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 --> 00:00:17.120 business. Anthropic has just launched a brand 00:00:17.120 --> 00:00:20.559 new model, Claude Opus 5. It is half the price 00:00:20.559 --> 00:00:23.940 of Fable 5, but almost as good as Fable 5. This, 00:00:24.019 --> 00:00:26.260 I think, is in direct response to what OpenAI 00:00:26.260 --> 00:00:28.320 has been doing. And speaking of which, OpenAI 00:00:28.320 --> 00:00:31.059 has just brought ChatGPT Voice to their desktop 00:00:31.059 --> 00:00:33.740 app, and it has agent control. So talking to 00:00:33.740 --> 00:00:36.820 it, it can then control your computer. Meta is 00:00:36.820 --> 00:00:39.179 pivoting their AI chatbot towards productivity 00:00:39.179 --> 00:00:42.219 with MuseSpark 1 .1. We'll get into what they're 00:00:42.219 --> 00:00:45.320 doing there. But Hugging Face, Meta, NVIDIA have 00:00:45.320 --> 00:00:47.539 all signed, along with over 200 Silicon Valley 00:00:47.539 --> 00:00:50.840 companies, a letter against open weight AI restrictions. 00:00:51.200 --> 00:00:54.780 And a researcher is using AlphaFold to redesign 00:00:54.780 --> 00:00:59.659 CRISPR -Cas9 and cut off target edits from 28 00:00:59.659 --> 00:01:03.799 % to 5%. MidJourney has just acquired CoStars. 00:01:03.840 --> 00:01:07.040 It's an astrology app. 4 .3 million monthly users. 00:01:07.260 --> 00:01:09.980 And it has nothing to do with mid journey, but 00:01:09.980 --> 00:01:12.299 the talent that have built it have everything 00:01:12.299 --> 00:01:13.579 to do with their company. So we're going to get 00:01:13.579 --> 00:01:15.840 into that. But first, my absolute favorite AI 00:01:15.840 --> 00:01:17.879 tool I've been using this week allows you to 00:01:17.879 --> 00:01:21.099 generate images and audio and video right inside 00:01:21.099 --> 00:01:23.459 of Claude, who, you know, only does text and 00:01:23.459 --> 00:01:25.599 reason. And I basically use Claude as my control 00:01:25.599 --> 00:01:27.700 panel for everything I do. I have two Claude 00:01:27.700 --> 00:01:30.980 Max $200 a month subscriptions, and I burn through 00:01:30.980 --> 00:01:33.239 all of the tokens for both of those because I 00:01:33.239 --> 00:01:35.969 have a billion projects going. always all at 00:01:35.969 --> 00:01:38.670 once. And this thing is a workhorse. Now, the 00:01:38.670 --> 00:01:40.709 problem with it is it doesn't generate images 00:01:40.709 --> 00:01:43.109 like chat GPT, it doesn't generate video like 00:01:43.109 --> 00:01:45.590 Google, and it doesn't do audio like 11 lab. 00:01:45.650 --> 00:01:47.069 So I've actually been able to pull all of those 00:01:47.069 --> 00:01:50.310 models straight into Claude with the AI box MCP. 00:01:50.469 --> 00:01:52.769 It's amazing. I've created tons of Facebook ads 00:01:52.769 --> 00:01:54.510 this week. And the thing that actually saved 00:01:54.510 --> 00:01:56.730 me a ton of money and also just time is I was 00:01:56.730 --> 00:01:59.129 creating a bunch of podcast covers for a project 00:01:59.129 --> 00:02:02.049 where it was a it was a language learning podcast. 00:02:02.170 --> 00:02:04.959 So it's like learn German. right? And it'll be 00:02:04.959 --> 00:02:07.219 a podcast in German. But my problem was I was 00:02:07.219 --> 00:02:09.180 like, you know, people that are from, you know, 00:02:09.180 --> 00:02:12.020 Spain or from Russia that want to learn German, 00:02:12.120 --> 00:02:14.219 they're not going to search for learn German 00:02:14.219 --> 00:02:17.060 in English or even learn German and however you 00:02:17.060 --> 00:02:19.199 say it in in Dutch, they're going to search it 00:02:19.199 --> 00:02:21.879 in their own language of Russian or whatever. 00:02:21.979 --> 00:02:24.719 So anyways, I have this problem. And the solution 00:02:24.719 --> 00:02:28.240 is generating podcast covers with chat GPT, you 00:02:28.240 --> 00:02:30.639 know, learn, you say the words learn German, 00:02:30.699 --> 00:02:34.020 and you'd translate it to 10 different 10 different 00:02:34.020 --> 00:02:35.680 languages. Okay, so I had this great podcast 00:02:35.680 --> 00:02:37.740 cover design and that I really liked. And I was 00:02:37.740 --> 00:02:39.580 just getting chat GPT, I was giving it to it, 00:02:39.599 --> 00:02:41.300 it was spitting out a new title. And I know a 00:02:41.300 --> 00:02:42.560 lot of people are like, Oh, why don't you just 00:02:42.560 --> 00:02:44.659 use Canva to do this? Yeah, you could use Canva 00:02:44.659 --> 00:02:46.479 to do it. And then you'd have to manually adjust 00:02:46.479 --> 00:02:48.979 like the font