WEBVTT 00:00:02.239 --> 00:00:04.799 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 --> 00:00:11.320 breaking down what it means for your life and 00:00:11.320 --> 00:00:30.199 business. I think there's some pros and cons 00:00:30.199 --> 00:00:32.020 to this. I actually make AI -generated music, 00:00:32.100 --> 00:00:35.520 and I make about $2 ,500 a month from it. So 00:00:35.520 --> 00:00:37.439 I want to tell you a little bit about what's 00:00:37.439 --> 00:00:40.219 going on with this case. Merendil is signing 00:00:40.219 --> 00:00:44.299 a $100 million Google Cloud deal to train self 00:00:44.299 --> 00:00:48.340 -improving AI. And Google Maps is turning Ask 00:00:48.340 --> 00:00:50.359 Maps, that kind of button that they have in there, 00:00:50.439 --> 00:00:53.100 into an agent that orders food. It can book hotels. 00:00:53.219 --> 00:00:54.880 It's going to be able to do way more than just 00:00:54.880 --> 00:00:58.090 asking Google Maps something with Gemini. AI 00:00:58.090 --> 00:01:00.909 moderation is not always perfect, and recently 00:01:00.909 --> 00:01:05.120 it has hit Reddit. discord and tumblr on false 00:01:05.120 --> 00:01:08.140 positives also a company called naive has just 00:01:08.140 --> 00:01:11.659 raised 28 .5 million dollars to let ai agents 00:01:11.659 --> 00:01:14.379 spin up and run entire companies i wanted to 00:01:14.379 --> 00:01:16.900 do a startup spotlight and they've raised almost 00:01:16.900 --> 00:01:18.719 30 million dollars this is an interesting one 00:01:18.719 --> 00:01:21.640 on the podcast i talk a lot about different models 00:01:21.640 --> 00:01:23.939 that you may not have tried like mistral from 00:01:23.939 --> 00:01:26.780 france cohere from canada deep seek from china 00:01:26.780 --> 00:01:31.040 and Grok and Perplexity and even Meta's AI, which 00:01:31.040 --> 00:01:33.200 just came out with a bunch of new coding features 00:01:33.200 --> 00:01:35.439 and a new coding AI. Now, my question to you 00:01:35.439 --> 00:01:37.980 is, have you tried all of these different AI 00:01:37.980 --> 00:01:40.079 models? Because if you want to, it typically 00:01:40.079 --> 00:01:42.400 is going to cost you $20 to get a subscription 00:01:42.400 --> 00:01:44.500 to all of those different platforms. And I will 00:01:44.500 --> 00:01:46.620 be honest, some of those models are better at 00:01:46.620 --> 00:01:49.140 answering specific types of questions than others. 00:01:49.219 --> 00:01:52.480 For a lot of what I do, I use Anthropic and I 00:01:52.480 --> 00:01:55.260 use OpenAI and I use Google as some of my main 00:01:55.260 --> 00:01:57.370 powerhouses, but I use some of the other models 00:01:57.370 --> 00:01:59.609 for different tasks. DeepSeek, honestly, for 00:01:59.609 --> 00:02:01.390 example, is something that can get things done 00:02:01.390 --> 00:02:03.870 very fast and very cheaply if you run out of 00:02:03.870 --> 00:02:05.989 your token credits. I would love for you to check 00:02:05.989 --> 00:02:08.189 out all of those different models and try them 00:02:08.189 --> 00:02:10.810 all and compare them all side by side for your 00:02:10.810 --> 00:02:13.469 next project or your next question, just to see 00:02:13.469 --> 00:02:15.509 what the results are. If you want to try that, 00:02:15.569 --> 00:02:18.889 go check out AIbox .ai, my own startup, where 00:02:18.889 --> 00:02:21.330 you can test over 80 different AI models. And 00:02:21.330 --> 00:02:23.110 you can also do a bunch of different image, audio, 00:02:23.250 --> 00:02:25.710 and video models as well. You can compare all 00:02:25.710 --> 00:02:27.919 of the results. side by side and pick what your 00:02:27.919 --> 00:02:30.460 favorite one is. If you're working on