WEBVTT 00:00:02.239 --> 00:00:04.780 Welcome to the AI Chat Podcast. I'm your host, 00:00:04.860 --> 00:00:07.179 Jaden Schaefer. Every day I cover cutting -edge 00:00:07.179 --> 00:00:09.320 AI news and talk with the leaders behind it, 00:00:09.339 --> 00:00:11.300 breaking down what it means for your life and 00:00:11.300 --> 00:00:17.329 business. Anthropic is finally letting Claude 00:00:17.329 --> 00:00:20.129 Cowork go to your mobile device, your phone on 00:00:20.129 --> 00:00:22.010 the Claude app. You've probably noticed you can 00:00:22.010 --> 00:00:24.410 get Claude code over there, but Claude Cowork 00:00:24.410 --> 00:00:26.250 was missing. So they're going to be releasing 00:00:26.250 --> 00:00:29.410 that soon. DeepSeek is planning their own inference 00:00:29.410 --> 00:00:32.670 chips to basically cut NVIDIA and Huawei out 00:00:32.670 --> 00:00:34.090 of the deal for them. They're trying to get around 00:00:34.090 --> 00:00:36.880 them and they rely on them heavily. SK Hynix 00:00:36.880 --> 00:00:41.700 is planning a $28 billion US IPO or listing because 00:00:41.700 --> 00:00:45.719 the demand for their memory has made a 260 % 00:00:45.719 --> 00:00:48.560 stock jump for them. So this company is growing 00:00:48.560 --> 00:00:52.799 like crazy. Solos is launching AirGo A6. These 00:00:52.799 --> 00:00:55.520 are smart glasses. They're only 19 grams and 00:00:55.520 --> 00:00:57.880 essentially they're smart glasses with voice 00:00:57.880 --> 00:01:01.039 AI, but there is no camera in them. Vercel's 00:01:01.039 --> 00:01:04.200 Guillaume Rauch is betting on splitting AI models 00:01:04.200 --> 00:01:06.700 from agents. So he thinks that you're going to 00:01:06.700 --> 00:01:08.959 get more bang for your buck and more efficiency 00:01:08.959 --> 00:01:11.180 if you actually take the AI model away from the 00:01:11.180 --> 00:01:12.980 agent and have these as two separate things. 00:01:13.120 --> 00:01:15.219 In completely unrelated news, I got a text today 00:01:15.219 --> 00:01:17.500 from my city in North Carolina that said that 00:01:17.500 --> 00:01:19.219 we are in a drought and I'm only allowed to water 00:01:19.219 --> 00:01:22.620 my lawn if my house number is even every other 00:01:22.620 --> 00:01:25.200 day of the week. And then it has proceeded to 00:01:25.200 --> 00:01:26.900 rain all day long. I'm not sure if we were supposed 00:01:26.900 --> 00:01:29.019 to be saving water so that AI data centers don't 00:01:29.019 --> 00:01:30.780 run out. but I think we have enough rain now. 00:01:30.819 --> 00:01:32.930 The drought might be officially over. Okay, I'm 00:01:32.930 --> 00:01:34.969 super excited. Maybe the most excited I've been 00:01:34.969 --> 00:01:38.010 in a while. Anthropic has just allowed ClaudeCowork 00:01:38.010 --> 00:01:41.530 to go on your mobile device on the Claude app 00:01:41.530 --> 00:01:43.750 on your phone. Previously, it was only on your 00:01:43.750 --> 00:01:45.469 desktop computer. And it's actually pretty annoying 00:01:45.469 --> 00:01:47.150 because if you're doing something with ClaudeCode, 00:01:47.310 --> 00:01:49.909 you could remote control it from your phone, 00:01:50.030 --> 00:01:51.849 although it had some limitations. One of them 00:01:51.849 --> 00:01:53.969 being I found I had like all the permissions 00:01:53.969 --> 00:01:56.549 were set the same way from my phone. So I still 00:01:56.549 --> 00:01:58.250 had to keep approving things over and over again, 00:01:58.310 --> 00:02:00.530 like the olden days, which is kind of brutal. 