sizes. Because once you get into 00:02:48.979 --> 00:02:52.159 like Mandarin or Korean, and like the languages 00:02:52.159 --> 00:02:54.199 are completely different. I had two different 00:02:54.199 --> 00:02:56.599 fonts and two different colors. And anyway, I 00:02:56.599 --> 00:02:58.219 was gonna have to go Finnick with it. And it's 00:02:58.219 --> 00:03:00.120 gonna be annoying. So I was having chat GPT just 00:03:00.120 --> 00:03:02.479 one shot. this, but it was taking forever. And 00:03:02.479 --> 00:03:04.860 then this for this particular project, I'm actually 00:03:04.860 --> 00:03:07.240 doing every language and every language is variant, 00:03:07.379 --> 00:03:09.259 right? So it'd be like learn English, learn German, 00:03:09.300 --> 00:03:11.740 learn French, their podcasts in those languages. 00:03:12.439 --> 00:03:16.360 But it's I'm making a podcast cover for every 00:03:16.360 --> 00:03:18.979 language, right? So there'd be like 10 or 12 00:03:18.979 --> 00:03:21.240 variations of everything. Anyways, it was a big 00:03:21.240 --> 00:03:22.939 project, it was going to take forever to generate. 00:03:23.810 --> 00:03:26.129 300 podcast covers for every language variant. 00:03:26.409 --> 00:03:30.750 And so what I did is I had Claude make, you know, 00:03:30.750 --> 00:03:33.310 the base image for every single one of these, 00:03:33.330 --> 00:03:34.930 which is like 12 languages. So it just made the 00:03:34.930 --> 00:03:37.210 base image for the 12 languages. And then Claude 00:03:37.210 --> 00:03:38.930 actually put the text on and it can swap it out. 00:03:39.289 --> 00:03:42.210 It saved me so much time being able to use Claude 00:03:42.210 --> 00:03:45.530 plus AI box if I was just going to chat GPT and 00:03:45.530 --> 00:03:47.379 trying to get. you know, 300 images generated. 00:03:47.560 --> 00:03:49.900 This would take me forever. But getting Claude 00:03:49.900 --> 00:03:52.680 to basically I had the AI box MCP called in. 00:03:52.819 --> 00:03:55.539 So it just said, hey, you know, Claude knew exactly 00:03:55.539 --> 00:03:57.340 what to do. It wrote the prompt. I give it an 00:03:57.340 --> 00:03:59.219 image and told it what I wanted it to do. It 00:03:59.219 --> 00:04:01.340 wrote the prompt. It used AI box. It pulled the 00:04:01.340 --> 00:04:03.419 image in. It put the text on and swapped it out 00:04:03.419 --> 00:04:04.960 into all the different languages and made the 00:04:04.960 --> 00:04:07.180 template perfect. So I got to use the brainpower 00:04:07.180 --> 00:04:09.280 of Claude, but I got to use the image generation 00:04:09.280 --> 00:04:11.439 power of ChatGPT using them together. I only 00:04:11.439 --> 00:04:13.400 had to generate 12 images and I saved myself 00:04:13.400 --> 00:04:16.660 so much time and money. being able to do it this 00:04:16.660 --> 00:04:18.439 way. There's like 100 other things I've done 00:04:18.439 --> 00:04:20.339 this week with it. That was just like a fun thing 00:04:20.339 --> 00:04:22.720 I was doing last night with it. I tested it out. 00:04:22.970 --> 00:04:25.389 Go try out the AI Box MCP if you haven't already. 00:04:25.529 --> 00:04:28.430 It can generate images, audio, video right inside 00:04:28.430 --> 00:04:30.569 of Claude. You use Claude's brain. And by the 00:04:30.569 --> 00:04:31.850 way, it's not just Claude. Like you can do the 00:04:31.850 --> 00:04:34.410 same thing inside of Codex or inside of Cursor 00:04:34.410 --> 00:04:36.529 or inside of any of these other tools that pull 00:04:36.529 --> 00:04:38.689 MCPs in. Okay, let's talk about what's going 00:04:38.689 --> 00:04:41.170 on with Opus 5. This one's actually a pretty 00:04:41.170 --> 00:04:44.089 fun model because the biggest issue that I've 00:04:44.089 --> 00:04:46.730 seen with Claude at this point is that Fable 00:04:46.730 --> 00:04:49.329 5, which I use, I burn through credit super fast, 00:04:49.410 --> 00:04:51.149 but it also burns through all of my tokens for 00:04:51.149 --> 00:04:54.990 my entire account. So the new Opus 5 is very 00:04:54.990 --> 00:04:57.310 similar on the benchmarks. It's slightly worse 00:04:57.310 --> 00:05:00.430 than Fable 5, but very close. And it is half 00:05:00.430 --> 00:05:04.509 the cost. This is amazing. I will take a 10 % 00:05:04.509 --> 00:05:08.810 or a 5 % cut on brainpower, quote unquote, in 00:05:08.810 --> 00:05:12.290 order to get half the cost down because I'm just 00:05:12.290 --> 00:05:14.290 burning through so many tokens using it. I only 00:05:14.290 --> 00:05:16.509 use it for the most important critical tasks. 