WEBVTT 00:00:02.140 --> 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 --> 00:00:11.300 breaking down what it means for your life and 00:00:11.300 --> 00:00:17.920 business. Anthropix Mythos AI is finding Microsoft 00:00:17.920 --> 00:00:21.019 bugs faster than Microsoft's engineers can patch 00:00:21.019 --> 00:00:24.480 them. Google's SynthID watermark survived 300 00:00:24.480 --> 00:00:27.890 edits. but fragmentation might actually make 00:00:27.890 --> 00:00:29.609 this completely useless. We'll get into that. 00:00:29.890 --> 00:00:32.890 Encore AI just raised $30 million to turn sales 00:00:32.890 --> 00:00:35.170 call transcripts into voice agents. This is an 00:00:35.170 --> 00:00:37.869 interesting pipeline as far as the data goes. 00:00:38.210 --> 00:00:40.450 Writers right now are embracing typos, myself 00:00:40.450 --> 00:00:43.250 included, and all of the idiosyncrasies basically 00:00:43.250 --> 00:00:46.770 to fight the AI writing that we're seeing everywhere 00:00:46.770 --> 00:00:49.630 that has just become, you know, basically a plague 00:00:49.630 --> 00:00:52.820 on the entire internet. Sayer is going to acquire 00:00:52.820 --> 00:00:55.880 Oasis security for a billion dollars to police 00:00:55.880 --> 00:00:58.799 AI agents identities. If you have tasks that 00:00:58.799 --> 00:01:01.579 you do over and over again using AI tools, I'd 00:01:01.579 --> 00:01:03.439 love for you to check out the AI box builder 00:01:03.439 --> 00:01:06.079 platform. This is my own startup at AI box dot 00:01:06.079 --> 00:01:08.480 AI, and it allows you to link together multiple 00:01:08.480 --> 00:01:11.719 AI models and put prompts in and automate entire 00:01:11.719 --> 00:01:14.159 processes of repetitive tasks. So you don't have 00:01:14.159 --> 00:01:16.260 to do them and, you know, put in your prompts 00:01:16.260 --> 00:01:18.500 and copy and paste between different documents 00:01:18.500 --> 00:01:20.799 or different AI models over and over again. Cool 00:01:20.799 --> 00:01:23.000 thing is that all the different AI models, we 00:01:23.000 --> 00:01:24.900 have over 80 on the platform, work together. 00:01:25.019 --> 00:01:27.500 So you can have 11 labs creating audio. You can 00:01:27.500 --> 00:01:30.659 have Google VO3 creating video. You can use ChatGPT 00:01:30.659 --> 00:01:32.680 to generate images, and you can use Cloud for 00:01:32.680 --> 00:01:34.939 your writing. You can build out workflows and 00:01:34.939 --> 00:01:37.719 really cool tools all on the AI Box Builder platform. 00:01:38.019 --> 00:01:40.040 It's linked in the description, and it's only 00:01:40.040 --> 00:01:43.560 $8 .99 a month. In addition, you can chat with 00:01:43.560 --> 00:01:45.879 all the different AI models in one place for 00:01:45.879 --> 00:01:47.739 that same price. Let's talk about what's going 00:01:47.739 --> 00:01:50.200 on with Anthropix Mythos. They have a... Bug 00:01:50.200 --> 00:01:53.780 Hunting AI, and it found 90 critical and 141 00:01:53.780 --> 00:01:56.359 different important bugs in Microsoft SharePoint 00:01:56.359 --> 00:01:59.780 just in April. And this is basically the problem 00:01:59.780 --> 00:02:01.599 here is that it's finding these bugs faster than 00:02:01.599 --> 00:02:04.099 Microsoft's engineers can patch them. So Microsoft 00:02:04.099 --> 00:02:09.360 shipped 600 fixes on July 14th, and their internal 00:02:09.360 --> 00:02:12.199 meetings show that the company was racing a May 00:02:12.199 --> 00:02:15.360 31st deadline before hostile governments build 00:02:15.360 --> 00:02:17.960 equivalent tools. And I'll also say that this 00:02:17.960 --> 00:02:22.310 isn't the only thing. Microsoft SharePoint isn't 00:02:22.310 --> 00:02:24.490 the only thing it's been working on. It has found 00:02:24.490 --> 00:02:28.150 hundreds of critical bugs in Microsoft. 