00:02:00.870 --> 00:02:03.079 But I'm really excited. excited because now you 00:02:03.079 --> 00:02:05.719 can fire something off on your computer, leave 00:02:05.719 --> 00:02:08.680 and you can keep, you know, get iterating with 00:02:08.680 --> 00:02:11.020 Claude cowork on your phone. And I think it's 00:02:11.020 --> 00:02:12.520 actually going to be able to take control of 00:02:12.520 --> 00:02:14.699 your computer still and do stuff. So this is 00:02:14.699 --> 00:02:17.259 actually insanely useful for so many things. 00:02:17.319 --> 00:02:19.219 At the same time, Claude has also published a 00:02:19.219 --> 00:02:21.919 list of what people are using Claude cowork for. 00:02:22.000 --> 00:02:24.580 So in case you need any ideas, or you're curious, 00:02:24.719 --> 00:02:27.539 there's one graph in particular, they said that 00:02:27.539 --> 00:02:30.780 33 % of people are using Claude cowork for business 00:02:31.120 --> 00:02:34.360 processes and operations. 16 % are using it for 00:02:34.360 --> 00:02:37.979 content creation and copywriting and only 8 .7%. 00:02:37.979 --> 00:02:39.860 So basically like 9 % are using it for software 00:02:39.860 --> 00:02:41.479 development. I actually use it for software development 00:02:41.479 --> 00:02:44.340 a lot. Um, but I kind of toggle between Claude 00:02:44.340 --> 00:02:47.120 cowork and Claude code. There's like pros and 00:02:47.120 --> 00:02:50.060 cons to both of them. I find if I'm doing something 00:02:50.060 --> 00:02:53.159 where I need to be like logging in on the browser 00:02:53.159 --> 00:02:55.099 to stuff and clicking around and grabbing files 00:02:55.099 --> 00:02:57.060 off my computer, I'm going to use Claude cowork. 00:02:57.199 --> 00:02:59.599 Um, if I'm doing something that's more in like, 00:02:59.680 --> 00:03:01.610 you know, just making pushes to my GitHub and 00:03:01.610 --> 00:03:04.210 stuff, I'm going to use Claude code. So anyways, 00:03:04.289 --> 00:03:06.189 there's pros and cons. And I will say the biggest 00:03:06.189 --> 00:03:08.009 pro with Claude code is that you just get way 00:03:08.009 --> 00:03:09.930 more mileage out of it. I burned through all 00:03:09.930 --> 00:03:11.569 my Claude co -work credits. I actually made a 00:03:11.569 --> 00:03:13.750 new Claude max account just because I didn't 00:03:13.750 --> 00:03:15.530 want to run out of all my fable credits, which 00:03:15.530 --> 00:03:19.080 shout out also Claude. Claude said that they're 00:03:19.080 --> 00:03:23.139 extending Fable 5, the use inside of the Claude 00:03:23.139 --> 00:03:24.580 app. If you have like a Claude Mac subscription, 00:03:24.960 --> 00:03:27.120 they're extending it to the 12th. They were supposed 00:03:27.120 --> 00:03:29.099 to end today on the 7th, but they're extending 00:03:29.099 --> 00:03:31.219 it to the 12th. So if you have Claude, you're 00:03:31.219 --> 00:03:33.319 going to get way more Fable 5. Now, the problem 00:03:33.319 --> 00:03:35.819 is if you're anything like me, you burnt all 00:03:35.819 --> 00:03:38.139 of your Fable 5 credits already. I made a new 00:03:38.139 --> 00:03:40.680 account yesterday and I have officially burnt 00:03:40.680 --> 00:03:43.439 through all of my weekly credits already. So 00:03:43.439 --> 00:03:46.180 that's definitely a problem. I'm trying to work 00:03:46.180 --> 00:03:48.990 on that and I have to wait. till tomorrow, Wednesday, 00:03:49.169 --> 00:03:51.330 before it reset. The next things that people 00:03:51.330 --> 00:03:53.909 are using Claude for is DevOps and infrastructure, 00:03:54.189 --> 00:03:58.490 7%, research and intelligence, 6 .4%, data analysis 00:03:58.490 --> 00:04:01.889 and business intelligence, 5 .8%, document processing 00:04:01.889 --> 00:04:04.289 and extraction, 4%, sales and revenue operations, 00:04:04.590 --> 00:04:06.830 4%, and then everything else is about 14%. So 00:04:06.830 --> 00:04:08.729 the biggest thing is business processes and operations, 00:04:08.969 --> 00:04:12.409 content creation, and copywriting. That is basically 