October 26, 2026 2026-10-26 https://www.youtube.com/watch?v=QE1RSFy1544 [0:11] We go live in ten. [0:25] Every morning, the future moves faster. [0:30] The models, the money, the power, and the people building what comes next. [1:09] From Austin, Texas to the world. [1:16] AI news, real context. [1:19] This is the Austin Wilson Show. [1:21] I'm Austin Wilson, your human in the loop. [1:30] Good morning, and welcome to the Austin Wilson Show, the number one daily AI news show in the world. [1:38] We are covering the breaking AI headlines that matter to you. [1:43] Today. [1:44] I'm Austin Wilson coming to you live from the AI capital of the world, Austin, Texas, and we have a full docket for you today. [1:51] We will jump right into it. [1:53] First of all, Anthropic is finally filed to go public. [1:58] Claude was found pulling Georgia voter data without permission. [2:04] Texas is suing OpenAI to keep ChatGPT from its children. [2:10] American Express is cutting jobs partly for AI. [2:14] Texas football's AI fourth down claim. [2:17] And finally, OpenAI sealed off AI agents that still talked. [2:22] And lastly, we've got a special guest for you guys today. [2:25] Excited to introduce you shortly. [2:29] Let's get into it. [2:30] Right at the top, uh, it's official. [2:33] The company behind Claude just filed to sell stock to the public. [2:37] Anthropic made its filing public Monday morning. [2:41] It wants to trade on the Nasdaq as Ant, a n t. [2:45] It hasn't set a price or a share count. [2:48] NPR says the sale could come as soon as November. [2:53] We've been talking about this quite a bit lately. [2:56] And finally, regular investors could soon own a piece of Claude's maker, Anthropic. [3:01] Reports value it at 1,500,000,000,000 to $2,000,000,000,000. [3:07] Anthropic says its engineers now add over 45 times more code per day than in 2024. [3:15] Sales hit $16,300,000,000 in the first half of twenty twenty six. [3:21] That's 14 times the same stretch last year, but it lost $80,700,000,000 in the same six month period. [3:29] Most of that is a paper charge tied to its soaring valuation. [3:35] Wow. [3:35] We're finally here. [3:36] We've been talking about this for quite some time, and now Anthropic has actually filed their s one. [3:42] They've decided it is time to go public, and they're gonna, um, make an effort to raise a ton of money by making their stock available to the general public. [3:53] This is a very interesting timing. [3:56] Uh, as we heard last week that, uh, OpenAI has made their own announcement to try to raise funds privately. [4:04] Anthropic, of course, is in an effort to go public to raise their own capital to continue to fuel this AI frenzy that we are seeing, and it is not slowing down at all. [4:17] Even in the past couple weeks that we've seen chip stocks pull back, there's been a lot of bearish talk in the news. [4:24] Anthropic is still ready to get rocking and rolling and put their put their product out on the line and and let let the public investors come in and make their bets there as well too. [4:39] This is this is great timing. [4:42] This is an exciting time. [4:43] This is really interesting, especially around when we see the kind of the election and political cycle coming in as well too. [4:52] Very, very interesting timing. [4:53] There's rumors rumors that a Fable five five model will be released imminently as well too, which means that Anthropic has their next iteration of intelligence to drop on the general public, which could be a huge boost for their IPO depending upon performance and the impact. [5:15] And, of course, last week, we saw them release Claude Duet, their personal agent, which is another, uh, kind of effort to make their company more valuable. [5:29] How many more products and features can they release before they go public? [5:33] How much hype can they raise and continue to fuel the fire of frenzy to get investors interested and willing to put their money on the line. [5:45] Let me ask you guys, would you buy Anthropic stock on day one? [5:53] I think I'm gonna hold off. [5:56] I'm bullish, but I think I'm gonna hold off just because I think the hype is too much on a named stock like Anthropic. [6:04] We saw the same thing with SpaceX. [6:06] It came out and skyrocketed on day one and then had a major pullback. [6:13] Not saying that we'll see the same thing with Anthropic, but that is something that I would absolutely look out for and maybe try to time it a little bit. [6:24] I don't know. [6:25] Alright. [6:25] Moving on to the next story. [6:27] Anthropic again. [6:29] Anthropic pulled Georgia voter data. [6:33] Nobody asked it to. [6:34] It got around a paywall on its own. [6:36] So Claude was able to finagle its way. [6:39] Axios Axios broke the story Sunday. [6:42] Claude was doing an internal statistics project, and then it found an old copy of a state web page that showed how to get an access pass to some secured data. [6:54] So Claude requested one. [6:56] It pulled voter statistics the state normally charges for. [7:00] This looks like a kind of inadvertent hack here. [7:04] Election day is eight days away, and