365, 00:02:28.689 --> 00:02:31.789 Microsoft Teams, Microsoft Copilot, all of that 00:02:31.789 --> 00:02:33.330 since the beginning of the year. So I think they 00:02:33.330 --> 00:02:37.229 have about 300 moderate severity SharePoint flaws 00:02:37.229 --> 00:02:40.050 that are still queued for patching. So I mean, 00:02:40.050 --> 00:02:43.110 this is kind of a big deal. And on July of this 00:02:43.110 --> 00:02:46.009 month, or the 14th of the month, Microsoft released 00:02:46.009 --> 00:02:50.069 patches for over 600 bugs, only seven were low 00:02:50.069 --> 00:02:52.830 or moderate severity, and one was already being 00:02:52.830 --> 00:02:55.009 exploited by hackers in the wild. So I mean, 00:02:55.030 --> 00:02:57.360 it literally found something that was actively 00:02:57.360 --> 00:03:00.520 being hacked. The Five Eyes Intelligence Alliance 00:03:00.520 --> 00:03:03.979 warned in late June last month that the window 00:03:03.979 --> 00:03:06.960 for defenders to outpace AI -equipped attackers 00:03:06.960 --> 00:03:09.300 would close within months. Internal Microsoft 00:03:09.300 --> 00:03:11.759 documents show that it might have already closed, 00:03:11.900 --> 00:03:13.719 right, because we already have one exploit being 00:03:13.719 --> 00:03:16.919 used. The real danger, I think, is that Mythos 00:03:16.919 --> 00:03:20.479 can chain multiple small bugs into working exploits 00:03:20.479 --> 00:03:23.740 that humans miss. So Vin Nugent, who's a senior 00:03:23.740 --> 00:03:26.259 anthropic advisor and a former NSA AI chief, 00:03:26.319 --> 00:03:28.560 said that microsoft's standard triage system 00:03:28.560 --> 00:03:32.520 which is patch critical flaws first low severity 00:03:32.520 --> 00:03:35.900 last no longer works when ai can weaponize the 00:03:35.900 --> 00:03:38.360 bottom of the queue so the race is definitely 00:03:38.360 --> 00:03:41.340 on and speed alone might not be enough right 00:03:41.340 --> 00:03:44.340 now we also have google's synth id watermark 00:03:44.340 --> 00:03:46.520 so basically this is an invisible watermark embedded 00:03:46.520 --> 00:03:49.180 on ai generated images primarily from google 00:03:49.180 --> 00:03:51.039 but there's some other a bunch of other companies 00:03:51.039 --> 00:03:52.879 have kind of all worked together to adopt this 00:03:52.879 --> 00:03:56.629 not everybody has i think grok if I'm remembering 00:03:56.629 --> 00:03:58.849 correctly, does not do this, but many others 00:03:58.849 --> 00:04:02.280 do. So, um, when they were, when they did this, 00:04:02.300 --> 00:04:05.560 these AI generated images, um, when it had the 00:04:05.560 --> 00:04:08.340 synth ID watermark, apparently it actually survived 00:04:08.340 --> 00:04:12.159 300 rounds of compression and resizing and real 00:04:12.159 --> 00:04:16.399 testing. And so, and it still was able to detect 00:04:16.399 --> 00:04:18.920 that an image was AI generated. So people might 00:04:18.920 --> 00:04:20.600 say like, Oh, look, I'm going to, you know, take 00:04:20.600 --> 00:04:22.639 this AI generated images image. I'm going to 00:04:22.639 --> 00:04:24.779 compress it. I'm going to resize it. I'm going 00:04:24.779 --> 00:04:27.699 to, um, zoom in and like, do you, you can. try 00:04:27.699 --> 00:04:29.759 to edit it in all sorts of ways. But that watermark 00:04:29.759 --> 00:04:32.399 is still detectable, which is really cool. And 00:04:32.399 --> 00:04:35.019 that Google says proves that their tech is way 00:04:35.019 --> 00:04:37.240 more durable than they were expecting. I think 00:04:37.240 --> 00:04:40.259 there's still one big problem. And that's Google's 00:04:40.259 --> 00:04:43.680 detector can't read watermarks for open AI runway 00:04:43.680 --> 00:04:46.180 or NVIDIA. So I believe if I'm correct, they're 00:04:46.180 --> 00:04:48.639 working with meta on this, and maybe a couple 00:04:48.639 --> 00:04:50.819 other players. But even though they're using 00:04:50.819 --> 00:04:53.920 the exact same technology, OpenAI Runway and 00:04:53.920 --> 00:04:57.060 NVIDIA, like the detector doesn't go cross platform, 00:04:57.279 --> 00:05:00.120 which is kind of weird. So SynthID is obviously 00:05:00.120 --> 00:05:03.240 really impressive. And with those 300 compressed 00:05:03.240 --> 00:05:07.379 cycles on both fully AI generated and AI edited 00:05:07.379 --> 00:05:11.639 images. It was only breaking after a 20 % border 00:05:11.639 --> 00:05:13.759 crop was added on top. That was the only thing 00:05:13.759 --> 00:05:16.399 that was killing it. Google limits synth ID verification 00:05:16.399 --> 00:05:19.540 to about 10 image checks per day per user, and 