17.480 --> 00:39:22.040 GPS tracking. And to have centimeter level, you 00:39:22.040 --> 00:39:27.335 know, precision is, in my opinion, insane. You 00:39:27.335 --> 00:39:30.075 know? We'll definitely start to see more of 00:39:31.335 --> 00:39:34.935 interesting statistics or, like, you know, average steps 00:39:34.935 --> 00:39:37.735 taken in a lifetime, average distance traveled in 00:39:37.735 --> 00:39:41.140 a lifetime, average time spent in specific locations. 00:39:41.279 --> 00:39:44.480 You know? And you gotta wonder, that data 00:39:44.480 --> 00:39:47.119 is just not sitting there. It's definitely being 00:39:47.119 --> 00:39:53.535 used elsewhere for, you know, who knows what, 00:39:53.535 --> 00:39:58.655 data training, you know, new models. So the 00:39:58.655 --> 00:40:00.515 good and the bad, I guess, with that. 00:40:02.655 --> 00:40:07.230 Here's a really cool research, I guess, security 00:40:07.230 --> 00:40:10.590 slash research story. A researcher just won a 00:40:10.590 --> 00:40:14.510 Bitcoin for a quantum attack on crypto. So 00:40:14.510 --> 00:40:22.305 independent Italian researcher, Giancarlo Lelyuan, Project Eleven's one 00:40:22.305 --> 00:40:25.525 Bitcoin q day prize on April twenty fourth 00:40:25.825 --> 00:40:29.665 by deriving a fifteen bit elliptic curve private 00:40:29.665 --> 00:40:33.420 key from its public key. So they apparently 00:40:33.560 --> 00:40:35.640 achieved this by using a variant of Shor's 00:40:35.640 --> 00:40:40.220 algorithm on publicly accessible rented cloud quantum hardware. 00:40:40.600 --> 00:40:47.725 Wow. Wow. So the search space he covered 00:40:47.725 --> 00:40:50.365 was thirty two thousand seven hundred sixty seven 00:40:50.365 --> 00:40:55.025 possibilities. That's two to the fifteenth minus one. 00:40:55.805 --> 00:40:58.285 And the previous record was six bits set 00:40:58.285 --> 00:41:01.580 in September twenty five on an IBM one 00:41:01.580 --> 00:41:04.220 thirty three qubit machine. So this extends the 00:41:04.220 --> 00:41:07.900 record by a factor of five hundred and 00:41:07.900 --> 00:41:13.580 twelve times. Now don't get scared or start 00:41:13.580 --> 00:41:17.795 selling your Bitcoin because, that is fifteen bit 00:41:17.795 --> 00:41:20.994 elliptic curve, and Bitcoin uses two hundred fifty 00:41:20.994 --> 00:41:25.734 six elliptic curve cryptography. So they're not anywhere 00:41:25.795 --> 00:41:29.895 near breaking real Bitcoin yet. But I think 00:41:30.140 --> 00:41:32.460 it's kind of interesting to see the gap 00:41:32.460 --> 00:41:35.260 that project eleven is kinda, like, characterizing or 00:41:35.260 --> 00:41:40.620 highlighting here as a fundamental well, they're saying 00:41:40.620 --> 00:41:44.205 it's increasingly an engineering problem rather than a 00:41:44.445 --> 00:41:48.045 fundamental physics problem, which I'm not completely sure 00:41:48.045 --> 00:41:52.845 what they mean by that. But, apparently, this 00:41:52.845 --> 00:41:54.785 type of framing is what makes the milestone 00:41:54.925 --> 00:41:58.625 notable. It's that it an independent researcher was 00:41:59.150 --> 00:42:02.770 working on basically cloud accessible hardware, you know, 00:42:02.830 --> 00:42:05.630 no national lab, no private chips, nothing. This 00:42:05.630 --> 00:42:09.250 is just, you know, publicly available tools, and 00:42:11.265 --> 00:42:13.984 researchers have definitely had some pushback on this. 00:42:13.984 --> 00:42:17.505 You know, their criticism says that, you know, 00:42:17.505 --> 00:42:20.385 a regular laptop can brute force a fifteen 00:42:20.385 --> 00:42:24.640 bit bits search space in milliseconds. When you 00:42:24.640 --> 00:42:26.480 run a quantum machine on a problem that 00:42:26.480 --> 00:42:30.020 small, you can't actually tell whether Shor's algorithm 00:42:30.240 --> 00:42:32.800 did anything quantum or whether the result is 00:42:32.800 --> 00:42:35.120 what you'd get from random noise that happened 00:42:35.120 --> 00:42:41.165 to land on the right answer. So this 00:42:41.165 --> 00:42:43.805 pushback is basically saying that without showing measurable 00:42:43.805 --> 00:42:47.165 quantum advantage, speed, or behavior that classical hardware 00:42:47.165 --> 00:42:49.965 couldn't replicate, you don't really have a quantum 00:42:49.965 --> 00:42:52.365 attack. You have a quantum machine that returned 00:42:52.365 --> 00:42:55.890 the right number, and there's a huge difference 00:42:56.030 --> 00:43:00.190 between those two things conceptually. You know? And 00:43:00.190 --> 00:43:02.610 then project eleven kinda came back with the 00:43:02.750 --> 00:43:04.990 defense that the point isn't to break real 00:43:04.990 --> 00:43:07.790 keys at this scale. It's to demonstrate that 00:43:08.225 --> 00:43:10.705 demonstrate Shor's running end to end on real 00:43:10.705 --> 00:43:14.785 quantum hardware, you know, as a stepping stone 00:43:14.785 --> 00:43:18.145 for the ecosystem itself. So they're just really 00:43:18.145 --> 00:43:20.545 treating this as something cool or interesting that 00:43:20.545 --> 00:43:25.670 happened, And I think that they're less focused 00:43:25.670 --> 00:43:29.450 on the idea that, you know, that a 00:43:29.670 --> 00:43:31.510 laptop could brute force it or that it 00:43:31.510 --> 00:43:34.950 may have accidentally or randomly gotten the right 00:43:34.950 --> 00:43:41.175 answer. They're more interested in, you know, this 00:43:41.175 --> 00:43:43.415 as a case study in itself. So I 00:43:43.415 --> 00:43:45.355 think that's a really good stance to have, 00:43:45.575 --> 00:43:47.895 and I'm sure we're gonna see more and 00:43:47.895 --> 00:43:51.880 more, quantum research and availability come around in 00:43:51.880 --> 00:43:54.840 the next five to ten years. Definitely one 00:43:54.840 --> 00:43:57.880 of the fields that I was not expecting 00:43:57.880 --> 00:43:59.900 to grow as fast as it did, but 00:44:00.920 --> 00:44:03.645 I think it's cool. I've been seeing those, 00:44:03.965 --> 00:44:08.685 quantum encryption warnings on, you know, my Git 00:44:08.685 --> 00:44:14.205 keys for a while now, and I definitely 00:44:14.205 --> 00:44:16.299 am looking forward to the day where those 00:44:16.299 --> 00:44:18.299 are no longer a thing, and I just 00:44:18.299 --> 00:44:21.420 have quantum and encrypted keys. I guess that's 00:44:21.420 --> 00:44:25.740 what they would be. Right? But let's move 00:44:25.740 --> 00:44:29.020 on. So kind of a interesting one from 00:44:29.020 --> 00:44:33.065 Google blog, Gemini now generates real files. And 00:44:33.065 --> 00:44:36.205 I know this is kinda sounds lame, but, 00:44:37.145 --> 00:44:39.385 you know, I think it's kinda cool. It 00:44:39.465 --> 00:44:42.744 Gemini is able to generate downloadable files directly 00:44:42.744 --> 00:44:45.085 from a chat prompt. So this includes PDFs, 00:44:45.420 --> 00:44:49.599 Word docs, Excel spreadsheets, and Google Workspace files. 00:44:52.540 --> 00:44:55.980 You know? And I'm not sure how exciting 00:44:55.980 --> 00:44:58.785 this is. Maybe it's the wide spread availability 00:44:59.805 --> 00:45:05.505 of different, file types. They've gone and covered. 