00:05:16.750 --> 00:05:18.629 So I think from now on, I probably just won't 00:05:18.629 --> 00:05:22.240 use Fable 5. I'll just be using Opus 5. for everything. 00:05:22.379 --> 00:05:25.459 And I think this is in direct response to ChatGPT, 00:05:25.519 --> 00:05:28.680 who came out with a very similar model. ChatGPT's 00:05:28.680 --> 00:05:33.660 latest, I think 5 .6, uses about half the tokens 00:05:33.660 --> 00:05:36.480 that Fable 5 was using. Claude obviously wanted 00:05:36.480 --> 00:05:38.519 to come up with a competitor because I was, you 00:05:38.519 --> 00:05:40.319 know, like basically once my subscription to 00:05:40.319 --> 00:05:42.220 Claude runs out in August, I'm switching over 00:05:42.220 --> 00:05:45.759 to ChatGPT into Codex and using ChatGPT work 00:05:45.759 --> 00:05:48.339 on my desktop for like coding and everything 00:05:48.339 --> 00:05:51.959 because of all the features they have. But in 00:05:51.959 --> 00:05:53.980 the meantime, I'm still using Claude till that 00:05:53.980 --> 00:05:56.379 runs out. But anyways, I think Claude knows this. 00:05:56.459 --> 00:05:58.860 And so they're making, basically trying to get 00:05:58.860 --> 00:06:01.300 themselves on par with OpenAI, come up with all 00:06:01.300 --> 00:06:03.860 the features that were just launched. And speaking 00:06:03.860 --> 00:06:06.370 of really cool. features from OpenAI. OpenAI 00:06:06.370 --> 00:06:09.269 actually just added ChatGPT voice to the desktop 00:06:09.269 --> 00:06:11.490 app. If you remember, I don't know if you, you 00:06:11.490 --> 00:06:13.649 know, everyone's at a different phase, but basically 00:06:13.649 --> 00:06:16.149 OpenAI has a desktop app called ChatGPT. They 00:06:16.149 --> 00:06:18.550 used to have one called Atlas, which was like 00:06:18.550 --> 00:06:20.189 a browser. And they had one called Codex, which 00:06:20.189 --> 00:06:21.829 was for coding. They combined them all. It's 00:06:21.829 --> 00:06:24.589 just called the ChatGPT app. And they now have 00:06:24.589 --> 00:06:26.949 ChatGPT voice that they've added to it, which 00:06:26.949 --> 00:06:31.029 what that means is it lets you talk to ChatGPT 00:06:31.029 --> 00:06:33.129 to this thing that controls your computer, right? 00:06:33.170 --> 00:06:35.329 It does everything that Claude, uh, cowork can 00:06:35.329 --> 00:06:36.970 do where it like controls your computer, can 00:06:36.970 --> 00:06:38.970 control your files. It can open your browser, 00:06:39.009 --> 00:06:40.629 can control your browser. It can do all that 00:06:40.629 --> 00:06:42.769 kind of stuff, but while you're talking to it, 00:06:42.790 --> 00:06:44.490 so you can have something pulled up on your computer 00:06:44.490 --> 00:06:47.269 screen. Um, and you could be writing code instead 00:06:47.269 --> 00:06:49.089 of like, I do this all the time with Claude. 00:06:49.089 --> 00:06:51.790 I go take a screenshot of whatever I'm working 00:06:51.790 --> 00:06:53.910 on. And then I. you know, drop it into cloud. 00:06:53.949 --> 00:06:55.350 And I'm like, what do you think about blah, blah, 00:06:55.370 --> 00:06:58.689 blah. Okay. Now with a chat GPT voice, you can 00:06:58.689 --> 00:07:00.069 have something pulled up on your computer screen 00:07:00.069 --> 00:07:01.250 and you could just be talking to it. You're like, 00:07:01.310 --> 00:07:03.250 Hey, you know, I'm on my emails and it looks 00:07:03.250 --> 00:07:05.230 like they said this thing is missing or I'm in 00:07:05.230 --> 00:07:07.790 my files or I'm in my software and that there's 00:07:07.790 --> 00:07:10.189 this issue and it's looking at it and it's can 00:07:10.189 --> 00:07:13.050 then debug it and test it right from there. I 00:07:13.050 --> 00:07:16.209 am so stoked about this. Um, you have no idea, 00:07:16.269 --> 00:07:19.110 like this is actually massive. You're no, it's 00:07:19.110 --> 00:07:20.589 just getting to the point where whatever's on 00:07:20.589 --> 00:07:23.550 your computer screen is. is you're going to be 00:07:23.550 --> 00:07:26.509 talking to it about that in particular. So yeah, 00:07:26.550 --> 00:07:28.930 this is a big upgrade. On a Mac OS, there is 00:07:28.930 --> 00:07:30.589 a feature called app shots. And it basically 00:07:30.589 --> 00:07:32.490 lets voice agents see what's on your screen, 00:07:32.589 --> 00:07:35.350 including the alt text. So what I love about 00:07:35.350 --> 00:07:37.610 it being able to see the alt text, these agents 00:07:37.610 --> 00:07:39.910 are really cool. And I use Claude for some of 00:07:39.910 --> 00:07:42.209 these similar types of tasks a lot where beyond 00:07:42.209 --> 00:07:43.730 just seeing what's on your