a new ad 00:02:30.460 --> 00:02:32.000 or you need to create a bunch of new Facebook 00:02:32.000 --> 00:02:34.500 ads, for example, you can get different video 00:02:34.500 --> 00:02:37.120 models to create the same prompt. And I've tried 00:02:37.120 --> 00:02:39.159 this and been pretty impressed with different 00:02:39.159 --> 00:02:42.620 things like Sora versus Google VO3. And you can 00:02:42.620 --> 00:02:44.900 actually run those different ads side by side 00:02:44.900 --> 00:02:47.099 against each other in Facebook to see which one 00:02:47.099 --> 00:02:49.240 converts the best, but you can also see them 00:02:49.240 --> 00:02:50.979 side by side. So if you want to go check it out, 00:02:51.000 --> 00:02:54.060 it is $8 and 99 cents to get started on AI box. 00:02:54.219 --> 00:02:55.960 It's super cheap and you get access. to over 00:02:55.960 --> 00:02:57.860 80 different AI models. I'm going to leave a 00:02:57.860 --> 00:03:00.770 link in the description to go try it out. Suno 00:03:00.770 --> 00:03:04.110 is now rolling out the new watermarking and they're 00:03:04.110 --> 00:03:06.550 doing audio fingerprints for their AI music generator. 00:03:06.669 --> 00:03:08.310 This is something that I use all the time. I 00:03:08.310 --> 00:03:10.270 love Suno. I've made tons of music there and 00:03:10.270 --> 00:03:13.870 actually make $2 ,500 a month from my music catalog. 00:03:13.990 --> 00:03:17.430 A ton of it is either AI generated or it's just 00:03:17.430 --> 00:03:21.069 AI empowered. So I do a lot of meditation music 00:03:21.069 --> 00:03:23.030 is like my niche. I started it in college. And 00:03:23.030 --> 00:03:25.490 so I'll still drop albums on there every once 00:03:25.490 --> 00:03:28.129 in a while on the weekends. And in order to do 00:03:28.129 --> 00:03:30.530 that, I actually just use Suno. to generate the 00:03:30.530 --> 00:03:32.009 meditation music in the background. And it's 00:03:32.009 --> 00:03:34.550 my voice. I'm talking over top of it. But yeah, 00:03:34.569 --> 00:03:37.509 from that, I still make $2 ,500 to $3 ,000 a 00:03:37.509 --> 00:03:39.610 month, super, super passively. And it's kind 00:03:39.610 --> 00:03:42.870 of a fun side project. But something that is 00:03:42.870 --> 00:03:44.750 happening is that a lot of different lawsuits 00:03:44.750 --> 00:03:47.189 have been leveled against Suno. And to be fair, 00:03:47.250 --> 00:03:49.610 I think they did scrape out a ton of different 00:03:49.610 --> 00:03:52.969 people's music that wasn't licensed properly. 00:03:53.710 --> 00:03:55.789 Now they're getting sued. I think they made a 00:03:55.789 --> 00:03:58.770 settlement with Warm Music Group. And the other 00:03:58.770 --> 00:04:01.330 part of this whole settlement is the goal to 00:04:01.330 --> 00:04:03.909 try to stop low quality AI songs from flooding 00:04:03.909 --> 00:04:06.509 Spotify and Apple Music without disclosure. So 00:04:06.509 --> 00:04:08.189 without people knowing that they're AI generated. 00:04:08.449 --> 00:04:11.129 Deezer apparently has a tool that can detect 00:04:11.129 --> 00:04:14.050 AI generated music and it's removing any AI generated 00:04:14.050 --> 00:04:16.089 music from its platform that doesn't get a certain 00:04:16.089 --> 00:04:19.009 threshold of streams. So if it's a popular song, 00:04:19.050 --> 00:04:21.250 they leave it. But if it's an AI generated song 00:04:21.250 --> 00:04:23.170 that has no streams, a bunch of people are just... 00:04:23.149 --> 00:04:24.970 posting it, they're actually cleaning it off 00:04:24.970 --> 00:04:26.810 the platform, they just don't want it on there. 