00:05:19.540 --> 00:05:22.079 they throttle faster lookups on similar images 00:05:22.079 --> 00:05:25.000 to block attackers from reverse engineering their 00:05:25.000 --> 00:05:27.360 system. It's kind of like when you, when Anthropic 00:05:27.360 --> 00:05:30.240 is complaining about everyone going and doing 00:05:30.240 --> 00:05:32.220 distillation attacks, right? They're like, everyone's 00:05:32.220 --> 00:05:34.740 asking Anthropic questions, getting the answers 00:05:34.740 --> 00:05:36.279 and training their AM models on it. Google's 00:05:36.279 --> 00:05:38.720 worried that people are going to use the verification 00:05:38.720 --> 00:05:40.800 like hey is this was this generated with synth 00:05:40.800 --> 00:05:44.079 id to reverse engineer how they're doing i don't 00:05:44.079 --> 00:05:45.620 know if that's really a long -term play because 00:05:45.620 --> 00:05:47.560 eventually right you're going to get enough data 00:05:47.560 --> 00:05:51.019 and figure it out but um for now there it's kind 00:05:51.019 --> 00:05:52.899 of a closely guarded secret it feels kind of 00:05:52.899 --> 00:05:55.019 like it's going to be like these captchas how 00:05:55.019 --> 00:05:57.420 captcha has had to evolve a lot right it used 00:05:57.420 --> 00:05:59.060 to just be like really simple letters and then 00:05:59.060 --> 00:06:00.879 it got more complicated and then it got turned 00:06:00.879 --> 00:06:03.420 into like identify all the bridges in this photo 00:06:03.420 --> 00:06:06.959 and you know zoom this uh you know spin the three 00:06:07.050 --> 00:06:08.889 model till it lines up in the image. Like there's 00:06:08.889 --> 00:06:10.490 all these crazy things that have been invented 00:06:10.490 --> 00:06:12.490 for captures. This is kind of what I feel like 00:06:12.490 --> 00:06:14.870 synth ID is going to become opening. I runway 00:06:14.870 --> 00:06:17.129 and NVIDIA have each built their own versions 00:06:17.129 --> 00:06:19.829 of watermarks. Um, but they all have a little 00:06:19.829 --> 00:06:21.089 bit different of implementation. That's what 00:06:21.089 --> 00:06:22.730 I was saying. Like it doesn't actually go cross 00:06:22.730 --> 00:06:25.129 platform. And so there's a bit of a fragmented 00:06:25.129 --> 00:06:27.430 landscape right now. And these detectors, they 00:06:27.430 --> 00:06:30.170 can only really detect their own watermarks, 00:06:30.230 --> 00:06:32.649 not everyone else's watermarks. The thing that 00:06:32.649 --> 00:06:34.110 I'm the most curious about, and I don't actually 00:06:34.110 --> 00:06:36.310 have a direct answer to this, that, that I I 00:06:36.310 --> 00:06:38.170 would be curious, though, is if it's actually 00:06:38.170 --> 00:06:40.589 able to, you know, you go to like Google generate 00:06:40.589 --> 00:06:42.610 an image, maybe it looks like a great image, 00:06:42.709 --> 00:06:44.689 just go to an open source model, give it that 00:06:44.689 --> 00:06:48.089 image and say regenerate this. If it would kind 00:06:48.089 --> 00:06:51.170 of kill the synth ID on top, and I would guess 00:06:51.170 --> 00:06:54.269 that it might actually kill it. So For that reason, 00:06:54.509 --> 00:06:56.370 I mean, half the time I'm generating images, 00:06:56.470 --> 00:06:59.009 I'm like screenshotting, you know, another image 00:06:59.009 --> 00:07:00.889 or another graphic or something that I like out 00:07:00.889 --> 00:07:02.610 in the wild. And I'm like, hey, use this as like 00:07:02.610 --> 00:07:05.269 an inspiration and change it slightly. And so 00:07:05.269 --> 00:07:07.769 that's how I use like chat GPT to generate images 00:07:07.769 --> 00:07:09.670 a lot of times. And I'm worried it's going to 00:07:09.670 --> 00:07:12.170 be the same thing, which basically makes these 00:07:12.170 --> 00:07:14.829 image generating synth ID things useless, because 00:07:14.829 --> 00:07:18.230 for a majority of, you know, bad actors out there, 00:07:18.329 --> 00:07:20.310 they'll be smart enough to figure that out. And 00:07:20.310 --> 00:07:22.730 99 % of people that just use these AI image. 