00:45:05.725 --> 00:45:10.525 So docs, sheets, slides, CSV, LaTeX, plain text, 00:45:10.525 --> 00:45:14.609 RTF, and markdown. So you can also either 00:45:14.609 --> 00:45:17.250 download to a device or export it straight 00:45:17.250 --> 00:45:20.930 to Google Drive. And this is just globally 00:45:20.930 --> 00:45:23.010 available now. And to yeah. Just a nice 00:45:23.010 --> 00:45:26.130 UX edition to kinda see coming into, another 00:45:26.130 --> 00:45:30.615 LLM tool. I feel like Claude's had all 00:45:30.615 --> 00:45:34.775 this for a while. So I really don't 00:45:34.775 --> 00:45:37.195 know how to be excited about this one, 00:45:37.655 --> 00:45:43.380 but but definitely nice to see, you know, 00:45:43.380 --> 00:45:46.500 forward movement on tooling like this. I think 00:45:46.500 --> 00:45:50.020 it's cool. Definitely good use of their time 00:45:50.020 --> 00:45:53.255 to make that available, but I definitely think 00:45:53.474 --> 00:45:56.835 last week's story on Google's design MD spec 00:45:56.835 --> 00:46:01.494 was definitely a lot more interesting and industry 00:46:01.635 --> 00:46:05.869 driving than generating real files from Gemini, but 00:46:06.009 --> 00:46:08.250 definitely a nice convenience hack for those of 00:46:08.250 --> 00:46:12.329 you that are using Gemini and like that 00:46:12.329 --> 00:46:16.430 like that work stack. Google did another thing. 00:46:17.210 --> 00:46:19.425 They released I don't know if they released 00:46:19.425 --> 00:46:21.425 it, but they have this tool called Google 00:46:21.425 --> 00:46:24.625 Illuminate, which is an experimental lab tool that 00:46:24.625 --> 00:46:28.325 turns academic papers and technical documents into roughly 00:46:28.465 --> 00:46:32.145 five minute two voice podcast style audio with 00:46:32.145 --> 00:46:36.280 interactive click to jump transcripts. So generation time 00:46:36.280 --> 00:46:39.819 is about thirty seconds. There's a twenty generation 00:46:40.119 --> 00:46:42.700 per day cap and thirty day library retention, 00:46:43.560 --> 00:46:47.400 four style presets, casual, formal, guided, and free 00:46:47.400 --> 00:46:53.315 form. So I guess now it accepts arbitrary 00:46:53.455 --> 00:46:57.155 URLs and topic searches and still requires Gmail 00:46:57.215 --> 00:46:59.375 login and wait list access. So I guess 00:46:59.375 --> 00:47:01.855 it's not publicly available yet, but this is 00:47:01.855 --> 00:47:04.895 very similar to the notebook l m tool 00:47:04.895 --> 00:47:10.810 that generates a podcast. And, you know, I 00:47:10.810 --> 00:47:16.010 have opinions on this, especially in terms of, 00:47:16.010 --> 00:47:20.270 you know, the explosion of agentically generated podcast, 00:47:21.305 --> 00:47:25.704 now in the ecosystem. Definitely feels much more 00:47:25.704 --> 00:47:29.305 spammy. Definitely feels a lot less human. I'm 00:47:29.305 --> 00:47:32.265 not sure if I love it, but it 00:47:32.265 --> 00:47:34.285 is kinda cool to be able to get 00:47:36.660 --> 00:47:39.619 very dense technical information, you know, parsed down 00:47:39.619 --> 00:47:45.780 into simple language that's easy to understand as 00:47:45.780 --> 00:47:47.780 a primer. I think I think as a 00:47:47.780 --> 00:47:51.194 primer, this we'll see this kind of tool 00:47:51.194 --> 00:47:55.994 shining, but I don't I, you know, I 00:47:55.994 --> 00:47:58.795 I don't see the point of using something 00:47:58.795 --> 00:48:02.095 like this to replace actually reading the research 00:48:02.154 --> 00:48:06.839 paper. You know? It it defeats the purpose 00:48:06.839 --> 00:48:11.960 of the depth that the paper provides. You 00:48:11.960 --> 00:48:13.880 know? And so, you know, there's two sides 00:48:13.880 --> 00:48:16.599 of this. Right? So my boss at my 00:48:16.599 --> 00:48:19.535 weekend