AI reached state election data with no human sign off. [7:09] Anthropic says it has since added blocks to stop moves like this. [7:15] Georgia's deputy secretary of state confirmed it. [7:18] Anthropic reported the incident October ninth ninth, but left out which state was impacted. [7:24] But Georgia calls it public data. [7:26] Claude never reached voter files or the registration system. [7:31] It was all actually just public data. [7:33] So not a huge breach here, but something that we're seeing a trend in cybersecurity if you... [7:41] If we wanna say that this is in the the realm of cyber security and that the way the Internet was built and has been architected up to this point was under the understanding that humans had to manually try to hack into these systems. [7:59] And a lot of systems, the payoff to, you know, hack into them is not worth the effort that it would take to actually do the hacking. [8:11] Right? [8:11] So we built a lot... [8:13] This structure that it was like, I mean I mean, you think of maybe an analogy where... [8:20] This is maybe a great, like, medieval times. [8:23] You know, we built this... [8:25] The Internet like a a medieval castle. [8:27] We've got a moat. [8:29] We've got walls. [8:31] We've got, you know, defenses set up around the perimeter of any website. [8:36] Right? [8:37] And previously, we built these defenses to guard against, you know, bow and arrow and, you know, maybe some catapults and some kind of archaic military, uh, or or archaic medieval military tools. [8:53] Right? [8:55] The problem is now somehow we've developed an airplane that could just fly over and drop right into the castle, and we haven't built up the defenses to protect against that, if that analogy makes sense. [9:09] So we have a lot of websites out there where it was like, it wouldn't ever make sense or or be worth the effort for people to manually spend, you know, maybe days, months, years trying to hack into a website to get, you know, public data. [9:28] That being said, now we have AI that's smart enough to figure out its way around and do that really quickly and can get into these sources. [9:37] Now in this case, the the consequences were were not that big of a deal. [9:42] Right? [9:42] This this was publicly available data according to the state of Georgia, but Claude was able to find a way in pretty simply, it sounds like. [9:52] And so what we need to do and what we're starting to do is build up these defenses around traditional Internet architecture and how that was built and start to fill in these gaps that is gonna make it harder for AI to break into these even inadvertently where it sounds like in this case, it wasn't nefarious at all. [10:13] Claude was doing a, you know, quote unquote research project and happened to find a backdoor into this system and found its way in, got the data it needed. [10:27] It didn't really maybe realize that it was doing something it shouldn't do. [10:33] Question to the audience, should AI Labs have to say which entities or states they impacted in this case? [10:45] Also, who's liable in this case? [10:49] Is this the state of Georgia's fault, or is this on Anthropic? [10:54] That's a question we'll be having to answer as we move forward with this progress. [11:01] Alright. [11:02] Texas just sued OpenAI and Sam Altman himself. [11:06] It wants ChatGPT off limits to kids in the state of Texas. [11:12] Attorney General Ken Paxton filed it Friday. [11:15] He also wants a court approved reviewer to clear new AI models before Texans get them. [11:22] He seeks $10,000 per violation. [11:25] Paxton is running for US senate. [11:30] This is a very interesting breaking headline here. [11:35] Kids use chatbots for homework and for personal company. [11:40] A a win could could force real age checks on the use of ChatGPT and agents like it. [11:47] Paxton says ChatGPT encouraged suicide and got kids hooked with no parental oversight. [11:54] The suit cites Adam Rain. [11:56] He was 16 and died by suicide after long ChatGPT chats. [12:02] But Elon Musk is the biggest back... [12:04] Backer of senate run. [12:07] Musk is also Altman's long time rival. [12:11] That's interesting. [12:12] Is there a conflict of interest here? [12:15] This is something that society is going to have to decide on pretty quickly here. [12:21] Again, coming back to what we talked about in the last story, but who is liable for adverse effects of AI incidents and who's responsible for protecting, especially in this case, you know, kids, the most vulnerable. [12:39] It's really interesting to think about these... [12:44] These are difficult problems to solve and who should be the the entity or who should solve these problems. [12:51] That's something that we need to decide pretty quickly. [12:55] The problem with this is, as I pointed out, AI is not political. [13:04] Right? [13:05] But this is becoming a political problem that a lot of politicians on either side are kind of jumping onto one bandwagon or another, and it's really confusing too because it's not straightforward. [13:21] It's not, you know, the Democrats on one side and the Republicans on on the other