00:04:27.129 --> 00:04:29.449 So in this deal, and what soon is rolling out 00:04:29.449 --> 00:04:31.870 right now, downloads are soon going to be restricted 00:04:31.870 --> 00:04:34.670 to paying customers, there's going to be a monthly 00:04:34.670 --> 00:04:36.889 cap on how many you can download. So I think 00:04:36.889 --> 00:04:38.970 people were downloading like 1000s in a month. 00:04:39.149 --> 00:04:41.310 And that was causing you know, issues when you 00:04:41.310 --> 00:04:44.290 try to upload 1000s a month to Spotify, it was, 00:04:44.350 --> 00:04:47.490 you know, very spammy. So there's a monthly cap. 00:04:47.589 --> 00:04:49.810 And if I'm being honest, to me, I don't think 00:04:49.810 --> 00:04:51.430 this is gonna this doesn't really bother me, 00:04:51.470 --> 00:04:53.839 I don't make an insane volume of music. And I 00:04:53.839 --> 00:04:56.339 think it tends to be better when you make an 00:04:56.339 --> 00:04:58.339 AI generated song, which I'm not against, by 00:04:58.339 --> 00:05:00.540 the way, if you make AI generated music, I think 00:05:00.540 --> 00:05:03.740 spending a lot of time doing multiple regenerations, 00:05:03.740 --> 00:05:06.139 mixing your different bits of the song together, 00:05:06.279 --> 00:05:08.660 doing covers of something. So like a lot of times, 00:05:08.740 --> 00:05:11.740 I've done stuff where I'll play something, I'll 00:05:11.740 --> 00:05:13.819 upload the file of me playing, maybe I'm not 00:05:13.819 --> 00:05:16.680 incredibly good pianist, but I might play a little 00:05:16.680 --> 00:05:20.920 melody, or I had my son, he made an entire guitar 00:05:20.920 --> 00:05:24.579 album. or he he plucked some little tunes on 00:05:24.579 --> 00:05:27.000 a ukulele and had it turn into like this relaxing 00:05:27.000 --> 00:05:29.160 guitar album that he gets a bunch of streams 00:05:29.160 --> 00:05:31.720 from. So I think that's like a great way to go. 00:05:31.839 --> 00:05:34.300 But you don't need to do crazy volumes for stuff 00:05:34.300 --> 00:05:35.860 like that. So when they're putting this kind 00:05:35.860 --> 00:05:38.120 of monthly download cap under it, I'm not super 00:05:38.120 --> 00:05:40.779 worried about that. That is all in the terms 00:05:40.779 --> 00:05:44.459 of this lawsuit that Suno has settled with Warner. 00:05:44.600 --> 00:05:47.420 This all kind of they made this deal last year 00:05:47.420 --> 00:05:50.910 and it's starting to roll out now. Udio has another 00:05:50.910 --> 00:05:53.529 settlement with Warner Music Group, and they 00:05:53.529 --> 00:05:56.050 are ending downloads entirely, which basically 00:05:56.050 --> 00:05:58.370 makes Suno's approach, keeping downloads but 00:05:58.370 --> 00:06:01.529 gated behind a subscription, a much better outcome 00:06:01.529 --> 00:06:04.069 for its business. With Udio, you can't even download 00:06:04.069 --> 00:06:06.529 your music anymore. So Suno, you can continue 00:06:06.529 --> 00:06:08.750 to use it for generating music. I'm happy that's 00:06:08.750 --> 00:06:11.370 the platform I've invested in the most. And then 00:06:11.370 --> 00:06:13.290 they're both doing watermarking, which basically 00:06:13.290 --> 00:06:16.610 lets distributors like Spotify or Apple track 00:06:16.610 --> 00:06:19.459 which ones are AI. generated when they're actually 00:06:19.459 --> 00:06:21.279 uploaded right now, there's a there's like a 00:06:21.279 --> 00:06:23.600 little personal opt in. So when I'm uploading 00:06:23.600 --> 00:06:26.000 stuff, it's like, is this partially AI generated, 00:06:26.120 --> 00:06:28.279 completely AI generated, or no AI generation 00:06:28.279 --> 00:06:30.220 at all, and you have like a disclosure on there. 