job has, you know, recently been generating 00:48:19.535 --> 00:48:25.695 training videos, basically, like, by making instructions, you 00:48:25.695 --> 00:48:28.095 know, an instruction list, the notebook l m 00:48:28.095 --> 00:48:29.855 will turn it into a training video and 00:48:29.855 --> 00:48:31.535 and can even make it themed. So they 00:48:31.535 --> 00:48:33.855 they do mostly anime themed, which I find 00:48:33.855 --> 00:48:38.450 funny, but that's a whole topic. But the 00:48:38.450 --> 00:48:40.530 point of that is, you know, it's it's 00:48:40.530 --> 00:48:44.710 very there's not much technicality involved in these 00:48:45.250 --> 00:48:47.275 tasks that are being trained on. You know, 00:48:47.275 --> 00:48:50.315 they're very general information with a very interactive 00:48:50.315 --> 00:48:52.715 u a UI that's simple. It's not something 00:48:52.715 --> 00:48:56.315 like complex math or clustering logic. It's it's 00:48:56.395 --> 00:49:00.869 you know, those types of documentation real need 00:49:00.869 --> 00:49:03.910 real reads and not a summary by AI 00:49:03.910 --> 00:49:08.470 podcasters. So I think the right framing for 00:49:08.470 --> 00:49:11.270 me personally is, you know, something more of 00:49:11.270 --> 00:49:14.305 an index rather than a substitute. You know? 00:49:14.305 --> 00:49:15.665 You can even use it to kinda figure 00:49:15.665 --> 00:49:18.145 out which papers you actually wanna sit down 00:49:18.145 --> 00:49:20.465 with or use it to refresh on papers 00:49:20.465 --> 00:49:22.885 you might have read in the past. But 00:49:24.385 --> 00:49:26.465 the I don't know. I I just feel 00:49:26.465 --> 00:49:28.645 like you lose out on too much by 00:49:28.890 --> 00:49:32.170 turning, you know, a dense paper into a 00:49:32.170 --> 00:49:36.809 five minute summary. And I'm sure that the 00:49:36.809 --> 00:49:39.450 twenty generations a day is actually way more 00:49:39.450 --> 00:49:41.450 than enough. You know, that's a really generous 00:49:41.450 --> 00:49:43.055 cap. I don't know if you should be 00:49:43.055 --> 00:49:45.375 trying to process more than, you know, twenty 00:49:45.375 --> 00:49:48.335 research papers a day. Our brains are literally 00:49:48.335 --> 00:49:51.474 not built to consume information like that. So 00:49:53.135 --> 00:49:55.394 who knows? I don't think I would want 00:49:55.860 --> 00:49:58.580 AI podcaster to, you know, represent any sort 00:49:58.580 --> 00:50:00.420 of work that I've done or papers that 00:50:00.420 --> 00:50:05.140 I've done, especially with the amount of drift 00:50:05.140 --> 00:50:08.724 and context overlap. I think that there's a 00:50:08.724 --> 00:50:11.765 high chance of misinterpreting the material that's trying 00:50:11.765 --> 00:50:16.484 to be conveyed. So that's my take on 00:50:16.484 --> 00:50:17.925 it, but I don't know. What do you 00:50:17.925 --> 00:50:20.165 guys think? You guys gonna be out here 00:50:20.165 --> 00:50:24.040 generating podcasts to learn, you know, deep material 00:50:24.260 --> 00:50:28.900 very superficially, let me know. But I don't 00:50:28.900 --> 00:50:31.220 want to kinda go back or go ahead 00:50:31.220 --> 00:50:33.859 and look back on these years as, you 00:50:33.859 --> 00:50:36.840 know, the age of AI summarization where everything 00:50:36.980 --> 00:50:41.295 went from being long form long form available, 00:50:42.235 --> 00:50:46.495 short form consumed. You know? Rather than actually 00:50:46.555 --> 00:50:50.575 going in-depth on topics, we are just arbitrarily 00:50:50.795 --> 00:50:54.910 skimming skimming over the surface. So who knows? 