00:04:12.409 --> 00:04:15.930 50 % of what everybody is using it for. I know 00:04:15.930 --> 00:04:17.209 you'll be able to do all of that from your phone, 00:04:17.269 --> 00:04:20.170 which is a huge unlock. It's also on 24 -7. So 00:04:20.170 --> 00:04:21.750 you can actually give it a task on your phone 00:04:21.750 --> 00:04:23.949 and then go to bed and it's done in the morning, 00:04:24.029 --> 00:04:25.529 which is really nice. I think I've had issues 00:04:25.529 --> 00:04:27.930 in the past when working with a similar type 00:04:27.930 --> 00:04:29.829 of software from my phone where it like if you 00:04:29.829 --> 00:04:31.810 turn your phone screen off, then it stops running, 00:04:31.949 --> 00:04:34.009 which is bad. So anyways, it's always on. It's 00:04:34.009 --> 00:04:36.329 from your phone. This will be fantastic. It is 00:04:36.329 --> 00:04:39.509 launching today. It is in beta and it is for 00:04:39.509 --> 00:04:41.769 Max Plan subscribers. If you're paying $100 a 00:04:41.769 --> 00:04:44.269 month, it's also going to roll out to the pro 00:04:44.269 --> 00:04:47.930 users at $20 a month later on. CoWorks demo that 00:04:47.930 --> 00:04:49.819 they showed with this. This basically showed 00:04:49.819 --> 00:04:52.319 it pulling data from email, Slack, and a bunch 00:04:52.319 --> 00:04:54.160 of different meeting transcripts, and it was 00:04:54.160 --> 00:04:56.500 writing a document, drafting emails, and it was 00:04:56.500 --> 00:04:58.540 all doing that from your phone, which is really 00:04:58.540 --> 00:05:01.420 cool, but it was using your computer to, like 00:05:01.420 --> 00:05:03.300 your desktop, to do all those things, which is 00:05:03.300 --> 00:05:06.199 pretty sweet. DeepSeek, the popular Chinese LLM, 00:05:06.259 --> 00:05:08.959 is going to be designing their own chips to run 00:05:08.959 --> 00:05:11.160 inference. This is something that essentially 00:05:11.160 --> 00:05:13.860 they're doing to get away from relying on NVIDIA 00:05:13.860 --> 00:05:16.379 and, of course, their big competitor, Huawei, 00:05:16.540 --> 00:05:18.579 over in China, which currently dominates. It's 00:05:18.579 --> 00:05:21.379 about half of China's data center chip market. 00:05:21.680 --> 00:05:24.000 What I will say is that inference chips are cheaper 00:05:24.000 --> 00:05:26.779 and faster to iterate on than training chips. 00:05:26.920 --> 00:05:29.480 So that's kind of DeepSeq's something that's 00:05:29.480 --> 00:05:31.740 going to help their own profit margin on every 00:05:31.740 --> 00:05:33.680 response that their model is actually generating. 00:05:34.060 --> 00:05:36.680 Alibaba and Baidu are both kind of pushing for 00:05:36.680 --> 00:05:39.120 similar custom chips. OpeningEye and Broadcom 00:05:39.120 --> 00:05:41.240 have released their own inference chip, which 00:05:41.240 --> 00:05:43.920 is called Jalapeno. As far as the timeline goes, 00:05:44.180 --> 00:05:46.439 custom accelerators typically take about three 00:05:46.439 --> 00:05:49.620 to five years from design to production shipment. 00:05:49.819 --> 00:05:51.660 And DeepSeq, they haven't. said you know what 00:05:51.660 --> 00:05:53.560 any of their like foundry partners haven't said 00:05:53.560 --> 00:05:54.899 anything about their timeline they haven't said 00:05:54.899 --> 00:05:56.899 anything about the performance targets so while 00:05:56.899 --> 00:05:58.459 they are starting to develop this i don't think 00:05:58.459 --> 00:06:01.300 we're going to see it anytime very soon SK Hynix 00:06:01.300 --> 00:06:03.279 is going to be selling shares on the US stock 00:06:03.279 --> 00:06:06.019 market this week. They're raising about $28 billion 00:06:06.019 --> 00:06:08.699 and they're letting anyone in the United States 00:06:08.699 --> 00:06:12.060 get access to the world's second largest memory 00:06:12.060 --> 00:06:14.240 chip maker. This is the first time this is happening 00:06:14.240 --> 00:06:16.439 and if you remember my last episode, Samsung, 00:06:16.699 --> 00:06:19.240 SK Hynix, and one company out of Idaho are basically 00:06:19.240 --> 00:06:22.420 the only three people that can mass produce These 00:06:22.420 --> 00:06:24.720 custom memory chips for AI training at scale. 