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 --> 00:00:17.269 business. Thinking Machines by Miriam Marotti, 00:00:17.370 --> 00:00:20.370 who famously left OpenAI after she was the CEO 00:00:20.370 --> 00:00:22.589 when Sam Altman got kicked out, have released 00:00:22.589 --> 00:00:25.589 their first open -weight AI model called Inkling. 00:00:25.769 --> 00:00:28.609 OpenAI has built something called GPT -RED. It's 00:00:28.609 --> 00:00:31.329 an LLM super hacker, and they built this to harden 00:00:31.329 --> 00:00:34.850 their own models. AI Box has shipped an MPC server 00:00:34.850 --> 00:00:37.490 that brings 80 different AI models into Cloud, 00:00:37.609 --> 00:00:41.829 ChatGPT, and Gemini. AWS is committing $1 billion 00:00:41.829 --> 00:00:44.409 to a forward -deployed engineering organization. 00:00:45.020 --> 00:00:47.200 OpenAI and Anthropic are both scaling theirs. 00:00:47.460 --> 00:00:50.340 Apple Intelligence has been cleared for a China 00:00:50.340 --> 00:00:53.859 launch with Alibaba's Quen and Meta plans to 00:00:53.859 --> 00:00:57.140 sell and resell their AI compute following what 00:00:57.140 --> 00:00:59.960 SpaceX is doing and kind of copying their playbook. 00:01:00.060 --> 00:01:02.200 Let's kick this off with Thinking Machines. This 00:01:02.200 --> 00:01:04.379 is a company that I'm really rooting for. They've 00:01:04.379 --> 00:01:06.480 raised, you know, over a billion dollars, but 00:01:06.480 --> 00:01:08.560 they haven't come out with anything super groundbreaking. 00:01:08.760 --> 00:01:10.719 But what I will say when it comes to a lot of 00:01:10.719 --> 00:01:13.099 these models, they're so expensive and they take 00:01:13.099 --> 00:01:15.670 so much time that when you think of something 00:01:15.670 --> 00:01:17.349 like even Anthropic, it felt like OpenAI had 00:01:17.349 --> 00:01:19.150 just run away from them and they were never going 00:01:19.150 --> 00:01:21.930 to catch up. And little by little, they pick 00:01:21.930 --> 00:01:23.829 their lane, they make their product better, and 00:01:23.829 --> 00:01:25.489 they're able to get market share until the point 00:01:25.489 --> 00:01:27.409 where Anthropic has now exploded and is making 00:01:27.409 --> 00:01:29.489 more money than OpenAI. It surpassed them in 00:01:29.489 --> 00:01:32.209 revenue. I think we might see a lot of that same 00:01:32.209 --> 00:01:34.150 strategy played out by a lot of different AI 00:01:34.150 --> 00:01:36.250 companies if they don't kind of shrivel up and 00:01:36.250 --> 00:01:39.090 die or get sold off for parts or, you know, an 00:01:39.090 --> 00:01:40.870 acqui -hire or something like that. And so Thinking 00:01:40.870 --> 00:01:42.290 Machines is one of these companies that I'm really 00:01:42.290 --> 00:01:45.060 excited about. I hope that they really go places. 00:01:45.219 --> 00:01:47.700 But this is run by Miriam Roddy, who is famously 00:01:47.700 --> 00:01:51.120 the CEO of OpenAI when Sam Altman left. And she 00:01:51.120 --> 00:01:53.280 has created this thing called Inkling. It's an 00:01:53.280 --> 00:01:56.879 open weight AI model. It has 975 billion parameters. 