00:06:30.319 --> 00:06:32.420 And it's kind of voluntary, but you know, there's 00:06:32.420 --> 00:06:34.540 nothing that enforces that no one's actually 00:06:34.540 --> 00:06:37.000 able to tell. And so this, they're actually going 00:06:37.000 --> 00:06:39.079 to be able to do that. I think streaming platforms 00:06:39.079 --> 00:06:41.860 right now have started to purge AI tracks that 00:06:41.860 --> 00:06:44.560 are just trying to get farmed royalties, there's 00:06:44.560 --> 00:06:46.699 a lot of spam and stuff on there. But to be fair, 00:06:46.740 --> 00:06:50.079 I think you do Spam in, you know, even even before 00:06:50.079 --> 00:06:53.360 all of this came out, the old tactic that the 00:06:53.360 --> 00:06:54.899 spammers would do is they would just go find 00:06:54.899 --> 00:06:58.600 like random obscure YouTube channels that would 00:06:58.600 --> 00:07:00.540 publish good music. Right. Maybe you're like 00:07:00.540 --> 00:07:03.160 not very famous on YouTube. They would rip your 00:07:03.160 --> 00:07:05.120 stuff and they would upload it like like their 00:07:05.120 --> 00:07:06.839 own. It was just on YouTube. It wasn't already. 00:07:06.920 --> 00:07:08.439 They try to find something that wasn't already 00:07:08.439 --> 00:07:10.899 on like music platforms. And then they would, 00:07:10.899 --> 00:07:13.420 you know, you could do there's all sorts of spam 00:07:13.420 --> 00:07:15.620 where they would like make it the same name as 00:07:15.620 --> 00:07:17.480 an artist that actually exists. or they would 00:07:17.480 --> 00:07:19.379 put like keyword stuffing in the title. Like 00:07:19.379 --> 00:07:21.399 there's a whole bunch of stuff that those people 00:07:21.399 --> 00:07:23.300 would do and try to release thousands of songs 00:07:23.300 --> 00:07:27.040 that way. So I don't think like. the spam element 00:07:27.040 --> 00:07:29.560 of it and the AI gets a little bad rep for it. 00:07:29.600 --> 00:07:31.360 But the spammers are going to spam and they've 00:07:31.360 --> 00:07:33.579 been spamming for a long time. Maybe this made 00:07:33.579 --> 00:07:35.600 it a little bit easier. But at the end of the 00:07:35.600 --> 00:07:38.040 day, I think we also need separate like spam 00:07:38.040 --> 00:07:39.939 control tools. And I think the platforms have 00:07:39.939 --> 00:07:41.819 been working on that and are working on that. 00:07:41.899 --> 00:07:45.259 So hopefully that will continue. Merendil is 00:07:45.259 --> 00:07:48.319 a startup that was founded by some. alumni from 00:07:48.319 --> 00:07:50.879 Anthropic. They just signed a $100 million, I 00:07:50.879 --> 00:07:52.139 actually think it's even more than that, but 00:07:52.139 --> 00:07:54.720 about $100 million deal with Google Cloud to 00:07:54.720 --> 00:07:57.980 train self -improving AI systems. This deal they 00:07:57.980 --> 00:08:00.319 put together is about half the startup's entire 00:08:00.319 --> 00:08:03.040 seed funding. So I think they were able to get 00:08:03.040 --> 00:08:05.160 $200 million. They gave $100 million of that 00:08:05.160 --> 00:08:10.240 straight to Google. And it is a huge bet on compute 00:08:10.240 --> 00:08:13.519 hardware to build AI that can iterate, improve 00:08:13.519 --> 00:08:16.600 itself, and automate research. Specifically, 00:08:16.910 --> 00:08:18.269 think they're going to be working on medicine 00:08:18.269 --> 00:08:21.910 and biology. Their software has a bunch of different 00:08:21.910 --> 00:08:24.750 stages of AI self -improvement, and it's going 00:08:24.750 --> 00:08:27.870 to be basically routing those to whatever chip 00:08:27.870 --> 00:08:31.290 works best. So Google's TPUs or NVIDIA's GPUs, 00:08:31.290 --> 00:08:32.669 it's going to let the startup kind of optimize 00:08:32.669 --> 00:08:36.190 the costs and also potentially resell the orchestration 00:08:36.190 --> 00:08:37.850 layer to