00:06:24.860 --> 00:06:27.680 This listing in particular is going to, I mean, 00:06:27.680 --> 00:06:28.939 they're really just trying to capitalize on the 00:06:28.939 --> 00:06:32.279 fact that in the last couple months, their stock 00:06:32.279 --> 00:06:35.680 has surged 260%. All of this is because of AI 00:06:35.680 --> 00:06:38.519 data centers, which desperately need high -speed 00:06:38.519 --> 00:06:41.100 memory chips to train and run AI models. SK Hynix 00:06:41.100 --> 00:06:44.339 is going to sell 17 .8 million shares. And a 00:06:44.339 --> 00:06:46.019 lot of people are calling this entire situation 00:06:46.019 --> 00:06:49.000 Ramageddon because when they came out with their 00:06:49.000 --> 00:06:51.579 first quarter revenue, they're up about 200%. 00:06:51.560 --> 00:06:54.019 100 % year over year, just because all of the 00:06:54.019 --> 00:06:57.740 AI workloads have this massive shortage of RAM. 00:06:57.980 --> 00:07:00.259 And if you take into account Samsung and everything 00:07:00.259 --> 00:07:02.939 going on there, SK Hynix and Samsung are both 00:07:02.939 --> 00:07:06.519 spending over $550 billion to build new memory 00:07:06.519 --> 00:07:08.620 factories. This is one of the largest coordinated 00:07:08.620 --> 00:07:10.759 equipment investments in semiconductor history. 00:07:10.899 --> 00:07:13.660 So they're working together and there is a massive 00:07:13.660 --> 00:07:16.699 investment that is going in right now. Solos 00:07:16.699 --> 00:07:20.160 is launching Ergo A6 smart glasses. They're only 00:07:20.160 --> 00:07:23.060 19. and they've decided to cut out the camera 00:07:23.060 --> 00:07:26.759 entirely, but they do have voice AI built into 00:07:26.759 --> 00:07:28.759 them. This is a really interesting play because 00:07:28.759 --> 00:07:30.699 we have the competitors, which are Meta. We have 00:07:30.699 --> 00:07:32.360 Apple working on smart glasses. We have Google 00:07:32.360 --> 00:07:34.819 that has smart glasses going on right now. But 00:07:34.819 --> 00:07:37.040 Meta and Google both have cameras inside of theirs. 00:07:37.120 --> 00:07:39.199 I'm not exactly sure what's going on with Apple 00:07:39.199 --> 00:07:40.319 because, you know, they're not going to come 00:07:40.319 --> 00:07:43.439 out for a few years and we'll see that in a couple 00:07:43.439 --> 00:07:44.560 of years because they're pretty slow with all 00:07:44.560 --> 00:07:46.879 of this. But it's kind of cool that in the meantime, 00:07:47.019 --> 00:07:48.040 while we're waiting for a lot of these other 00:07:48.040 --> 00:07:50.160 big players to come out with their smart. glasses, 00:07:50.480 --> 00:07:53.240 we're getting some new firms in here making some 00:07:53.240 --> 00:07:56.160 cool stuff. This is super light, they say that 00:07:56.160 --> 00:07:58.360 they're privacy first design, and it's going 00:07:58.360 --> 00:08:00.000 to solve one of the biggest problems that smart 00:08:00.000 --> 00:08:02.519 glasses have. I think a lot of people I've seen 00:08:02.519 --> 00:08:04.399 complaining on LinkedIn and other places about 00:08:04.399 --> 00:08:05.939 the cameras. And you know, if you're walking 00:08:05.939 --> 00:08:07.279 around and people have these matter Ray Bans, 00:08:07.339 --> 00:08:09.199 you never know if you're being recorded, it might 00:08:09.199 --> 