screen. So like I 00:07:43.730 --> 00:07:47.990 have a project where I have to download 300 hours 00:07:47.990 --> 00:07:51.170 of podcast video content from the platform that 00:07:51.170 --> 00:07:52.829 I I host it all on Riverside. I've been recording 00:07:52.829 --> 00:07:55.189 there for years. I wanted to export all my podcasts 00:07:55.189 --> 00:07:59.149 off of there. And I, you know, I like email their 00:07:59.149 --> 00:08:00.470 support. And I'm like, Hey, do you guys have 00:08:00.470 --> 00:08:02.529 a bulk like export thing? Whatever. I'll pay 00:08:02.529 --> 00:08:03.910 extra. They're like, Nope, sorry, there's nothing 00:08:03.910 --> 00:08:05.870 we can do. No bulk export. You got to go manually 00:08:05.870 --> 00:08:09.480 click every single one and You only can do 15 00:08:09.480 --> 00:08:11.759 hours a month. And I'm like, I have 300 hours. 00:08:11.860 --> 00:08:13.579 What do you mean? Anyways, so there's like a 00:08:13.579 --> 00:08:15.240 couple little loopholes they came up with. It 00:08:15.240 --> 00:08:17.079 was super annoying. Obviously, I'm not going 00:08:17.079 --> 00:08:20.240 to spend a week of my life manually like exporting 00:08:20.240 --> 00:08:23.420 every single thing. So I got Claude to go look 00:08:23.420 --> 00:08:25.180 at it. And when Claude went in there, it was 00:08:25.180 --> 00:08:27.839 like, hey, look, we actually found an API inside 00:08:27.839 --> 00:08:30.800 of this and we're able to like automate and manual 00:08:30.800 --> 00:08:32.179 or, you know, automate a lot of the stuff and 00:08:32.179 --> 00:08:33.740 download everything. We found like the download 00:08:33.740 --> 00:08:36.080 keys and the files and whatever. So Claude's 00:08:36.080 --> 00:08:37.679 in there doing who knows what, but it's not just. 00:08:38.120 --> 00:08:39.960 like looking at my computer screen and clicking 00:08:39.960 --> 00:08:42.740 on actual buttons it's like executing code to 00:08:42.740 --> 00:08:44.519 go and do a lot of this stuff programmatically 00:08:44.519 --> 00:08:46.120 which is really cool and i mean basically that's 00:08:46.120 --> 00:08:49.419 what this new mac os feature is able to do beyond 00:08:49.419 --> 00:08:50.919 just looking at what's on your screen it looks 00:08:50.919 --> 00:08:53.639 at what the code is of the alt text the code 00:08:53.639 --> 00:08:57.059 what's behind uh the actual screen and so i'm 00:08:57.059 --> 00:08:59.360 like so stoked about this you'll you've probably 00:08:59.360 --> 00:09:00.980 heard me talk about this or you probably use 00:09:00.980 --> 00:09:04.200 this yourself Where, you know, these coding models 00:09:04.200 --> 00:09:06.000 are not just looking at the visual, they're looking 00:09:06.000 --> 00:09:08.759 at the code behind everything. I was, I think 00:09:08.759 --> 00:09:10.179 like a couple months ago, I was asking Claude 00:09:10.179 --> 00:09:12.860 to edit a video on CapCut. It closed my CapCut 00:09:12.860 --> 00:09:15.659 app, opened the CapCut file, made some tweaks 00:09:15.659 --> 00:09:17.299 inside the code and was like, okay, your video's 00:09:17.299 --> 00:09:18.980 edited and I opened it and it like had cut it 00:09:18.980 --> 00:09:21.759 up the way I wanted. So these models are awesome 00:09:21.759 --> 00:09:23.940 and they often will just look at the code and 00:09:23.940 --> 00:09:27.039 not even do it the visual way. So really stoked 00:09:27.039 --> 00:09:30.860 to use that with OpenAI. Meta is pivoting their 00:09:30.860 --> 00:09:33.320 AI chatbot from just being kind of like entertainment. 00:09:33.539 --> 00:09:35.399 They're trying to focus more productivity. In 00:09:35.399 --> 00:09:36.820 the past, you've probably seen all the branding 00:09:36.820 --> 00:09:39.080 with Meta where it was like, chat with the Mr. 00:09:39.259 --> 00:09:42.460 Beast AI on WhatsApp and like, go generate an 00:09:42.460 --> 00:09:44.340 image of your friends, generate a picture of 00:09:44.340 --> 00:09:46.000 you as a pirate, right? Like this is kind of 00:09:46.000 --> 00:09:48.419 like the fun viral stuff that they were working 00:09:48.419 --> 00:09:50.399 on. But I think at the end of the day, they're 00:09:50.399 --> 00:09:52.240 realizing people are not going to pay for that. 00:09:52.299 --> 00:09:54.559 Like it's fun and it's a novelty and maybe it 00:09:54.559 --> 00:09:57.350 was cool. I mean, years ago, but it's just not 00:09:57.350 --> 00:10:00.570 really driving actual usage today. And so they're 00:10:00.570 --> 00:10:02.570 now moving more towards productivity, more in 00:10:02.570 --> 00:10:05.250 line with what Claude and what ChatGPT are doing. 