00:01:57.140 --> 00:01:59.540 And companies can download this and customize 00:01:59.540 --> 00:02:02.040 it themselves. So you don't have to go and pay 00:02:02.040 --> 00:02:05.200 for API access to OpenAI or... Anthropic, you 00:02:05.200 --> 00:02:07.640 can actually just go and fine tune this all on 00:02:07.640 --> 00:02:09.500 your own. You can fine tune it on your own data. 00:02:09.599 --> 00:02:11.659 And the bet that they want people to take is 00:02:11.659 --> 00:02:14.400 that that is going to be better than a one size 00:02:14.400 --> 00:02:17.900 fits all model that OpenAI or Anthropic or Gemini 00:02:17.900 --> 00:02:20.699 or any of these other big labs are selling. Inkling 00:02:20.699 --> 00:02:23.960 was trained on 45 trillion tokens of text, image, 00:02:24.020 --> 00:02:26.620 and audio and video. They did it in about nine 00:02:26.620 --> 00:02:29.319 months, which is way faster than OpenAI's five 00:02:29.319 --> 00:02:32.000 -year timeline or Anthropic's three years. In 00:02:32.000 --> 00:02:34.990 a test with Bridgewater Associates, The financial 00:02:34.990 --> 00:02:37.590 model trained on the hedge fund's own expertise 00:02:37.590 --> 00:02:41.370 scored 84 .7 % on financial reasoning benchmarks 00:02:41.370 --> 00:02:45.729 at roughly 1 14th the cost of the top models 00:02:45.729 --> 00:02:48.129 from Anthropic and OpenAI. So we're seeing some 00:02:48.129 --> 00:02:50.069 massive improvements when you're taking this 00:02:50.069 --> 00:02:51.650 kind of these kind of models and you're fine 00:02:51.650 --> 00:02:54.090 tuning it on your own data. The model activates 00:02:54.090 --> 00:02:57.310 only 41 billion of its 975 billion parameters 00:02:57.310 --> 00:02:59.830 per task, and users can dial up thinking efforts 00:02:59.830 --> 00:03:02.750 to trade speed for accuracy. So I'm rooting for 00:03:02.750 --> 00:03:04.449 them, but time will tell how well this model 00:03:04.449 --> 00:03:06.610 does. We just learned that OpenAI built something 00:03:06.610 --> 00:03:09.110 called GPT -RED. This is an AI model trained 00:03:09.110 --> 00:03:11.849 to attack their own systems, and they use it 00:03:11.849 --> 00:03:14.729 to catch security flaws before release. So the 00:03:14.729 --> 00:03:18.009 model cuts successful attacks on GPT -5 .6 from 00:03:18.009 --> 00:03:21.930 over 90 % down to 23%, which basically makes 00:03:21.930 --> 00:03:24.259 it one of OpenAI's most secure releases yet. 00:03:24.300 --> 00:03:27.099 One particularly interesting attack that it was 00:03:27.099 --> 00:03:30.139 able to discover is called a novel fake chain 00:03:30.139 --> 00:03:32.219 of thought, which basically is tricking a model 00:03:32.219 --> 00:03:35.400 into making up fake reasoning steps. So it's 00:03:35.400 --> 00:03:37.319 kind of similar to convincing someone that, you 00:03:37.319 --> 00:03:40.240 know, one plus one equals three. And then, you 00:03:40.240 --> 00:03:41.919 know, saying that I already checked the math, 00:03:42.060 --> 00:03:44.120 that's what equals and if it equals that, then 00:03:44.120 --> 00:03:45.979 therefore and you, you know, go trick them on 00:03:45.979 --> 00:03:47.580 the next thing. So they found that when they 00:03:47.580 --> 00:03:50.819 gave the same task as a human red teamer who 00:03:50.819 --> 00:03:55.300 tested GPT -5, in 2025, GPT -RED found more effective 00:03:55.300 --> 00:03:57.539 attacks than the humans. They also got it to 00:03:57.539 --> 00:04:00.400 go and hack like this third party vending machine 00:04:00.400 --> 00:04:02.259 agent, which is kind of funny. It does have a 00:04:02.259 --> 00:04:04.159 bunch of limitations, so it's not very good at 00:04:04.159 --> 00:04:06.379 back and forth conversation attacks, and it's 00:04:06.379 --> 00:04:09.240 not very good at exploiting images that are embedded 00:04:09.240 --> 00:04:12.319 into malicious text. The idea behind this is 00:04:12.319 --> 00:04:15.479 actually functionally working is that GPT -RED 00:04:15.479 --> 00:04:17.819 is automating how all of the security holes are 00:04:17.819 --> 00:04:20.540 found because it puts an attacker model against 00:04:20.540 --> 00:04:22.060 a defender. model, and they put them in these 00:04:22.060 --> 00:04:23.860 kind of simulated real world environments, and 00:04:23.860 --> 00:04:26.279 they're battling it out. OpenAI is not releasing 00:04:26.279 --> 00:04:27.860 this externally. So they're not giving this to 00:04:27.860 --> 00:04:29.639 other people to test, which is interesting, right? 