00:07:22.889 --> 00:07:25.029 generators in a you know responsible or good 00:07:25.029 --> 00:07:27.410 way well it doesn't really matter and so at the 00:07:27.410 --> 00:07:29.370 end of the day yeah i'm not sure if it's gonna 00:07:29.370 --> 00:07:31.269 make a huge difference if i'm you know being 00:07:31.269 --> 00:07:34.170 honest sorry for the pessimism okay encore ai 00:07:34.170 --> 00:07:36.290 has just raised 30 million dollars to build voice 00:07:36.290 --> 00:07:38.129 agents they're all going to be trained on real 00:07:38.129 --> 00:07:41.029 sales calls that have actually closed deals which 00:07:41.029 --> 00:07:43.110 is fascinating right because we have all these 00:07:43.110 --> 00:07:44.990 data sets i have a friend in particular and he's 00:07:44.990 --> 00:07:47.129 got a company and he's got a bunch of sales people 00:07:47.129 --> 00:07:49.410 that help close deals for his cleaning company 00:07:49.410 --> 00:07:51.569 and what's interesting is he uses go high level 00:07:51.569 --> 00:07:54.009 and every single call is recorded for the whole 00:07:54.009 --> 00:07:56.189 conversation. So he has transcripts of every 00:07:56.189 --> 00:07:58.529 single call. And yeah, he knows which of those 00:07:58.529 --> 00:08:00.569 calls actually ended up closing. You can also 00:08:00.569 --> 00:08:03.110 even if he wasn't tracking that bit of information 00:08:03.110 --> 00:08:05.250 in particular, you could usually just go read 00:08:05.250 --> 00:08:06.990 the transcript or have AI read the transcript 00:08:06.990 --> 00:08:09.209 and decide which ones were closed. But that is 00:08:09.209 --> 00:08:11.889 fascinating, right? It's like determining who 00:08:11.889 --> 00:08:13.790 the top sales reps are, which ones are actually 00:08:13.790 --> 00:08:16.290 closing, using those transcripts and training 00:08:16.290 --> 00:08:18.810 models to be more like those. So fine tuning 00:08:18.810 --> 00:08:21.209 models to actually be good at closing deals. 00:08:21.319 --> 00:08:24.339 This startup in particular, Encore AI, they're 00:08:24.339 --> 00:08:26.639 getting their call recordings and they're mining 00:08:26.639 --> 00:08:28.639 all of that to identify what the top performers 00:08:28.639 --> 00:08:32.220 say that move customers forward. And so, yeah, 00:08:32.259 --> 00:08:34.870 this is... in my mind, really, really smart. 00:08:35.250 --> 00:08:37.649 The annual recurring revenue has grown more than 00:08:37.649 --> 00:08:40.389 5x since they started and since they did their 00:08:40.389 --> 00:08:42.509 seed round, which was about 18 months ago. They 00:08:42.509 --> 00:08:45.149 now serve over 40 enterprise customers. Most 00:08:45.149 --> 00:08:46.730 of those are financial institutions that they're 00:08:46.730 --> 00:08:49.570 servicing right now. This was started in 2022 00:08:49.570 --> 00:08:54.110 as ESAIT .io, and then they rebranded to Encore. 00:08:54.210 --> 00:08:56.570 Later on, I think one of their advisors recommended 00:08:56.570 --> 00:09:00.269 it because ESAIT is not, I don't know, very recognizable. 00:09:00.750 --> 00:09:03.679 In any case, they... analyze different customer 00:09:03.679 --> 00:09:06.899 interactions stage by stage to find revenue leaks 00:09:06.899 --> 00:09:08.639 and friction points. And they feed those insights 00:09:08.639 --> 00:09:11.879 into both AI agents and human employees for coaching. 00:09:12.000 --> 00:09:14.000 So what's interesting to me is it doesn't just 00:09:14.000 --> 00:09:17.139 grab that just like the pure transcript. and 00:09:17.139 --> 00:09:19.159 try to copy the whole thing. But it breaks it 00:09:19.159 --> 00:09:21.179 into stages. It's like, okay, we're in, you know, 00:09:21.179 --> 00:09:23.259 maybe it's like a, I don't know if they do cold 00:09:23.259 --> 00:09:24.679 calling, but let's say it's like cold calling. 00:09:24.720 --> 00:09:26.340 It's like, you got to get the person to stay 00:09:26.340 --> 00:09:27.779 on the phone. And it's like, then you got to 00:09:27.779 --> 00:09:28.919 get them to be interested in the product. Then 00:09:28.919 --> 00:09:30.639 you got to give them like a hook. Then you got 00:09:30.639 --> 00:09:34.340 to like, you know, get rid of any sort of issues 00:09:34.340 --> 00:09:35.720 they might have with your product. Then you got 00:09:35.720 --> 00:09:36.899 to sell them on the price, right? Like these 00:09:36.899 --> 00:09:39.059 are these different stages in the sale that you 00:09:39.059 --> 00:09:41.379 got to go through. And so it's breaking the transcripts 00:09:41.379 --> 00:09:44.039 down into all of those different pieces to, you 00:09:44.039 --> 00:09:45.779 know, identify it that way and kind of break 00:09:45.779 --> 00:09:47.830 it into chunks. I think this is so fascinating. 