customers. It's something they're going 00:08:37.850 --> 00:08:39.870 to be able to make money from. The co -founders 00:08:39.870 --> 00:08:43.629 are Benham Neshaber and Harsh Mehta. Both of 00:08:43.629 --> 00:08:45.230 them came from Anthropic, but they're not the 00:08:45.230 --> 00:08:47.700 only ones working. on this, their two main competitors 00:08:47.700 --> 00:08:49.980 are going to be recursive superintelligence and 00:08:49.980 --> 00:08:52.000 recursive intelligence. Both of them are basically 00:08:52.000 --> 00:08:54.259 doing the identical thing, this kind of self 00:08:54.259 --> 00:08:56.720 -improvement approach. They closed their startup 00:08:56.720 --> 00:08:59.899 round in late June at a $1 billion valuation. 00:09:00.320 --> 00:09:04.139 So this $100 million compute deal is very aggressive 00:09:04.139 --> 00:09:06.639 spending for a company that's brand new. They 00:09:06.639 --> 00:09:08.320 just closed a seed round to give half the money 00:09:08.320 --> 00:09:10.879 away. Their bet, I think, is showing that recursive 00:09:10.879 --> 00:09:13.740 self -improvement is they're trying to move this 00:09:13.740 --> 00:09:15.519 from just being this kind of research idea. to 00:09:15.519 --> 00:09:17.940 an actual engineering problem and they think 00:09:17.940 --> 00:09:20.080 that the real moat is controlling the software 00:09:20.080 --> 00:09:23.059 that the orchestration and orchestrating all 00:09:23.059 --> 00:09:25.379 of this between expensive hardware and it's probably 00:09:25.379 --> 00:09:27.620 not compute that's what they're saying so i'm 00:09:27.620 --> 00:09:30.779 curious to see where this goes Google Maps Ask 00:09:30.779 --> 00:09:33.639 Maps Assistant can now order food, book hotels, 00:09:33.799 --> 00:09:36.080 and buy event tickets directly in the app. I 00:09:36.080 --> 00:09:37.960 love this. I think that Google obviously has 00:09:37.960 --> 00:09:40.799 some incredible assets and adding AI in a really 00:09:40.799 --> 00:09:44.240 thoughtful way makes them much more useful. I've 00:09:44.240 --> 00:09:47.059 been excited by how they've added AI into a lot 00:09:47.059 --> 00:09:49.340 of the different platforms and tools. I think 00:09:49.340 --> 00:09:51.779 they can do better in others, but with Maps in 00:09:51.779 --> 00:09:54.720 particular, it's kind of cool because, I mean, 00:09:54.720 --> 00:09:57.559 you have so many different businesses on Google 00:09:57.559 --> 00:10:00.360 Maps, you know, of like having like hotels and 00:10:00.360 --> 00:10:02.840 restaurants and everything, Google Maps really 00:10:02.840 --> 00:10:05.480 could turn into like an Airbnb, for example, 00:10:05.500 --> 00:10:07.679 where you're like, you're searching how to get 00:10:07.679 --> 00:10:10.240 to a certain city. And then all of a sudden, 00:10:10.259 --> 00:10:12.059 it's like, well, here are all of the hotels in 00:10:12.059 --> 00:10:13.960 that city, and you can go book it there directly. 00:10:14.059 --> 00:10:16.120 They're already kind of competing with Expedia, 00:10:16.139 --> 00:10:18.539 or you can compete with Expedia in a way like 00:10:18.539 --> 00:10:21.019 I've, I've searched and found, you know, depending 00:10:21.019 --> 00:10:23.740 on if I'm just in a certain geographic location, 00:10:23.860 --> 00:10:26.120 like what's the closest hotels around me, I've 00:10:26.120 --> 00:10:28.080 seen them on Google before, or sometimes I'll 00:10:28.169 --> 00:10:30.809 Look them up on Expedia and then go look up Google 00:10:30.809 --> 00:10:32.990 for different reviews or different things. So 00:10:32.990 --> 00:10:34.389 I think they have an incredible asset there. 