00:04:29.699 --> 00:04:31.660 Because we had the whole anthropic mythos model, 00:04:31.759 --> 00:04:33.300 which was really good at security exploits, and 00:04:33.300 --> 00:04:35.199 they gave it out to all the top labs and said, 00:04:35.259 --> 00:04:37.079 hey, like, go harden all your security with this. 00:04:37.139 --> 00:04:39.660 So OpenAI is not giving this out externally beyond 00:04:39.660 --> 00:04:42.139 just using it for themselves. And I mean, it's 00:04:42.139 --> 00:04:45.399 got a huge massive drop in the success rate of 00:04:45.399 --> 00:04:47.430 a lot of these exploits. I think it shows AI 00:04:47.430 --> 00:04:49.930 powered red teaming can be just as effective 00:04:49.930 --> 00:04:52.129 or more effective than humans. I mean, you can 00:04:52.129 --> 00:04:55.410 just brute force way more tests than humans. 00:04:55.470 --> 00:04:56.850 And I mean, it's trained off of what humans are 00:04:56.850 --> 00:04:58.689 doing. But at this point, it's doing better than 00:04:58.689 --> 00:05:01.129 what a lot of the humans are doing. AI box, which 00:05:01.129 --> 00:05:03.810 full disclosure is my own startup has just released 00:05:03.810 --> 00:05:07.149 an MCP server and we are allowing people to plug 00:05:07.149 --> 00:05:09.730 80 different AI models straight into cloud chat, 00:05:09.790 --> 00:05:12.350 GPT, Gemini or cursor. Basically, it lets you 00:05:12.350 --> 00:05:15.069 use any model inside of whatever assistant you 00:05:15.069 --> 00:05:16.750 already are used to using. Personally, I use 00:05:17.509 --> 00:05:19.209 clawed all day long, although I'm kind of switching 00:05:19.209 --> 00:05:21.689 over to ChatGPT with their new ChatGPT app. It's 00:05:21.689 --> 00:05:24.370 really amazing. But both either way, you get 00:05:24.370 --> 00:05:27.829 the AI Box MCP and it allows you to access all 00:05:27.829 --> 00:05:30.170 of the images that... OpenAI can generate. It 00:05:30.170 --> 00:05:32.329 allows you, if you're on Claude or ChatGPT, to 00:05:32.329 --> 00:05:34.769 generate all of the videos that Google V03 can 00:05:34.769 --> 00:05:36.769 make. And no matter what platform you're using, 00:05:36.829 --> 00:05:39.350 you can access what Eleven Labs can do with audio. 00:05:39.670 --> 00:05:42.089 I spent basically the entire day today getting 00:05:42.089 --> 00:05:44.649 ready for a big Facebook campaign that we're 00:05:44.649 --> 00:05:46.750 launching here at AI Box. And in the past, doing 00:05:46.750 --> 00:05:49.370 Facebook campaigns usually meant hiring a Facebook 00:05:49.370 --> 00:05:52.230 ads team. It meant creating tons of creative, 00:05:52.350 --> 00:05:54.889 which just takes a lot of time. And it meant, 00:05:54.949 --> 00:05:57.009 you know, spending a ton of time on landing pages. 