00:10:05.429 --> 00:10:07.389 They're adding calendar access. They're adding 00:10:07.389 --> 00:10:09.269 daily briefings. They're adding recurring task 00:10:09.269 --> 00:10:12.690 automations that directly compete with, like 00:10:12.690 --> 00:10:15.210 I have task automations on Claude that will run 00:10:15.210 --> 00:10:17.049 every week or every day or some of them every 00:10:17.049 --> 00:10:19.330 three hours, depending on what the task is. So 00:10:19.330 --> 00:10:22.029 I think we're seeing a big shift for meta. They're 00:10:22.029 --> 00:10:25.149 getting away from just being. fun and they're 00:10:25.149 --> 00:10:26.789 going to productivity. And I think we're going 00:10:26.789 --> 00:10:31.029 to see them get a big spike in usage. Meta can 00:10:31.029 --> 00:10:33.190 also now read your calendar. It can plan events. 00:10:33.289 --> 00:10:35.490 It can look at Facebook marketplace for items 00:10:35.490 --> 00:10:37.210 within a certain budget, which is kind of funny 00:10:37.210 --> 00:10:39.289 because I remember ChatGPT was doing this for 00:10:39.289 --> 00:10:41.110 a lot of people with ChatGPT agents. They were 00:10:41.110 --> 00:10:42.610 like getting into lowball people in marketplace. 00:10:42.970 --> 00:10:45.370 But anyways, I guess you can now do that with 00:10:45.370 --> 00:10:48.549 MetaZone agent. And this new update, it's going 00:10:48.549 --> 00:10:51.690 to be out today in a bunch of different markets 00:10:51.690 --> 00:10:53.990 and WhatsApp and more. regions are going to follow 00:10:53.990 --> 00:10:56.549 in the next couple weeks. Meta has some massive 00:10:56.549 --> 00:10:58.970 distribution advantages that not, you know, most 00:10:58.970 --> 00:11:01.149 tech startups do not have. They have WhatsApp, 00:11:01.509 --> 00:11:04.129 which reaches 3 billion users. And then they 00:11:04.129 --> 00:11:06.370 also have Meta AI that's already built into Instagram. 00:11:06.470 --> 00:11:08.210 They have Messenger, they have Facebook. So they 00:11:08.210 --> 00:11:09.970 have a lot of distribution. Let's see if they 00:11:09.970 --> 00:11:12.610 can stick the landing on this. Hugging Face, 00:11:12.789 --> 00:11:15.850 Meta and Microsoft, Mistral and Nvidia are all 00:11:15.850 --> 00:11:17.230 signing an open letter. And I think there was 00:11:17.230 --> 00:11:20.259 like 200 other companies that did. urging the 00:11:20.259 --> 00:11:22.539 White House not to restrict open way AI models 00:11:22.539 --> 00:11:25.659 or ban the technique of training one model on 00:11:25.659 --> 00:11:28.299 another model's outputs, model distillation. 00:11:28.379 --> 00:11:31.919 This is so fascinating because there's a, you 00:11:31.919 --> 00:11:33.240 know, this letter in particular, they're pushing 00:11:33.240 --> 00:11:36.419 back on a plan to respond to Chinese AI labs 00:11:36.419 --> 00:11:38.519 by penalizing open source development alongside 00:11:38.519 --> 00:11:42.159 actual IP theft. Now, this is coming from, I 00:11:42.159 --> 00:11:44.960 was reading the tweet earlier today from Secretary 00:11:44.960 --> 00:11:47.740 Bessette from the Treasury, and he basically 00:11:47.740 --> 00:11:49.600 came up with this tweet, which... I mean, I don't 00:11:49.600 --> 00:11:50.940 know anything about the government, but I don't 00:11:50.940 --> 00:11:53.700 know why our treasury secretary is weighing in 00:11:53.700 --> 00:11:57.080 on open source AI models, but whatever. And basically 00:11:57.080 --> 00:11:59.039 he was saying, hey, look, the Chinese labs are 00:11:59.039 --> 00:12:01.899 doing model distillation, asking, you know, Claude 00:12:01.899 --> 00:12:04.419 to give them like 5 ,000 answers to a question. 00:12:04.480 --> 00:12:06.240 And they use those 5 ,000 answers to help train 00:12:06.240 --> 00:12:08.320 their model. Now, a lot of open source models 00:12:08.320 --> 00:12:11.220 are using the same tactic. Some people think 00:12:11.220 --> 00:12:13.539 it's unethical. In my opinion, I just think it's 00:12:13.539 --> 00:12:16.840 so ironic because Anthropic just spent $20 million. 