00:09:48.429 --> 00:09:50.470 Encore's edge over Salesforce and a bunch of 00:09:50.470 --> 00:09:53.389 other CRM giants is that the call recordings 00:09:53.389 --> 00:09:55.990 live outside of their platform. So financial 00:09:55.990 --> 00:09:59.090 services is, I think, one of the smartest areas 00:09:59.090 --> 00:10:01.169 that they could get into for this. Banks right 00:10:01.169 --> 00:10:03.950 now and also insurers already record every customer 00:10:03.950 --> 00:10:05.870 conversation for compliance. We've all heard 00:10:05.870 --> 00:10:06.970 it when we're on the phone and it's like, this 00:10:06.970 --> 00:10:08.950 call may be monitored for quality assurance purposes, 00:10:09.090 --> 00:10:11.309 right? And so Encore right now is going to be 00:10:11.309 --> 00:10:12.990 able to plug into that infrastructure without 00:10:12.990 --> 00:10:17.309 forcing a renegotiation of how CRM data gets 00:10:17.309 --> 00:10:19.350 handled across every customer deployment. And 00:10:19.350 --> 00:10:21.549 if they're able to actually train it on, like, 00:10:21.590 --> 00:10:22.850 let's say they go to Chase Bank and they're like, 00:10:22.909 --> 00:10:25.470 hey, Chase Bank, you already have all the transcripts. 00:10:25.470 --> 00:10:28.210 We're going to use your transcripts. So we understand 00:10:28.210 --> 00:10:30.750 how your top salespeople actually convert people. 00:10:30.870 --> 00:10:33.289 Right. So and if they can just keep that data 00:10:33.289 --> 00:10:35.210 with just Chase Bank and they're not like training 00:10:35.210 --> 00:10:38.049 off of that and giving it to, you know, Bank 00:10:38.049 --> 00:10:40.029 of America or something. else, that's a really 00:10:40.029 --> 00:10:42.230 valuable because Chase Bank gets, you know, the 00:10:42.230 --> 00:10:44.950 best their salespeople are able to do. And Bank 00:10:44.950 --> 00:10:46.529 of America gets the best their salespeople are 00:10:46.529 --> 00:10:47.909 able to do. And they're not worried about that 00:10:47.909 --> 00:10:50.149 data or that, you know, confidential information 00:10:50.149 --> 00:10:52.549 leaking. I think this is incredible. And so when 00:10:52.549 --> 00:10:54.769 you have something like Salesforce, currently 00:10:54.769 --> 00:10:57.669 Salesforce and others like that are a little 00:10:57.669 --> 00:10:59.029 bit more centralized, and they don't have those 00:10:59.029 --> 00:11:02.230 kind of out of the box, you know, custom client 00:11:02.230 --> 00:11:05.679 solution. So very cool. Right now, writers and 00:11:05.679 --> 00:11:08.019 editors are deliberately breaking conventional 00:11:08.019 --> 00:11:11.620 style rules. We're all getting rid of em dashes. 00:11:11.620 --> 00:11:14.500 We're all adding typos. Well, I mean, I actually 00:11:14.500 --> 00:11:16.399 just naturally have a lot of typos. So that's 00:11:16.399 --> 00:11:18.600 just me. My wife always rolls her eyes at me. 00:11:18.740 --> 00:11:20.639 But this is actually becoming something that's 00:11:20.639 --> 00:11:22.120 more and more popular. And it's kind of funny 00:11:22.120 --> 00:11:24.000 because in a weird way with this podcast, I've 00:11:24.000 --> 00:11:27.019 always kind of had that same mantra. A lot of 00:11:27.019 --> 00:11:30.269 when I first started this podcast. Prior to this, 00:11:30.309 --> 00:11:33.090 I actually ran a bunch of 100 % AI generated 00:11:33.090 --> 00:11:36.690 scripts and speaker podcasts for a different 00:11:36.690 --> 00:11:38.610 startup I had. And I was just doing it for, you 00:11:38.610 --> 00:11:40.649 know, lead gen or whatever. And they were kind 00:11:40.649 --> 00:11:42.929 of like educational episodes, evergreen content, 00:11:43.049 --> 00:11:45.289 not news like this is. But anyways, from the 00:11:45.289 --> 00:11:46.730 very beginning, I always knew that this was a 00:11:46.730 --> 00:11:49.009 possibility. And I always said, look, I got to 00:11:49.009 --> 00:11:50.950 have a real person, myself, my real voice on 00:11:50.950 --> 00:11:53.169 this podcast talking because people are going 00:11:53.169 --> 00:11:55.909 to want to hear my background, my stories, my 00:11:55.909 --> 00:11:58.970 anecdotes. And even to the point where I was. 00:11:59.080 --> 00:12:01.559 like pauses and ums and buts and like little 00:12:01.559 --> 00:12:04.019 i don't know human things that you put in there 00:12:04.019 --> 00:12:05.940 people want to hear that because they know that 00:12:05.940 --> 00:12:08.679 this isn't an ai talking to them i for one i 00:12:08.679 --> 00:12:10.679 mean even if the voice because the voices used 00:12:10.679 --> 00:12:12.100 to be bad and now they're actually really good 00:12:12.100 --> 00:12:14.559 like if you use 11 labs v3 the voices are incredible 00:12:14.559 --> 00:12:17.259 sometimes i can't tell it's a it's a real um 00:12:17.259 --> 00:12:20.080 or it's just an ai if if i'm not careful and 00:12:20.080 --> 00:12:23.620 i've had calls i've i've heard music um like 00:12:23.620 --> 00:12:26.080 i can detect it but guys i'm serious like i'll 00:12:26.080 --> 00:12:28.360 get I'll listen to a song the other day. I added 00:12:28.360 --> 00:12:32.149 it to my liked folder. on YouTube Music and realized 00:12:32.149 --> 00:12:34.149 probably like five listens in. I'm like, this 00:12:34.149 --> 00:12:37.149 is probably AI generated. So anyways, I think 00:12:37.149 --> 00:12:39.090 this happens more than we would like to admit. 00:12:39.309 --> 00:12:41.590 I mean, this stuff is getting a lot better. But 00:12:41.590 --> 00:12:43.330 my mantra has always been like, look, people 00:12:43.330 --> 00:12:45.190 are going to want a real human. And this is what's 00:12:45.190 --> 00:12:47.289 happening with writers now. They're adding literal 00:12:47.289 --> 00:12:50.289 typos and kind of these messy first person stories 00:12:50.289 --> 00:12:54.029 to make all of their writing look more unmistakably 00:12:54.029 --> 00:12:56.090 human. It's not even just like looking human, 00:12:56.129 --> 00:12:58.129 but unmistakable. Like we want people to know 00:12:58.129 --> 00:13:01.009 this is human. ChatGPT doesn't output text like 00:13:01.009 --> 00:13:03.629 this. It doesn't use all lowercase. So this is 00:13:03.629 --> 00:13:06.429 really interesting. One publication in particular 00:13:06.429 --> 00:13:08.789 said that they got a thousand pitches in May 00:13:08.789 --> 00:13:11.740 and almost all of them were AI generated. Pitchfork's 00:13:11.740 --> 00:13:14.940 editorial head has banned AI from review submissions. 00:13:15.259 --> 00:13:17.779 That was starting in 2024. And I think that was 00:13:17.779 --> 00:13:21.279 like two years ago. But novelists are actively 00:13:21.279 --> 00:13:23.940 avoiding a lot of different things. And I mean, 00:13:23.940 --> 00:13:26.100 you'll see it on Twitter. It's kind of funny 00:13:26.100 --> 00:13:28.740 or an X or whatever, where everyone like finds 00:13:28.740 --> 00:13:31.620 em dashes and old literature or they find like 00:13:31.620 --> 00:13:35.440 just, you know, funny glitches that Claude always 00:13:35.440 --> 00:13:37.480 says, you know, it's not this, it's that right? 00:13:37.580 --> 00:13:39.399 Like all those funny things, and they'll screenshot 00:13:39.399 --> 00:13:41.000 them from old books. and they're like, oh my 00:13:41.000 --> 00:13:43.399 gosh, I can't believe Jane Austen was using AI 00:13:43.399 --> 00:13:46.200 to write. So anyways, there's a lot going on 00:13:46.200 --> 00:13:51.360 inside of the writing world trying to be more 00:13:51.360 --> 00:13:55.740 human. Sierra is a $12 billion data security 00:13:55.740 --> 00:13:57.940 company and they're buying Oasis Security for 00:13:57.940 --> 00:14:01.759 $1 billion to manage AI agent identities inside 00:14:01.759 --> 00:14:04.559 of companies. So right now, basically every company 00:14:04.559 --> 00:14:06.500 is putting out all of these different agents 00:14:06.500 --> 00:14:09.279 to do work, but they need some tools to track. 