00:05:57.209 --> 00:06:00.319 With Claude, and with the AI Box MCP that we 00:06:00.319 --> 00:06:02.759 built into it, I actually got almost all of the, 00:06:02.860 --> 00:06:04.500 so basically what I did is I went and recorded 00:06:04.500 --> 00:06:06.759 a bunch of just selfies of me talking about the 00:06:06.759 --> 00:06:09.300 product on my phone. I dropped those into a folder 00:06:09.300 --> 00:06:11.720 and I had Claude go and edit all of them. And 00:06:11.720 --> 00:06:13.379 then I was able to say, hey, I need to create 00:06:13.379 --> 00:06:15.639 a bunch of dynamic, so like image generated ads. 00:06:15.819 --> 00:06:17.779 I gave it the style that I wanted. I have these 00:06:17.779 --> 00:06:19.259 different ideas. Some of them, it looks like 00:06:19.259 --> 00:06:21.439 kind of like breaking news images. Some of them, 00:06:21.480 --> 00:06:23.839 they actually look like, it's like me doing a 00:06:23.839 --> 00:06:25.819 FaceTime call and there's like a text message 00:06:25.819 --> 00:06:27.399 popping up on the screen and it's some sort. 00:06:27.500 --> 00:06:29.680 of text message about AI box. So anyways, I had 00:06:29.680 --> 00:06:31.180 all these different creative ideas that I went 00:06:31.180 --> 00:06:33.300 and got from a bunch of people and to go and, 00:06:33.339 --> 00:06:35.240 you know, create a template on Canva and then 00:06:35.240 --> 00:06:36.980 switch out tons of different variations of the 00:06:36.980 --> 00:06:39.300 title and the backgrounds and stuff like that 00:06:39.300 --> 00:06:41.459 takes a lot of time. And before Claude couldn't 00:06:41.459 --> 00:06:43.000 do it because it couldn't do any of the image 00:06:43.000 --> 00:06:45.240 generation. But now that AI box is embedded inside 00:06:45.240 --> 00:06:46.980 of Claude and it can do the image generation. 00:06:47.079 --> 00:06:49.220 One in particular that I did is I was showing, 00:06:49.259 --> 00:06:51.779 uh, I was showing the ability to generate images 00:06:51.779 --> 00:06:53.699 inside of Claude and I needed that generated 00:06:53.699 --> 00:06:56.079 as an ad. So you had to have like basically, 00:06:56.160 --> 00:06:59.259 uh, the design of a Claude interface. And then 00:06:59.259 --> 00:07:01.699 you had to be able to show images generated and, 00:07:01.699 --> 00:07:04.199 you know, have like AI box logos and stuff. So 00:07:04.199 --> 00:07:06.819 what I did is I was like, hey, go create this 00:07:06.819 --> 00:07:10.360 mockup inside of Claude. But for the image part, 00:07:10.439 --> 00:07:12.319 just go use AI box to generate the image and 00:07:12.319 --> 00:07:14.560 pull that inside of, it's like the image inside 00:07:14.560 --> 00:07:16.759 of the image that it's generating. Is able to 00:07:16.759 --> 00:07:18.180 do that in a couple seconds. And I said, this 00:07:18.180 --> 00:07:20.480 is awesome. Go think of like 10 variations or 00:07:20.480 --> 00:07:23.540 like 10 different use cases of my product. And, 00:07:23.540 --> 00:07:25.699 you know, and in this case, it's the product 00:07:25.699 --> 00:07:27.529 is. showing that you can create images. This 00:07:27.529 --> 00:07:30.389 is kind of a meta example, I know. But I was 00:07:30.389 --> 00:07:31.610 like, go think of like 10 different variations. 00:07:31.610 --> 00:07:32.930 So it's like, okay, well, you could use it for 00:07:32.930 --> 00:07:35.009 like UGC content, you could use it for like, 00:07:35.029 --> 00:07:37.970 coming up with product images for logos for newsletter 00:07:37.970 --> 00:07:39.689 banners, whatever came up with all these ideas. 00:07:39.910 --> 00:07:42.430 And inside of the image ad that it's creating, 00:07:42.610 --> 00:07:44.870 it would go to AI box anytime it had to generate 00:07:44.870 --> 00:07:47.550 one of the actual graphics inside of its image. 00:07:47.589 --> 00:07:50.209 So like the you know, the graphic of the logo 00:07:50.209 --> 00:07:53.569 or the graphic of the UGC person, whatever the 00:07:53.569 --> 00:07:55.889 actual image was, it would hit AI box, it would 00:07:55.889 --> 00:07:57.689 generate the image, it would pull inside of its 00:07:57.689 --> 00:07:59.410 own thing. So I didn't have to do anything. I 00:07:59.410 --> 00:08:01.269 literally just said, go, you know, we have the 00:08:01.269 --> 00:08:03.430 concept, we have the template that I like, go 00:08:03.430 --> 00:08:05.910 create 20 variations, think of a bunch of good 00:08:05.910 --> 00:08:07.870 use cases and use AI box for all the images. 