00:08:10.779 make people uncomfortable if you're talking to 00:08:10.779 --> 00:08:14.019 them. And so or you know, if it's if you're wearing 00:08:14.019 --> 00:08:15.600 them, and you're worried that the cameras could 00:08:15.600 --> 00:08:17.139 be hacked while you're wearing them all day, 00:08:17.160 --> 00:08:18.920 right, there's a whole bunch of issues. So cutting 00:08:18.920 --> 00:08:20.930 out the cameras on these. It feels great to not 00:08:20.930 --> 00:08:23.269 just be walking around with a giant camera on 00:08:23.269 --> 00:08:25.449 your head all day long. And you get a lot of 00:08:25.449 --> 00:08:27.769 the benefit of the AI, which is, you know, you 00:08:27.769 --> 00:08:29.990 can voice to text, talk to it and get answers 00:08:29.990 --> 00:08:32.129 back on what you're seeing now or what you're 00:08:32.129 --> 00:08:34.250 talking about. Now, I think, of course, the biggest 00:08:34.250 --> 00:08:37.389 drawback here is if I'm looking at my car engine 00:08:37.389 --> 00:08:39.789 and I'm like, where is the specific cap I'm looking 00:08:39.789 --> 00:08:41.570 for? If I have the glasses on, it can actually 00:08:41.570 --> 00:08:43.250 let me know. And in Google's case, they're going 00:08:43.250 --> 00:08:45.629 to be able to like highlight things so you can 00:08:45.629 --> 00:08:47.570 actually see, which is really cool. But this 00:08:47.570 --> 00:08:49.789 won't have that capability. These glasses in 00:08:49.789 --> 00:08:52.230 particular have voice AI and their assistant 00:08:52.230 --> 00:08:54.909 can handle anything from real -time translation, 00:08:55.149 --> 00:08:56.509 which is amazing. I think this is one of the 00:08:56.509 --> 00:08:59.690 greatest use cases of AI in these glasses. Calendar 00:08:59.690 --> 00:09:03.039 reminders, calls, they also let you... hear the 00:09:03.039 --> 00:09:05.120 surroundings through speakers behind your ear 00:09:05.120 --> 00:09:06.980 so they can amplify different things for you, 00:09:07.019 --> 00:09:08.799 which is really useful. I just might have to 00:09:08.799 --> 00:09:10.919 get myself a pair. Vercel, which is the platform 00:09:10.919 --> 00:09:14.120 where millions of AI agents deploy code daily. 00:09:14.259 --> 00:09:16.179 And when I say that, like personally, I'm one 00:09:16.179 --> 00:09:17.960 of those people. Basically, I think all of my 00:09:17.960 --> 00:09:21.080 different softwares and apps are using Vercel 00:09:21.080 --> 00:09:23.179 to host all of the different sites that I have 00:09:23.179 --> 00:09:26.019 running right now. And the reason why is because 00:09:26.019 --> 00:09:28.720 it works incredibly well. And I think I think 00:09:28.720 --> 00:09:30.019 there's like a bunch of other good ones like 00:09:30.019 --> 00:09:31.919 Netlify probably does this. And I think Amazon 00:09:31.919 --> 00:09:35.539 has a version as well. But Vercel seems to be 00:09:35.539 --> 00:09:37.240 one that has some incredible integrations and 00:09:37.240 --> 00:09:40.700 APIs and MCPs, which connect to your cloud. And 00:09:40.700 --> 00:09:43.059 so you can have cloud or codecs or whatever you're 00:09:43.059 --> 00:09:45.490 using to code. integrated really well, so it's 00:09:45.490 --> 00:09:48.429 easy to deploy code. So Vercel is betting that 00:09:48.429 --> 00:09:50.409 the future belongs to companies that are going 00:09:50.409 --> 00:09:53.529 to keep AI models and agents separate. So you're 00:09:53.529 --> 00:09:55.250 not bundling these together by one lab, which 00:09:55.250 --> 00:09:56.809 is basically what we've seen everyone doing so 00:09:56.809 --> 00:09:59.570 far. Their CEO, Guillermo Rauch, says that customers 00:09:59.570 --> 00:10:01.990 are already mixing OpenAI, Anthropic Gemini, 00:10:01.990 --> 00:10:04.070 and open models