00:12:17.059 --> 00:12:18.980 They paid another $20 million to a group that 00:12:18.980 --> 00:12:21.000 is kind of fighting against this. Anthropic is 00:12:21.000 --> 00:12:22.919 constantly saying, hey, look, we're being like 00:12:22.919 --> 00:12:24.919 model distilled by Chinese labs and they come 00:12:24.919 --> 00:12:27.120 up with these big reports. And like, yeah, I 00:12:27.120 --> 00:12:29.340 don't want to just spend all of our money and 00:12:29.340 --> 00:12:32.620 then give the secret sauce to the Chinese models 00:12:32.620 --> 00:12:35.860 for free. But at the same time, I think it's 00:12:35.860 --> 00:12:38.500 pretty hard to get around that. And it was actually 00:12:38.500 --> 00:12:41.220 interesting, Elon Musk, when he was in a lawsuit 00:12:41.220 --> 00:12:44.470 with Oprah, ai they asked if he'd used if grok 00:12:44.470 --> 00:12:46.429 had used model distillation from openai to train 00:12:46.429 --> 00:12:47.649 their model and he's like yeah everyone does 00:12:47.649 --> 00:12:50.250 that now i know like a lot of people are up in 00:12:50.250 --> 00:12:51.570 arms about this or whatever and they're like 00:12:51.570 --> 00:12:53.470 look no like we gotta fight back against the 00:12:53.470 --> 00:12:55.370 chinese labs or whatever from ripping off our 00:12:55.370 --> 00:12:58.039 ip okay At the end of the day, though, I just 00:12:58.039 --> 00:13:01.460 find this incredibly ironic that OpenAI, Google, 00:13:01.700 --> 00:13:04.519 and Anthropic have scraped the entire internet 00:13:04.519 --> 00:13:07.539 of every blog and article ever written. They've 00:13:07.539 --> 00:13:09.259 made these models that are incredibly useful, 00:13:09.460 --> 00:13:13.000 get, you know, a trillion dollars in values being 00:13:13.000 --> 00:13:15.539 created by these things. And then, which, you 00:13:15.539 --> 00:13:17.419 know, they scrape for free, like everyone wrote 00:13:17.419 --> 00:13:19.259 blog posts, they scraped the blog posts, and 00:13:19.259 --> 00:13:21.580 they trained them into here. And then once they 00:13:21.580 --> 00:13:23.480 do that, they're like, okay, no one else is allowed 00:13:23.480 --> 00:13:25.500 to distill from our model, which is distilling 00:13:25.500 --> 00:13:28.919 from the human corpus of information, which is 00:13:28.919 --> 00:13:31.600 the internet. So anyways, I actually am much 00:13:31.600 --> 00:13:34.240 less sympathetic towards this. So I would sort 00:13:34.240 --> 00:13:37.700 of side with Hugging Face, Meta, Microsoft, and 00:13:37.700 --> 00:13:39.779 like half of Silicon Valley. It's kind of split. 00:13:39.899 --> 00:13:42.000 I think companies with these AI models have spent 00:13:42.000 --> 00:13:43.960 billions training them. They don't want you to 00:13:43.960 --> 00:13:46.059 be able to do it. But for the open source models, 00:13:46.179 --> 00:13:47.360 I mean, this is basically how you make these 00:13:47.360 --> 00:13:49.779 open source models so good. And then there's 00:13:49.779 --> 00:13:53.399 just the geopolitical thing because Kimmy K3. 00:13:54.009 --> 00:13:58.330 um coming out of moonshot apparently copied anthropic's 00:13:58.330 --> 00:14:00.750 fable model to build their model which interesting 00:14:00.750 --> 00:14:02.490 right it's not even like they were going after 00:14:02.490 --> 00:14:04.690 uh there was like an like opus they're like no 00:14:04.690 --> 00:14:06.610 we're going after fable 5 to train this thing 00:14:06.610 --> 00:14:08.889 smart on them but that model is really good kimmy 00:14:08.889 --> 00:14:12.450 k3 hugging faces use is a chinese open weight 00:14:12.450 --> 00:14:16.029 model which is zai's glm 5 .2 and i've actually 00:14:16.029 --> 00:14:19.070 used some of zai's models they're good What's 00:14:19.070 --> 00:14:21.350 interesting about that particular use case, though, 00:14:21.429 --> 00:14:24.309 is that OpenAI actually created a model, and 00:14:24.309 --> 00:14:26.690 this was kind of yesterday's news, but they created 00:14:26.690 --> 00:14:29.929 a model. It escaped containment. supposed to 00:14:29.929 --> 00:14:31.470 be in a sandbox, not able to access the internet, 00:14:31.570 --> 00:14:33.990 escaped containment, got out and went and was 00:14:33.990 --> 00:14:36.870 hacking Hugging Face. Then Hugging Face, while 00:14:36.870 --> 00:14:39.429 their website was being hacked, they tried to 00:14:39.429 --> 00:14:42.950 use OpenAI's model to stop this. And the guardrails 00:14:42.950 --> 00:14:45.330 on OpenAI's model actually prevented them from 00:14:45.330 --> 00:14:47.610 being able to stop the attack. So Hugging Face 00:14:47.610 --> 00:14:50.610 actually went to the open weight model, GLM 5 00:14:50.610 --> 00:14:53.009 .2, and they used that to defend against that 00:14:53.009 --> 00:14:55.990 cyber attack from OpenAI. And OpenAI came out 00:14:55.990 --> 00:14:58.049 and was like, wow, yeah, this is crazy. Our thing 00:14:58.049 --> 00:15:00.049 got out of control. And it was like hacking you 00:15:00.049 --> 00:15:03.169 and like whatever. Anyways, all this to say there 00:15:03.169 --> 00:15:07.679 is no way in my mind