00:08:08.069 --> 00:08:10.810 And the cool thing is when when Claude was making 00:08:10.810 --> 00:08:14.620 this specific kind of like UI mockup. Um, because 00:08:14.620 --> 00:08:16.740 it's UI and it's not just like I went to chat 00:08:16.740 --> 00:08:18.939 GBT and got it to generate like one singular 00:08:18.939 --> 00:08:21.319 image, which isn't perfect. And the text might 00:08:21.319 --> 00:08:23.500 be funky and someone might be a little bit off 00:08:23.500 --> 00:08:25.160 or whatever, right? If you're just doing an image 00:08:25.160 --> 00:08:27.879 because Claude was kind of, uh, just creating 00:08:27.879 --> 00:08:30.379 it with code and then pulling an image into it. 00:08:30.420 --> 00:08:32.620 I said, okay, this is awesome. Now go put this 00:08:32.620 --> 00:08:34.940 actual graphic onto our website and animate it. 00:08:34.980 --> 00:08:36.799 So like where the chat bubble is, I'm like, have 00:08:36.799 --> 00:08:38.799 someone typing that out, have the bubble appear, 00:08:38.960 --> 00:08:40.740 have a loading screen that makes the image pop 00:08:40.740 --> 00:08:43.639 up. So it's so cool because Claude is able. to 00:08:43.639 --> 00:08:45.700 build these things. And because we're just pulling 00:08:45.700 --> 00:08:49.320 the image in with AI Box, it's not just a picture, 00:08:49.460 --> 00:08:51.720 but it's a full element that can be animated 00:08:51.720 --> 00:08:53.559 for the landing page. Then it can be turned into 00:08:53.559 --> 00:08:56.480 an ad for the Facebook ad and you can actually 00:08:56.480 --> 00:08:58.919 have the animated as a video ad. Anyways, so 00:08:58.919 --> 00:09:00.779 many possibilities. But if you're doing anything 00:09:00.779 --> 00:09:03.539 with Cloud, which doesn't have audio, video or 00:09:03.539 --> 00:09:06.460 image capabilities, just go get the AI Box MCP. 00:09:06.539 --> 00:09:09.259 It's like $7 .99 a month, I think, or $8 .99 00:09:09.259 --> 00:09:12.519 a month. And you get access to over 80 different... 00:09:12.720 --> 00:09:14.820 models and anything you need while you're using 00:09:14.820 --> 00:09:17.120 Claude, it can go and grab that and pull it in 00:09:17.120 --> 00:09:19.659 and save you so much time. No more back and forth. 00:09:19.740 --> 00:09:21.700 So go check it out. There's a link in the description 00:09:21.700 --> 00:09:25.559 to AI box dot AI slash MCP. If you want to get 00:09:25.559 --> 00:09:28.110 started with that. AWS is committing $1 billion 00:09:28.110 --> 00:09:31.009 to a new team of engineers who are going to be 00:09:31.009 --> 00:09:34.330 embedded inside of customers or in their customers' 00:09:34.509 --> 00:09:36.370 companies. And they're going to be building custom 00:09:36.370 --> 00:09:38.730 AI agents tailored to each business. And they're 00:09:38.730 --> 00:09:40.990 then going to hand off the working system when 00:09:40.990 --> 00:09:42.990 the project ends. There's been a ton of these 00:09:42.990 --> 00:09:46.289 companies being spun up by Anthropic and OpenAI. 00:09:46.490 --> 00:09:49.110 They're basically making these forward -deployed 00:09:49.110 --> 00:09:51.210 engineers. So you send the engineer into your 00:09:51.210 --> 00:09:53.470 customer's company. They build some sort of automation 00:09:53.470 --> 00:09:55.549 workflow. And I think it's kind of like OpenAI. 