00:10:34.509 --> 00:10:36.649 And now you're actually going to be able to use 00:10:36.649 --> 00:10:39.409 their assistant, their Ask Maps, to go and book 00:10:39.409 --> 00:10:40.789 all of this stuff. And it's not just hotels. 00:10:40.909 --> 00:10:43.669 It's also events. It's also tickets and food 00:10:43.669 --> 00:10:45.490 and all this other stuff. Really, really cool. 00:10:45.629 --> 00:10:48.110 You won't have to leave the actual Google Maps 00:10:48.110 --> 00:10:51.190 website. It also is able to read your Gmail and 00:10:51.190 --> 00:10:53.710 your calendar to personalize suggestions. It 00:10:53.710 --> 00:10:56.350 can recommend restaurants near your hotel or 00:10:56.350 --> 00:10:58.429 it can tell you when. your flight landing time 00:10:58.429 --> 00:10:59.809 is. So when you're actually going to book in 00:10:59.809 --> 00:11:01.789 a hotel, for example, it can put in the correct 00:11:01.789 --> 00:11:04.990 time or give you a good check in time. Food orders 00:11:04.990 --> 00:11:07.210 are all going to run through Square, Toast and 00:11:07.210 --> 00:11:10.190 Uber Eats. So it's not like, you know, I heard 00:11:10.190 --> 00:11:11.590 a lot of people like when they first heard the 00:11:11.590 --> 00:11:12.750 students were like, oh, my gosh, is it going 00:11:12.750 --> 00:11:15.289 to go kill Uber Eats? I think they do a lot of 00:11:15.289 --> 00:11:17.309 partnerships. Now, those partnerships, of course, 00:11:17.330 --> 00:11:19.809 are subject to them just deciding one day they 00:11:19.809 --> 00:11:21.509 don't want to do that particular partnership. 00:11:22.029 --> 00:11:24.190 But I think we can be pretty confident for antitrust 00:11:24.190 --> 00:11:26.330 reasons that probably won't happen. I remember 00:11:26.330 --> 00:11:27.870 when Google Flights came out, a lot of people 00:11:27.870 --> 00:11:30.649 were concerned. And to this day, if you go search 00:11:30.649 --> 00:11:32.690 Google Flights, you're still going to get recommendations 00:11:32.690 --> 00:11:35.090 from a ton of different travel sites below. So 00:11:35.090 --> 00:11:37.330 it didn't completely kill that entire industry. 00:11:38.370 --> 00:11:40.990 Personal intelligence, which is the Gmail and 00:11:40.990 --> 00:11:43.450 calendar feature, is going to ship by default. 00:11:43.549 --> 00:11:45.750 It's going to avoid a lot of the friction that 00:11:45.750 --> 00:11:48.210 apps have about sharing your personal data or, 00:11:48.250 --> 00:11:51.009 you know, going and syncing an MCP or syncing 00:11:51.009 --> 00:11:53.690 some sort of API or connection to Google or Gmail 00:11:53.690 --> 00:11:56.389 or your calendar. It's all built straight in. 00:11:56.509 --> 00:11:59.029 Ask Maps now remembers all of your past conversations. 00:11:59.029 --> 00:12:01.610 So if you ask, you know, what activities did 00:12:01.610 --> 00:12:03.350 you suggest for my Seattle trip? It's going to 00:12:03.350 --> 00:12:05.129 remember whatever you're talking to it about 00:12:05.129 --> 00:12:07.509 in the past on Gemini, and it's going to pull. 00:12:07.690 --> 00:12:10.029 that up. Maps right now, I think is really shifting 00:12:10.029 --> 00:12:12.110 from kind of just a navigation tool that we've 00:12:12.110 --> 00:12:15.149 all used for the last number of years into a 00:12:15.149 --> 00:12:16.990 transaction engine, they're trying to actually 00:12:16.990 --> 00:12:19.509 get you to make transactions on there. And of 00:12:19.509 --> 00:12:21.289 course, they can take a little piece of the pie, 00:12:21.389 --> 00:12:23.509 they can charge for ads, or they can charge transaction 00:12:23.509 --> 00:12:25.750 fees. So this is something that makes a lot of 00:12:25.750 --> 00:12:28.090 sense for them. Right now, social media platforms 00:12:28.090 --> 00:12:32.149 are aggressively using AI to catch spam and other 00:12:32.149 --> 00:12:34.629 things. And they're also getting real results 00:12:34.629 --> 00:12:38.570 caught up in this. So some of this wrongful banning, 00:12:38.570 --> 00:12:40.990 you could say, is getting thousands of innocent 00:12:40.990 --> 00:12:43.970 users. There's also like no way to appeal this 00:12:43.970 --> 00:12:46.669 before your account gets completely deleted. 