00:14:09.720 --> 00:14:12.700 who those agents are, and what they have access 00:14:12.700 --> 00:14:14.919 to what they're doing. This is a problem that 00:14:14.919 --> 00:14:17.100 I think a traditional security software would 00:14:17.100 --> 00:14:19.080 be tracking humans. But now that we have these 00:14:19.080 --> 00:14:20.879 agents, they also want to track that. And I think 00:14:20.879 --> 00:14:22.860 it's interesting, because I've seen some tweets 00:14:22.860 --> 00:14:25.460 on x where people were like, they're screenshotting, 00:14:25.460 --> 00:14:29.580 I think it was chat GPT, like 5 .6, or whatever. 00:14:29.679 --> 00:14:32.720 And It was on cursor. They were using cursor 00:14:32.720 --> 00:14:35.500 and they said they were working on a project. 00:14:35.580 --> 00:14:37.480 OK, stop for now. Finish this later. And it made 00:14:37.480 --> 00:14:39.360 a push after I told it to stop. Why did you do 00:14:39.360 --> 00:14:41.120 that? And I was like, oh, I just like was doing 00:14:41.120 --> 00:14:43.399 the push. Anyways, these things happen where 00:14:43.399 --> 00:14:44.840 these agents, you know, especially if you give 00:14:44.840 --> 00:14:46.559 them a task, especially if it's more open ended, 00:14:46.639 --> 00:14:49.120 like, hey, just get X, Y, Z thing done. They 00:14:49.120 --> 00:14:51.259 do it in their own way. And we don't always know 00:14:51.259 --> 00:14:53.360 what made the change, what made the edit, who 00:14:53.360 --> 00:14:55.820 did what. And so you can have some of these issues. 00:14:56.080 --> 00:14:58.919 Anyways, Oasis was founded in 2022. They raised 00:14:58.919 --> 00:15:01.470 one hundred and nine. million. They specialize 00:15:01.470 --> 00:15:04.990 in securing non -human identities and they do 00:15:04.990 --> 00:15:08.549 it through API, you know, API keys that are logging 00:15:08.549 --> 00:15:10.330 into corporate systems without any humans there. 00:15:10.429 --> 00:15:13.350 They just raised $600 million, a $12 billion 00:15:13.350 --> 00:15:16.509 valuation. And they're now making $150 million 00:15:16.509 --> 00:15:19.350 a year in revenue, but they're still not profitable. 00:15:19.629 --> 00:15:21.950 They've spent really aggressively. They have 00:15:21.950 --> 00:15:24.210 three acquisitions that they just did in quick, 00:15:24.330 --> 00:15:27.129 like really close together. This is their third 00:15:27.129 --> 00:15:29.470 acquisition in a few months, which is following 00:15:29.470 --> 00:15:32.450 them buying Rift and Genie Security. I think 00:15:32.450 --> 00:15:34.190 it's interesting when you have a company that's 00:15:34.190 --> 00:15:35.970 able to raise so much money. Obviously, they 00:15:35.970 --> 00:15:37.889 have a really good team. They have a really good 00:15:37.889 --> 00:15:40.470 visionary at the top. But when they make all 00:15:40.470 --> 00:15:43.519 these acquisitions, I mean, A, they're smart, 00:15:43.659 --> 00:15:46.879 but B, they're growing so fast and they have 00:15:46.879 --> 00:15:48.679 so much money, they have the bandwidth to do 00:15:48.679 --> 00:15:50.299 it and not to build it from scratch, which is 00:15:50.299 --> 00:15:53.159 interesting. There's a huge push right now to 00:15:53.159 --> 00:15:55.759 own the non -human identity market before a lot 00:15:55.759 --> 00:15:57.840 of these bigger companies like Palo Alto Networks 00:15:57.840 --> 00:16:00.179 or Okta are kind of moving into this. And so 00:16:00.179 --> 00:16:02.139 that's why I think they're jumping on that. Other 00:16:02.139 --> 00:16:03.679 news that happened this week that I just want 00:16:03.679 --> 00:16:06.500 to cover the headlines. Google's DeepMind has 00:16:06.500 --> 00:16:09.840 just shipped Lyria 3 .5 in flow music with some 00:16:09.840 --> 00:16:12.370 better vocals and lyrics. I've talked about Lyria 00:16:12.370 --> 00:16:14.570 in the past, not been impressed with it. When 00:16:14.570 --> 00:16:16.149 it was first came out, it was like only able 00:16:16.149 --> 00:16:18.070 to generate 30 second clips. Then it became, 00:16:18.169 --> 00:16:20.049 then it was able to do full songs. And now I 00:16:20.049 --> 00:16:21.590 think they're getting better at vocals and lyrics. 