00:09:55.720 --> 00:09:57.299 And Anthropic's way of saying, hey, look, we 00:09:57.299 --> 00:09:59.860 know you guys want AI, but you also don't know 00:09:59.860 --> 00:10:01.500 how to use it. We are the experts. We'll just 00:10:01.500 --> 00:10:03.799 send some people in there to build this thing 00:10:03.799 --> 00:10:05.799 for you. They're directly copying the playbook 00:10:05.799 --> 00:10:08.620 that OpenAI spent $4 billion on and Anthropic 00:10:08.620 --> 00:10:14.039 spent $1 .5 billion on so far. And in those cases, 00:10:14.220 --> 00:10:15.919 OpenAI and Anthropic both partnered with some 00:10:15.919 --> 00:10:18.600 private equity firms to actually fund the staff 00:10:18.600 --> 00:10:22.679 for all of this FDE teams. Anthropic is doing 00:10:22.679 --> 00:10:25.159 it entirely with internal Amazon resources. instead. 00:10:25.500 --> 00:10:27.740 And I'm going to be honest, this might be an 00:10:27.740 --> 00:10:30.919 interesting strategy. If I was any of these big 00:10:30.919 --> 00:10:33.460 AI companies, and you know, like, let's say meta, 00:10:33.519 --> 00:10:34.799 for example, and you're thinking about doing 00:10:34.799 --> 00:10:37.519 layoffs, well, maybe instead of meta doing layoffs, 00:10:37.539 --> 00:10:40.120 they should build one of these forward deployed 00:10:40.120 --> 00:10:43.740 engineer teams, get a bunch of money and go and 00:10:43.740 --> 00:10:46.539 get, you know, meta embedded into their customers 00:10:46.539 --> 00:10:49.100 businesses and just deploy the engineers over 00:10:49.100 --> 00:10:50.899 there. Anyways, I'm not sure if this is Amazon 00:10:50.899 --> 00:10:52.899 getting around layoffs, but it is an interesting 00:10:52.899 --> 00:10:54.960 strategy, you can imagine. because OpenAI and 00:10:54.960 --> 00:10:57.399 Anthropic definitely didn't have any extra engineers 00:10:57.399 --> 00:10:59.539 they had to partner with people for this. AWS 00:10:59.539 --> 00:11:02.639 has a really big advantage, I would say structurally, 00:11:02.799 --> 00:11:05.059 because when the project ends, the customer runs 00:11:05.059 --> 00:11:08.019 AI agents on AWS's cloud, and that's basically 00:11:08.019 --> 00:11:09.860 just creating long -term revenue. OpenAI and 00:11:09.860 --> 00:11:12.360 Anthropic make money per token used. That's a 00:11:12.360 --> 00:11:15.179 lot smaller of a payoff per engagement, but the 00:11:15.179 --> 00:11:19.000 cloud is a really big win for AWS. Also, I say 00:11:19.000 --> 00:11:22.519 all of this like AWS is copying OpenAI and Anthropic 00:11:22.519 --> 00:11:24.129 like they pioneered. this. But this is actually 00:11:24.129 --> 00:11:26.909 something that Palantir has spent the last 20 00:11:26.909 --> 00:11:29.330 years doing. And so I don't think this is, you 00:11:29.330 --> 00:11:30.809 know, something that's just brand new in the 00:11:30.809 --> 00:11:32.769 situation. It's something that's been done in 00:11:32.769 --> 00:11:34.950 the past. And it's it's kind of something that's 00:11:34.950 --> 00:11:37.429 been successful. We'll see how big Amazon is 00:11:37.429 --> 00:11:40.350 able to scale this. Apple has just won regulatory 00:11:40.350 --> 00:11:42.509 approval to launch Apple Intelligence in China. 00:11:42.509 --> 00:11:44.629 They're partnering with Alibaba to power all 00:11:44.629 --> 00:11:46.450 of the AI features there. It's interesting, right? 00:11:46.509 --> 00:11:48.190 Because Apple said, hey, look, like they were 00:11:48.190 --> 00:11:49.929 going to have their own AI in Syria. And yeah, 00:11:49.970 --> 00:11:51.710 that probably would have been a big pain for 00:11:51.710 --> 00:11:53.909 them with China. But then they said, look, we're 00:11:53.909 --> 00:11:57.610 just going to say anyone with an iPhone can go 00:11:57.610 --> 00:12:00.309 and use whatever AI model they choose. So in 00:12:00.309 --> 00:12:03.470 China, that's going to be something from Alibaba. 