side. [13:26] It kinda seems like both sides are for or against or... [13:29] But it kinda depends whose AI they're for and who's... [13:33] They're against, and it's not, you know, straightforward. [13:38] So this is really very very much murky waters. [13:44] The issue that I have with this is is, you know, one, Ken Paxton is running for senate and in a very contested race where AI is a major political topic. [14:00] And it's... [14:00] You know, the question is, are you just trying to win political points here or are you trying to have a major impact? [14:07] Of course, we're days away from the election. [14:09] So... [14:10] And he would vacate his seat as attorney general. [14:13] So maybe this is one of his final acts as attorney general would be this lawsuit. [14:19] And so it's hard to see like, is this something that he would actually follow through with? [14:26] Is this just to gain political points? [14:28] Hard to tell. [14:30] You know, my personal opinion is, like, this stuff does need to be litigated. [14:37] If OpenAI or Anthropic are culpable in any of, you know, nefarious activity of their product or their product creates harm that, uh, it did not warn against or the liability ends up being on those entities, then then they should be, you know, prosecuted properly. [15:00] But the fact that we would make it illegal in Texas for kids to use these tools, which are are gonna be super super powerful for both education, but also for for them learning how to develop their own skill set moving forward. [15:20] Like, that... [15:20] That's huge to do that. [15:22] And then the biggest piece of this, think, is that the state of Texas would have the right to review any new models before they're released. [15:30] That is outrageous in my opinion. [15:33] And especially, you know, when Ken Paxton is backed by other AI company providers like Elon Musk and SpaceX. [15:42] You you can't help but see the conflict of interest. [15:45] What do you guys think? [15:46] Should ChatGPT be off limits to kids? [15:50] Is the lawsuit about safety or is it about politics? [15:57] You guys be the judge. [15:58] And I guess we probably collectively all will be the judges as we come up on election day here shortly. [16:09] Up next, uh, last week, I got to sit down with Brynn TileLang. [16:14] She runs, uh, social sales link and trains salespeople on LinkedIn, and now she teaches them to use AI without losing their own voice. [16:26] Uh, one of her lines really stuck with me. [16:29] She said, the human is the loop, not the human in the loop. [16:34] Here's our conversation we recorded earlier. [16:40] Where do you think that kind of marketplace is with AI, and where do you think we're going from here? [16:46] Are we at the very beginning of Oh, yeah. [16:48] Kind of this tech revolution? [16:50] In the tech revolution, we're like day three of the big bang. [16:54] Okay. [16:55] Let's go. [16:56] I love that. [16:57] I think there's a vision of where this can go, but I also think we have no idea. [17:02] We can't even envision. [17:04] Yes. [17:04] Where it can go. [17:06] All the speculation in the world can kinda create excitement or fear, but we really don't know the evolution of this baby. [17:17] It's kinda like when Alexander Graham Bell created the phone. [17:21] There's no way they could have ever imagined an iPhone. [17:24] Yes. [17:25] Right? [17:25] And I think that's where we are today at right now. [17:28] That's amazing. [17:29] I love that analogy too. [17:30] It actually reminds me of Thomas Edison when he created the... [17:35] Oh, the phonograph. [17:36] Yeah. [17:36] Right. [17:36] He would never... [17:38] Yeah. [17:38] His use case was, oh, people would play sermons on it. [17:42] Right? [17:42] He never thought... [17:43] Like, much less Spotify, but never thought even, you know, radio or, you know, contemporary music, anything like that. [17:50] So that's really interesting. [17:51] I would agree with you too. [17:52] I think we're we're super early. [17:54] And I've said before, it's a paradigm shift. [17:57] Right? [17:57] A lot of people look at this as the next incremental step in the Internet. [18:01] So tell me a little bit about kind of the work that you're doing and the impact that you're seeing maybe on the day to day, and what would lead you to believe that that we're super early in this. [18:11] I'll start with what makes me believe we're super early in this is we're super early in this. [18:15] Right? [18:16] Like Yeah. [18:16] It just is. [18:18] You know, although AI has been around for for decades, it really has become mainstream in the last three or four years period. [18:27] Right? [18:28] So if you look at the evolution of anything, the first three or four years is, you know, training wheels. [18:33] Then while we are in a world where everything will get escalated based on technology, it's still the big bang. [18:43] Right? [18:43] It's expanding further every single day, and we have absolutely no idea when things would reverse back into a condensing. [18:54] So right now, we're in expanding mode. [18:57] I'm using it and teaching it at a very simple