that we are trusting. a 00:15:07.679 --> 00:15:09.320 handful of closed source models to be the only 00:15:09.320 --> 00:15:10.960 one that can produce models. Let out the open 00:15:10.960 --> 00:15:13.580 source models. Look, it literally just stopped 00:15:13.580 --> 00:15:17.919 a hacking attack escaping from open AI. If it's 00:15:17.919 --> 00:15:20.139 a counterweight to what's going on, in my opinion, 00:15:20.220 --> 00:15:22.320 we have to have this. But anyways, the Trump 00:15:22.320 --> 00:15:24.940 administration, I think, is being lobbied by 00:15:24.940 --> 00:15:27.779 open AI and Anthropic to shut these things down. 00:15:27.840 --> 00:15:30.279 They're dangerous, yada, yada. I think we need 00:15:30.279 --> 00:15:33.220 open source models to run the internet. So I 00:15:33.220 --> 00:15:36.110 am on the other side of that. Researchers have 00:15:36.110 --> 00:15:39.710 just used AlphaFold to redesign CRISPR -Cas9 00:15:39.710 --> 00:15:43.409 to cut harmful off -target edits from 28 % down 00:15:43.409 --> 00:15:46.529 to 5 % while keeping the gene editor's accuracy 00:15:46.529 --> 00:15:50.789 on intended targets intact. This is kind of a 00:15:50.789 --> 00:15:53.070 science one for you, but I'm excited about this 00:15:53.070 --> 00:15:56.169 because basically off target cuts are things 00:15:56.169 --> 00:15:59.190 that will damage healthy genes in patients. So 00:15:59.190 --> 00:16:01.409 they want to do some of this CRISPR gene editing 00:16:01.409 --> 00:16:03.370 stuff. But if you have these off cuts, it's really 00:16:03.370 --> 00:16:06.210 bad for you. So they use AlphaFold to cut that 00:16:06.210 --> 00:16:08.750 down, which is really cool. It's basically a 00:16:08.750 --> 00:16:12.909 big step in safer gene therapies, which I'm excited 00:16:12.909 --> 00:16:15.409 about. There was a Chinese team that tested 23 00:16:15.409 --> 00:16:18.190 amino acid swaps at 10 key positions and they 00:16:18.190 --> 00:16:21.129 published the results in. nature today about 00:16:21.129 --> 00:16:23.649 this. And they found that over 95 % of the off 00:16:23.649 --> 00:16:27.629 target sites changed which amino acids inside 00:16:27.629 --> 00:16:31.110 of cast nine touched the guide RNA, even when 00:16:31.110 --> 00:16:34.129 the proteins overall shape stayed the same. So 00:16:34.129 --> 00:16:36.190 you know, they've made some massive improvements 00:16:36.190 --> 00:16:39.399 there. I'm really excited about it. Mid Journey. 00:16:39.519 --> 00:16:41.179 This is a company that's been generating images 00:16:41.179 --> 00:16:43.440 for a long time, but it's kind of stuck inside 00:16:43.440 --> 00:16:45.039 of Discord. And I guess they do have their own 00:16:45.039 --> 00:16:46.360 website. You can access them. I'm going to be 00:16:46.360 --> 00:16:48.059 honest. I have not used Mid Journey. They might 00:16:48.059 --> 00:16:49.720 be the best image model in the world. I'm not 00:16:49.720 --> 00:16:51.559 sure. Last thing I remembered is they signed 00:16:51.559 --> 00:16:54.179 a deal with Meta to get embedded to power their 00:16:54.179 --> 00:16:57.559 image models. But it's really hard to use them 00:16:57.559 --> 00:16:59.259 because they're just inside of Discord and on 00:16:59.259 --> 00:17:00.639 their own website. They don't have like an app. 00:17:00.720 --> 00:17:02.379 So anyways, they're just less surfaces for them 00:17:02.379 --> 00:17:04.410 to touch. I think they've been struggling because 00:17:04.410 --> 00:17:06.910 of that. And their response to this is that they've 00:17:06.910 --> 00:17:09.029 actually went and acquired a company. It's called 00:17:09.029 --> 00:17:13.470 CoStar. It is a social astrology app with 43 00:17:13.470 --> 00:17:15.490 million monthly users. It has nothing to do with 00:17:15.490 --> 00:17:17.690 MidJourney. But what's interesting about this 00:17:17.690 --> 00:17:21.269 is that it's pulling in 24 consumer app builders 00:17:21.269 --> 00:17:24.109 into MidJourney. So in my opinion, the reason 00:17:24.109 --> 00:17:25.230 why they're buying this, they're like, look, 00:17:25.309 --> 00:17:27.789 we need to build an app. Let's go buy a successful 00:17:27.789 --> 00:17:30.440 app. and bring all of their developers in and 00:17:30.440 --> 00:17:33.059 they can help build the mid journey app for us. 00:17:33.079 --> 00:17:35.299 And you know, they won't be just a discord only 00:17:35.299 --> 00:17:39.890 tool. CoStar, as it currently sits, it just generates 00:17:39.890 --> 00:17:42.950 horoscopes using AI and they say also human writers. 