like DeepSeek into infrastructure 00:10:04.070 --> 00:10:07.750 that treats models as a swappable part. And he 00:10:07.750 --> 00:10:09.860 says that companies that are facilitating that 00:10:09.860 --> 00:10:11.519 are the ones that are going to win. Vercel right 00:10:11.519 --> 00:10:13.879 now, they say that they process about 6 million 00:10:13.879 --> 00:10:17.159 deployments per day. That's roughly half triggered 00:10:17.159 --> 00:10:20.240 by coding agents and over 1 trillion tokens flowing 00:10:20.240 --> 00:10:23.019 through their AI gateway daily. So I mean, this 00:10:23.019 --> 00:10:25.519 really is a giant in the industry. They said 00:10:25.519 --> 00:10:27.539 that they shipped EBE, which is a framework for 00:10:27.539 --> 00:10:30.799 defining agents in plain language. And Vercel 00:10:30.799 --> 00:10:33.899 Sandbox lets you control what data internal agents 00:10:33.899 --> 00:10:35.860 can actually access, which I think is kind of 00:10:35.860 --> 00:10:37.799 the code leaking and training problem that a 00:10:37.799 --> 00:10:38.559 lot of people are working on. worried about. 00:10:38.860 --> 00:10:41.820 They have framed this whole new paradigm by saying 00:10:41.820 --> 00:10:43.929 that this is putting them in direct competition 00:10:43.929 --> 00:10:46.129 with AI labs like OpenAI, which are building 00:10:46.129 --> 00:10:48.549 their own hosting and deployment tools directly 00:10:48.549 --> 00:10:51.490 into their products. Now, I will say I would, 00:10:51.529 --> 00:10:54.070 you know, I don't think I would ever use a hosting 00:10:54.070 --> 00:10:57.570 platform exclusively owned by OpenAI or Claude 00:10:57.570 --> 00:10:59.570 or, you know, Anthropic. The same way I actually 00:10:59.570 --> 00:11:02.669 don't love Lovable or other people that like 00:11:02.669 --> 00:11:04.629 are pretty deeply integrated, even though Lovable 00:11:04.629 --> 00:11:08.289 uses Supabase as the database service. I still 00:11:08.289 --> 00:11:10.509 don't like how deeply integrated it was for me 00:11:10.509 --> 00:11:12.990 to get off of Lovable was a huge pain. So I think 00:11:12.990 --> 00:11:15.169 there's going to be always a place for companies 00:11:15.169 --> 00:11:18.230 like Vercel that are independent and separate 00:11:18.230 --> 00:11:20.730 from the actual labs because if you ever want 00:11:20.730 --> 00:11:22.649 to switch from OpenAI to Cloud or from Cloud 00:11:22.649 --> 00:11:25.470 to Codex, you want to have that kind of flexibility 00:11:25.470 --> 00:11:27.570 where it doesn't really matter what coding agent 00:11:27.570 --> 00:11:29.629 you're using. You've got to stay away from that 00:11:29.629 --> 00:11:31.629 vendor lock -in. And speaking of which, if you 00:11:31.629 --> 00:11:34.210 want to get access to over 80 different AI models 00:11:34.210 --> 00:11:36.190 all on the same account, I'd love for you to 00:11:36.190 --> 00:11:39.049 try out AIbox .ai. That's my own platform. I 00:11:39.049 --> 00:11:41.149 let you access everything from audio, image, 00:11:41.750 --> 00:11:44.350 video, even music generation models. There's 00:11:44.350 --> 00:11:46.090 over 80 of them. I mean, there's all the basic 00:11:46.090 --> 00:11:47.690 ones that you know, like Cloud and OpenAI and 00:11:47.690 --> 00:11:49.750 Gemini, tons of open source ones you haven't 00:11:49.750 --> 00:11:52.009 heard of or tried before, some really cool video 00:11:52.009 --> 00:11:54.889 and audio powered AI tools on there. So if you 00:11:54.889 --> 00:11:57.090 want to check out all of that for only $8 .99 00:11:57.090 --> 00:12:00.470 a month, go check out AIbox .ai. I'll leave a 00:12:00.470 --> 00:12:02.450 link in the description. Otherwise, I hope you 00:12:02.450 --> 00:12:04.330 all have a fantastic rest of your day.