00:12:46.870 --> 00:12:50.409 Discord alone locked out 8 ,400 accounts over 00:12:50.409 --> 00:12:53.389 the summer because its AI accidentally mistook 00:12:53.389 --> 00:12:57.610 chess boards for illegal content, which is a 00:12:57.610 --> 00:13:02.049 problem. And Reddit's AI tools removed like... 00:13:02.240 --> 00:13:05.759 over 10 year old posts from our ask historians, 00:13:05.980 --> 00:13:08.440 the Reddit thread, our ask historians, because 00:13:08.440 --> 00:13:10.860 it said that they were linked to a historical 00:13:10.860 --> 00:13:14.419 photo site, um, that the system had flagged as 00:13:14.419 --> 00:13:16.000 spam. So it's like, look, you you're linking 00:13:16.000 --> 00:13:18.080 out to this, or I don't even think it was all 00:13:18.080 --> 00:13:19.700 the posts, but sometimes like the comments on 00:13:19.700 --> 00:13:21.980 it, they're all linking out to a bunch of historical 00:13:21.980 --> 00:13:24.759 photo sites. They review that as spam. And then 00:13:24.759 --> 00:13:26.720 they go and delete all of the threads. Some of 00:13:26.720 --> 00:13:27.940 these are 10 years old, right? And they're like, 00:13:28.000 --> 00:13:30.539 Oh no, these are AI generated threads. And of 00:13:30.539 --> 00:13:33.090 course it's old thread. So it's it is not the 00:13:33.090 --> 00:13:36.350 case. discords AI was designed to flag content 00:13:36.350 --> 00:13:38.470 for human review, there was this kind of bug 00:13:38.470 --> 00:13:40.669 in there, I guess, and it let it bypass humans 00:13:40.669 --> 00:13:43.009 entirely. And it just auto banned a ton of users. 00:13:43.190 --> 00:13:45.389 Some of them were locked out for weeks, which 00:13:45.389 --> 00:13:49.029 if it's a tool that you use a lot, this is obviously 00:13:49.029 --> 00:13:51.409 not very fun. And it's actually interesting discord 00:13:51.409 --> 00:13:54.129 is a platform that my company AI box we use kind 00:13:54.129 --> 00:13:56.889 of like instead of slack, we use discord for 00:13:56.889 --> 00:13:59.389 all of our developers. And we have like a community 00:13:59.389 --> 00:14:01.029 over there. And we do a lot of different stuff 00:14:01.029 --> 00:14:03.549 on discord, I can only imagine and like we got 00:14:03.549 --> 00:14:06.169 to pay premium, right to run our company through 00:14:06.169 --> 00:14:08.149 there. So like we pay for it, but I can only 00:14:08.149 --> 00:14:10.830 imagine getting banned and locked out for weeks. 00:14:10.889 --> 00:14:12.769 I mean, I really just switched to another platform. 00:14:12.830 --> 00:14:14.850 So I think this is quite detrimental for some 00:14:14.850 --> 00:14:17.110 of those platforms. Reddit reports that their 00:14:17.110 --> 00:14:20.929 systems now block about 2 million fake votes 00:14:20.929 --> 00:14:24.789 daily and their enforcement is ramping up. They 00:14:24.789 --> 00:14:27.529 say it's on hate content. It's been ramped up 00:14:27.529 --> 00:14:32.370 over 200%. They use LLMs to spot any spam patterns 00:14:32.370 --> 00:14:34.149 that they can find. And of course, I'm all for 00:14:34.149 --> 00:14:36.450 blocking spam. So if they're able to get real 00:14:36.450 --> 00:14:39.909 spam and real AI generated junk cut out, I'm 00:14:39.909 --> 00:14:42.610 thrilled. But I hate the false positives. And 