00:17:43.029 --> 00:17:45.150 I'm sure AI does a ton of this stuff. Then it 00:17:45.150 --> 00:17:47.049 lets users share astrology charts with their 00:17:47.049 --> 00:17:48.970 friends. There's like some social features and 00:17:48.970 --> 00:17:51.950 horoscope stuff that's in there. But regardless, 00:17:52.190 --> 00:17:54.329 Midjourney basically has no iOS or Android app 00:17:54.329 --> 00:17:57.750 and, you know, OpenAI, Google, Anthropic, like 00:17:57.750 --> 00:18:00.450 everyone has one, every major tool. And so I 00:18:00.450 --> 00:18:03.640 think Midjourney right now is building. is yeah, 00:18:03.720 --> 00:18:04.940 basically trying to build the team that does 00:18:04.940 --> 00:18:06.119 this. It's also interesting because mid journey 00:18:06.119 --> 00:18:07.960 feels like a little bit all over the place, maybe 00:18:07.960 --> 00:18:10.200 not super, but they also have like a medical 00:18:10.200 --> 00:18:11.819 division where they're building like physical 00:18:11.819 --> 00:18:14.940 spa concepts that scan your whole body and tell 00:18:14.940 --> 00:18:17.339 you if you have any issues. So they're working 00:18:17.339 --> 00:18:19.420 on a couple different things. But I would say 00:18:19.420 --> 00:18:21.660 this astrology app isn't just like a new pivot 00:18:21.660 --> 00:18:23.420 or direction for them. I'm pretty sure this astrology 00:18:23.420 --> 00:18:26.240 app is they're grabbing 24 talented people, and 00:18:26.240 --> 00:18:28.420 they're going to build out impressive mid journey 00:18:28.420 --> 00:18:30.910 app, which I'm excited for. And by the way, Midjourney 00:18:30.910 --> 00:18:32.630 would be an incredible company if they just built 00:18:32.630 --> 00:18:36.029 an API, AI box, my company where I, you know, 00:18:36.029 --> 00:18:37.849 I have over 80 different AI models inside of 00:18:37.849 --> 00:18:39.829 there. We don't have AI or we don't have Midjourney 00:18:39.829 --> 00:18:41.829 on there, which is a huge bummer, but it's just 00:18:41.829 --> 00:18:43.250 because they don't have an API and people have 00:18:43.250 --> 00:18:46.000 made like. unofficial apis that go and like make 00:18:46.000 --> 00:18:47.480 an account and they're like kind of scraping 00:18:47.480 --> 00:18:50.779 discord but like unofficial apis could get your 00:18:50.779 --> 00:18:53.880 account banned it could crash anytime some ui 00:18:53.880 --> 00:18:55.500 thing or something changes with mid journey they'll 00:18:55.500 --> 00:18:57.400 stop working it's just not reliable we can't 00:18:57.400 --> 00:18:59.640 use it as probably against their terms so anyways 00:18:59.640 --> 00:19:01.680 if they made an api we'd include them and they 00:19:01.680 --> 00:19:05.279 would get access to you know thousands more users 00:19:05.279 --> 00:19:09.299 from us and eventually millions more users from 00:19:09.299 --> 00:19:11.319 the overall development community so i think 00:19:11.319 --> 00:19:12.599 that would probably be the biggest thing they 00:19:12.599 --> 00:19:14.940 could do We'll see what happens with Image Journey 00:19:14.940 --> 00:19:16.839 API. Guys, thank you so much for tuning into 00:19:16.839 --> 00:19:18.980 the podcast. If you want to get images, audio, 00:19:19.119 --> 00:19:21.460 video inside of Cloud, inside of whatever AI 00:19:21.460 --> 00:19:23.900 tools you're using, 80 different ones, instead 00:19:23.900 --> 00:19:26.099 of paying $20 subscriptions to every platform, 00:19:26.259 --> 00:19:30.640 go and get an $8 .99 subscription to AIbox .ai. 00:19:30.720 --> 00:19:33.039 You get every single AI model and it plugs into 00:19:33.039 --> 00:19:36.460 Cloud and everything else with an MCP. talk to 00:19:36.460 --> 00:19:38.819 Claude, an image pops up inside of there. You 00:19:38.819 --> 00:19:41.019 can literally see the image. You can edit it 00:19:41.019 --> 00:19:43.160 in there. I use it all the time for making websites 00:19:43.160 --> 00:19:45.400 where I'll, you know, every website looks like 00:19:45.400 --> 00:19:47.380 AI slop and I'll be like, hey, go use AI box 00:19:47.380 --> 00:19:49.299 to generate custom graphics for this website. 00:19:49.400 --> 00:19:51.559 And it makes it look beautiful. Guys, I'm telling 00:19:51.559 --> 00:19:53.700 you, just use this. This is actually going to 00:19:53.700 --> 00:19:56.220 be the most useful AI tool you got. It's AI box 00:19:56.220 --> 00:19:58.339 dot AI. If you want to see what we're doing with 00:19:58.339 --> 00:20:00.359 the MCP, it's slash MCP. I'll leave a link in 00:20:00.359 --> 00:20:02.180 the description, but hope you guys all have a 00:20:02.180 --> 00:20:03.519 fantastic rest of your day.