00:12:03.549 --> 00:12:05.370 And in America, that's probably going to be Anthropic 00:12:05.370 --> 00:12:07.350 or OpenAI or whatever the model that you probably 00:12:07.350 --> 00:12:09.649 pay for that you have. premium of you could probably 00:12:09.649 --> 00:12:12.450 plug that straight into siri i think this matters 00:12:12.450 --> 00:12:14.809 a lot because china is apple's second largest 00:12:14.809 --> 00:12:18.289 market 20 .5 billion dollars in sales last quarter 00:12:18.289 --> 00:12:21.210 and i think being able to close that gap helps 00:12:21.210 --> 00:12:23.730 keep apple their number two smartphone position 00:12:23.730 --> 00:12:27.070 there's rivals like huawei alibaba's quen model 00:12:27.070 --> 00:12:29.429 is going to handle text and image understanding 00:12:29.429 --> 00:12:31.750 and generation across all of the different you 00:12:31.750 --> 00:12:35.529 know ipad and mac and vision os and apple's exploring 00:12:35.529 --> 00:12:38.679 a deal with baidu deep seek and ByteDance. I 00:12:38.679 --> 00:12:41.279 think they finally have settled on Alibaba after 00:12:41.279 --> 00:12:43.440 kind of talking to all of the other options. 00:12:44.039 --> 00:12:47.460 Alibaba's US listed shares went up 6 % when that 00:12:47.460 --> 00:12:49.139 was announced. This deal is basically going to 00:12:49.139 --> 00:12:51.559 give Quen some built -in distribution to the 00:12:51.559 --> 00:12:54.139 very new iPhone that will be sold in China. I'm 00:12:54.139 --> 00:12:56.259 going to be honest, I've tried some Quen models 00:12:56.259 --> 00:12:58.940 and been really impressed. Their text -to -speech 00:12:58.940 --> 00:13:01.480 model in particular is really good. But kind 00:13:01.480 --> 00:13:03.539 of the bigger story for me on all of this is 00:13:03.539 --> 00:13:07.179 that for not just Apple, but for any of the Western 00:13:07.179 --> 00:13:10.000 companies, China is not going to approve any 00:13:10.000 --> 00:13:12.419 sort of Western AI service if there isn't a local 00:13:12.419 --> 00:13:15.240 model partner that's locked in and that is powering 00:13:15.240 --> 00:13:18.139 everything. Meta is launching a cloud business 00:13:18.139 --> 00:13:19.679 called Meta Compute. They're going to be selling 00:13:19.679 --> 00:13:22.299 AI compute powered AI models to other companies, 00:13:22.419 --> 00:13:24.379 and they're doing this directly to compete with 00:13:24.379 --> 00:13:27.220 Amazon, Google, and Microsoft. And all this is 00:13:27.220 --> 00:13:30.840 right after Meta committed a whopping $182 .9 00:13:30.840 --> 00:13:34.019 billion to AI infrastructure, which is basically 00:13:34.019 --> 00:13:36.019 the SpaceX strategy. They're turning all of their 00:13:36.019 --> 00:13:38.759 excess data center capacity into revenue. It's 00:13:38.759 --> 00:13:40.740 interesting, right? Because Meta, I think, hopes 00:13:40.740 --> 00:13:42.759 that or would have hoped that their AI model 00:13:42.759 --> 00:13:45.480 was way more used and way more popular. But having 00:13:45.480 --> 00:13:47.700 all All of the extra AI infrastructure is, I 00:13:47.700 --> 00:13:49.759 mean, basically a goldmine. It's money that they're 00:13:49.759 --> 00:13:51.460 not spending and now money that they're actually 00:13:51.460 --> 00:13:54.580 making, similar to Grok, I think. Meta is going 00:13:54.580 --> 00:13:57.039 to sell both raw computing capacity, basically 00:13:57.039 --> 00:13:59.379 just like CoreWeave does, and they're also going 00:13:59.379 --> 00:14:01.340 to have access to their own AI models that they'll 00:14:01.340 --> 00:14:03.600 sell, including the recently launched Muse Spark 00:14:03.600 --> 00:14:05.559 model, which has gotten a lot of attention because 00:14:05.559 --> 00:14:08.220 it kind of finally pushes Meta to the forefront. 