00:16:21.669 --> 00:16:23.470 So I think this is going to be a big player, 00:16:23.509 --> 00:16:26.370 challenging Yudio and Suno and stuff. When it 00:16:26.370 --> 00:16:28.129 first came out, it wasn't much to talk about, 00:16:28.210 --> 00:16:30.070 but I think Lyria is getting better. I still 00:16:30.070 --> 00:16:32.029 like Suno more because they have the entire AI 00:16:32.029 --> 00:16:34.549 studio and they have like music covers you can 00:16:34.549 --> 00:16:36.649 do. And there's just so much to do with Suno. 00:16:36.730 --> 00:16:39.769 But I am excited that the underlying model behind 00:16:40.360 --> 00:16:43.080 Google is getting better. Panagram just raised 00:16:43.080 --> 00:16:45.779 $9 million to detect AI generated text. It apparently 00:16:45.779 --> 00:16:47.840 has 99 % accuracy. It's interesting, you know, 00:16:47.840 --> 00:16:49.860 in light of what we were talking about with everyone 00:16:49.860 --> 00:16:52.899 using more, I don't know, typos or whatever you 00:16:52.899 --> 00:16:56.240 want to call it to seem less AI generated. So 00:16:56.240 --> 00:16:58.080 now we have these tools that are really good. 00:16:58.139 --> 00:17:00.580 It's interesting because OpenAI in the very early 00:17:00.580 --> 00:17:04.920 days. came out, they had their own AI text detector. 00:17:05.519 --> 00:17:07.779 And they actually shut the project down because 00:17:07.779 --> 00:17:09.920 they said it was too hard to detect AI generated 00:17:09.920 --> 00:17:12.700 text. Now, I think we all kind of have AI generated 00:17:12.700 --> 00:17:16.440 text detection built into our brains now, where 00:17:16.440 --> 00:17:18.460 we're understanding a lot of the phrases that 00:17:18.460 --> 00:17:20.339 are used, we're understanding em dashes, but 00:17:20.339 --> 00:17:22.339 those things that AI models are going to be able 00:17:22.339 --> 00:17:25.150 to get around very easily. I actually have a 00:17:25.150 --> 00:17:27.609 whole list of things that AI models frequently 00:17:27.609 --> 00:17:31.430 say that annoy me. And whenever I get AI to write 00:17:31.430 --> 00:17:33.789 a podcast script or if I get it to go write an 00:17:33.789 --> 00:17:37.069 article or I go get it to write a blog post or 00:17:37.069 --> 00:17:39.710 content or like basically any work I'm doing 00:17:39.710 --> 00:17:42.339 or even like. even websites i very frequently 00:17:42.339 --> 00:17:44.740 will say like run through this whole list and 00:17:44.740 --> 00:17:46.819 remove every single m dash from the website every 00:17:46.819 --> 00:17:49.859 single time that we say x y or z and i just get 00:17:49.859 --> 00:17:52.960 it to go like run this kind of audit on my content 00:17:52.960 --> 00:17:54.980 now eventually hopefully the models will just 00:17:54.980 --> 00:17:58.140 get away from that but uh it is interesting all 00:17:58.140 --> 00:18:01.299 this ai generated text detection stuff. I think 00:18:01.299 --> 00:18:03.799 I like I hope that it's good, but I'm always 00:18:03.799 --> 00:18:05.299 dubious of it because I feel like it's always 00:18:05.299 --> 00:18:07.759 a cat and mouse game. Okay, there's funny news. 00:18:07.880 --> 00:18:10.339 Martha Stewart has co founded hint, which is 00:18:10.339 --> 00:18:13.259 an AI assistant for homeowners. And she has $10 00:18:13.259 --> 00:18:15.240 million in funding for this project. So that's 00:18:15.240 --> 00:18:17.220 kind of interesting. If you haven't left a review 00:18:17.220 --> 00:18:19.779 on the podcast yet, it would help the show out 00:18:19.779 --> 00:18:22.259 a ton. I would super, super appreciate it. It 00:18:22.259 --> 00:18:24.220 helps more incredible people like yourself find 00:18:24.220 --> 00:18:26.839 the show. And so yeah, it's the number one thing 00:18:26.839 --> 00:18:28.759 that could help this show reach more a bigger 00:18:28.759 --> 00:18:30.779 audience. If you wouldn't mind if you haven't 00:18:30.779 --> 00:18:32.859 already done it, it would mean the world to me. 00:18:32.920 --> 00:18:34.519 But thanks so much for tuning in. I hope you 00:18:34.519 --> 00:18:36.119 guys have an incredible day. Make sure to go 00:18:36.119 --> 00:18:39.119 check out AI box .ai if you want to get 80 different 00:18:39.119 --> 00:18:41.619 AI models all on one platform, and I'll see you 00:18:41.619 --> 00:18:42.420 in the next episode.