00:14:42.610 --> 00:14:44.330 I think a lot of people are, you know, leaves 00:14:44.330 --> 00:14:47.080 bad taste in their mouth. if that happens naive 00:14:47.080 --> 00:14:50.799 has just raised 28 .5 million dollars to give 00:14:50.799 --> 00:14:53.480 ai agents the power to basically set up and run 00:14:53.480 --> 00:14:56.100 an entire company so they're going to have a 00:14:56.100 --> 00:14:58.779 startup package it'll do business formation it'll 00:14:58.779 --> 00:15:00.539 do payments it'll do accounting it'll do cloud 00:15:00.539 --> 00:15:03.059 infrastructure all that is in a single api that 00:15:03.059 --> 00:15:06.389 their agent can call So developer can basically 00:15:06.389 --> 00:15:09.269 give an AI prompt and then it can have this, 00:15:09.350 --> 00:15:12.110 their tool spin up like an LLC. It can make a 00:15:12.110 --> 00:15:13.549 bank account. It can make emails. It can make 00:15:13.549 --> 00:15:15.610 phone numbers. I think this is really cool. And 00:15:15.610 --> 00:15:17.870 personally, I love this because it takes out 00:15:17.870 --> 00:15:20.269 a lot of the pain and start in making these startups. 00:15:20.309 --> 00:15:22.669 And they've been very successful and popular. 00:15:22.809 --> 00:15:25.070 They have about 30 ,000 developers that signed 00:15:25.070 --> 00:15:27.210 up as customers when in the first few months 00:15:27.210 --> 00:15:30.009 after they launched their annual revenue 10x 00:15:30.009 --> 00:15:33.230 over six months to the low double digit millions 00:15:33.230 --> 00:15:36.259 with just. full time employees, which is absolutely 00:15:36.259 --> 00:15:38.600 phenomenal. They have a bunch of customers that 00:15:38.600 --> 00:15:41.379 are AI agencies, faceless content creators, and 00:15:41.379 --> 00:15:43.740 they have at least one autonomous rental car 00:15:43.740 --> 00:15:46.940 business. An example is a TikTok account uses 00:15:46.940 --> 00:15:49.919 them to generate and post AI videos of dancing 00:15:49.919 --> 00:15:52.279 cats and dogs. The company is building infrastructure 00:15:52.279 --> 00:15:55.379 to cut the cost of running AI agents at scale. 00:15:55.460 --> 00:15:57.840 So they have sandboxes, they have a model router 00:15:57.840 --> 00:15:59.679 that picks the cheapest AI model for whatever 00:15:59.679 --> 00:16:01.820 task you're using. They have a memory layer, 00:16:01.960 --> 00:16:04.379 they have a serverless runtime. So customers 00:16:04.379 --> 00:16:06.399 only pay when agents are actually working. There's 00:16:06.399 --> 00:16:08.240 a lot of really cool features in there. And so 00:16:08.240 --> 00:16:09.940 yeah, no surprise that they're able to raise 00:16:09.940 --> 00:16:12.440 almost $30 million. I think this is an interesting 00:16:12.440 --> 00:16:14.360 company, I definitely am going to keep my eye 00:16:14.360 --> 00:16:16.360 on naive. Guys, thank you so much for tuning 00:16:16.360 --> 00:16:18.559 into the podcast. If you want to try all of the 00:16:18.559 --> 00:16:20.779 different AI models, test them side by side, 00:16:20.799 --> 00:16:22.639 compare the results for whatever your latest 00:16:22.639 --> 00:16:25.940 project is. And also get Claude and any other 00:16:25.940 --> 00:16:28.519 AI model to to integrate with all of those. So 00:16:28.519 --> 00:16:30.440 like you get image, audio and video generation 00:16:30.440 --> 00:16:33.529 inside of Claude with our connection. go try 00:16:33.529 --> 00:16:37.909 AI box .ai. My own startup is $8 .99 to get started. 00:16:38.029 --> 00:16:41.529 We also have annual plans that are 20 % off of 00:16:41.529 --> 00:16:44.149 that and you can scale up your usage. I'll catch 00:16:44.149 --> 00:16:45.590 you guys all in the next episode.