00:14:08.440 --> 00:14:12.000 The Ohio data center that is described by Mark 00:14:12.000 --> 00:14:15.490 as a Manhattan -sized project. is expected to 00:14:15.490 --> 00:14:17.370 open this year, and it's going to provide all 00:14:17.370 --> 00:14:19.990 of the excess capacity that Meta needs to resell. 00:14:20.110 --> 00:14:22.610 And SpaceX basically showed this model works. 00:14:22.669 --> 00:14:24.889 XAI signed deals in May with Anthropic, Google, 00:14:25.009 --> 00:14:28.129 and Reflection AI to buy Compute Time, and they 00:14:28.129 --> 00:14:30.090 basically turned all of their idle capacity into 00:14:30.090 --> 00:14:33.110 immediate revenue, and people paid a lot of money. 00:14:33.149 --> 00:14:34.889 I think Anthropic is paying over a billion dollars 00:14:34.889 --> 00:14:37.350 a month for access to that. For companies with 00:14:37.350 --> 00:14:39.149 deep pockets, this seems to be a good strategy. 00:14:39.169 --> 00:14:41.990 If at any point Meta's AI models get super, super 00:14:41.990 --> 00:14:44.070 popular, right, they can go and use their own. 00:14:44.110 --> 00:14:46.409 capacity. But if not, they're just going to make 00:14:46.409 --> 00:14:49.389 all of the money from all of the other AI companies. 00:14:49.529 --> 00:14:50.950 And I think at the end of the day, the demand 00:14:50.950 --> 00:14:53.590 for AI is not going to decrease. So it's a pretty 00:14:53.590 --> 00:14:55.789 smart bet. Guys, that was everything for the 00:14:55.789 --> 00:14:57.970 podcast today. Thank you so much for tuning in. 00:14:58.009 --> 00:14:59.909 If you enjoyed today's episode, make sure to 00:14:59.909 --> 00:15:02.710 go check out AI box .ai. Like I mentioned, where 00:15:02.710 --> 00:15:05.259 you can get our MCP that gives you 80 different 00:15:05.259 --> 00:15:06.820 AI models. And more importantly, if you're using 00:15:06.820 --> 00:15:09.679 Claude, you get images, audio, and video generated 00:15:09.679 --> 00:15:12.059 right inside of Claude. It's like actually magical 00:15:12.059 --> 00:15:13.799 to use. I've been using it all day. It takes 00:15:13.799 --> 00:15:16.740 two seconds to connect and it's only $8. Please, 00:15:16.820 --> 00:15:18.720 I beg you, if you have Claude, you have to try 00:15:18.720 --> 00:15:20.779 this out. It will just make your life so much 00:15:20.779 --> 00:15:23.000 easier. Also, if you want to get all of these 00:15:23.000 --> 00:15:24.940 different stories that I talk about on the podcast 00:15:24.940 --> 00:15:28.039 here straight into your inbox, go check out AIChatDaily 00:15:28.039 --> 00:15:30.279 .com. That's my website, my news site that is 00:15:30.279 --> 00:15:32.820 tied to this. I have deep dive articles on every 00:15:32.820 --> 00:15:35.370 story that I covered here. And I also have links 00:15:35.370 --> 00:15:37.409 over there where you can go and read more about 00:15:37.409 --> 00:15:39.409 it. There's a subscribe tab at the top of that 00:15:39.409 --> 00:15:41.450 website where you can get this all as an email 00:15:41.450 --> 00:15:43.529 or you can go get the deep dives on the site. 00:15:43.570 --> 00:15:45.169 Guys, thank you so much for tuning in and I will 00:15:45.169 --> 00:15:46.370 catch you in the next episode.