00:50:54.910 --> 00:50:57.890 We'll kinda see how this changes over time. 00:50:58.110 --> 00:51:00.930 It is kinda interesting to see this, competitor, 00:51:01.230 --> 00:51:05.570 notebook l m. Notebook l m is definitely 00:51:06.110 --> 00:51:12.635 kind of, you know, geared towards research and, 00:51:12.635 --> 00:51:18.255 you know, mind maps, learning, flashcards. So this 00:51:18.315 --> 00:51:22.175 illuminate tool is just audio and the transcript. 00:51:22.315 --> 00:51:24.640 So, you know, definitely a lot more lightweight 00:51:24.700 --> 00:51:29.820 than NotebookLM and probably a better use case 00:51:29.820 --> 00:51:32.700 for, you know, how do you put it, 00:51:32.700 --> 00:51:36.305 like, fresh context reads. Whereas NotebookLM would, you 00:51:36.305 --> 00:51:38.625 know, keep a host of data that can 00:51:38.625 --> 00:51:41.365 be tied together for, you know, later podcast 00:51:41.425 --> 00:51:45.905 or whatever. But, regardless, definitely a nice cool 00:51:45.905 --> 00:51:48.785 innovation by Google. Actually, both of those were 00:51:48.785 --> 00:51:52.750 cool innovations by Google. So definitely excited to 00:51:52.750 --> 00:51:55.309 kinda see more of the tooling ecosystem growing 00:51:55.309 --> 00:51:57.650 out growing out in the next few years. 00:51:58.109 --> 00:52:01.630 Who knows? But big thanks to GoDaddy for 00:52:01.630 --> 00:52:03.390 over twenty years of keeping this show on 00:52:03.390 --> 00:52:05.775 the air. Support us at geek new central 00:52:05.835 --> 00:52:09.455 dot com forward slash GoDaddy. Every click counts. 00:52:10.155 --> 00:52:12.395 You know, using a promo code is like 00:52:12.395 --> 00:52:15.115 writing a check to the show. So definitely 00:52:15.115 --> 00:52:17.035 spread that around. If anyone you know is 00:52:17.035 --> 00:52:18.955 trying to make a nice cool website, they've 00:52:18.955 --> 00:52:21.180 got very awesome tooling to get you up 00:52:21.180 --> 00:52:24.300 and started as fast as possible. If you 00:52:24.300 --> 00:52:26.540 wanna become a GNC insider, go ahead and 00:52:26.540 --> 00:52:28.380 head over to geek news central dot com 00:52:28.380 --> 00:52:30.140 forward slash insider and sign up for our 00:52:30.140 --> 00:52:32.915 insider program to support the show. If you 00:52:32.915 --> 00:52:35.075 got questions or feedback, feel free to email 00:52:35.075 --> 00:52:37.075 us at geek news at g mail dot 00:52:37.075 --> 00:52:40.515 com. Subscribe via your favorite podcast app and 00:52:40.515 --> 00:52:42.435 sign up for the newsletter at geek news 00:52:42.435 --> 00:52:45.555 central dot com, and you can grab yourself 00:52:45.555 --> 00:52:47.875 a modern podcast app from podcast apps dot 00:52:47.875 --> 00:52:49.420 com. They've got a lot of cool ones. 00:52:49.420 --> 00:52:52.460 I'm currently using I always forget the name. 00:52:52.460 --> 00:52:54.540 It's I think it's Pocket Casts. Yeah. I'm 00:52:54.540 --> 00:52:55.980 really bad about it. I just look at 00:52:55.980 --> 00:52:58.080 the icon and just go, whoop, that one. 00:52:58.860 --> 00:53:01.660 But, yep, Pocket Casts has been serving me 00:53:01.660 --> 00:53:03.635 pretty well. It seems like they get their 00:53:03.635 --> 00:53:06.355 updates pretty fast. Other than them, the one 00:53:06.355 --> 00:53:10.215 that I've noticed that gets, their feed updates 00:53:10.435 --> 00:53:13.175 very fast has been true fans. So definitely 00:53:13.395 --> 00:53:15.715 a couple of podcast apps to check out 00:53:15.715 --> 00:53:17.250 for you guys, a lot of the modern 00:53:17.310 --> 00:53:21.230 podcast two point o features. But let us 00:53:21.230 --> 00:53:22.690 know what you guys thought of the show.