level. [19:02] I'm a sales trainer who focuses on LinkedIn and now bring... [19:09] Bringing AI into it. [19:11] There are a few things. [19:12] I represent three companies. [19:14] One is a company called Aggie, letaggiedoit.com, and we are a social media tool designed for localization. [19:27] And it's an AI tool, and it's amazing. [19:29] And we have a free grader, socialmediagrader.ai. [19:31] Anyone can use it to grade their Instagram and LinkedIn and give you ideas, and then it's like a $29 a month tool. [19:38] So I'm a a fractional chief revenue officer for that. [19:42] I'm another company called Craft Works, which is a sales training company. [19:46] It's really not a training company anymore. [19:48] It's changing training. [19:49] Training is going away. [19:51] Right? [19:51] It's now... [19:52] It's pure enablement called agent works where every single activity that you do with a prospect client has one source of truth. [20:02] Where in sales, CRM used to be the one source of truth that's shifted because there's many sources of truth. [20:08] Right? [20:09] There's activity that's happening on LinkedIn that's in your email, that's in your text. [20:15] And so agent works brings all the activity together and not only enables the salesperson to do the next thing, but it does it in a way that creates almost no friction. [20:33] So here's my calendar for the day. [20:35] Your next call is in fifteen minutes with Austin Wilson. [20:38] This is what you're gonna talk about, summary of the three email exchanges you had based on what we know. [20:44] Here is your goal. [20:45] What we know about Austin, this is likely his goal. [20:48] Right? [20:48] And there's a free call plan. [20:50] Everything is just fed. [20:52] So you can read it in three minutes before a call and be hyper ready. [20:58] It can also say, you know, Austin, you know, often talks about sci fi in his Instagram, you know, whatever. [21:05] Right? [21:05] All those things that can come together. [21:08] That's another... [21:08] And then there's my company, which is social sales. [21:11] Like, we're a LinkedIn sales training company, and now we've just enabled AI with prompts. [21:17] So, you know, how do we leverage our relationships and our voice? [21:24] So all that said, the lowest common denominator from my perspective is the human is the loop, not into the loop. [21:35] Right? [21:36] Like, how do we use AI and keep our voice? [21:40] That's using transcripts. [21:41] It's making sure that AI is learning you and giving you the opportunity to correct it, not make assumptions. [21:51] So a lot of it in the prompts that we use are around questioning. [21:56] If you wanna comment for this blog post, what did you like about that blog post? [22:01] What stood out to you? [22:02] So it's not just writing it. [22:06] It's helping you better articulate your thoughts, not Yes. [22:11] Substitute for your thoughts. [22:13] Wow. [22:14] That's great. [22:15] I love that. [22:16] You called out a couple of really key things there, in sales. [22:19] But every role, there's a lot of ancillary tasks that you have to take in order to do your actual job, and your actual job becomes a small portion of that. [22:29] And maybe that's what you're passionate about and what you love doing. [22:31] You love selling. [22:32] You love talking to people. [22:33] You love helping people, but a large portion of your time is updating the CRM, checking your notes, reading back... [22:40] Whatever, trying to keep everything together so that you can be productive. [22:44] And so I think you're totally right when we see tools like this that enable us. [22:48] And I love that it doesn't substitute the thoughts. [22:51] It helps you articulate your thoughts. [22:53] And I think that's what what moves us forward. [22:56] What's kind of your message to people that are maybe uncertain about AI or don't know the power? [23:03] Where where do you think they should get started, uh, to learn this and really experience it? [23:08] So I think the first thing they have to do is identify where the gaps are or where the challenges are in their world. [23:16] Right? [23:17] So I have an app that will give me my grocery shopping list for Costco based on the foods that I like to eat and how much protein I want in every meal. [23:30] And do I want green vegetables and, you know, do I like or not like cauliflower? [23:36] Right? [23:36] Like Yeah. [23:37] And so that has learned me, and now I can go in, and then it will give me all the recipes based on all the food I just... [23:44] And that's an AI, you know, program that's great for me. [23:48] Other people wanna be creative and don't wanna plan their meals. [23:51] Right? [23:52] So you have to identify where in your life you have a challenge or a gap or some kind of need for support. [24:03] Right? [24:03] And and I think you have to start there, whether it's personal or confessional. [24:08] Where would you like help? [24:10] One of the questions that I'll often ask a new client is if you could hire a VA and wave a magic wand and that VA could take over tasks in your life, what would they be? [24:24] Right? [24:24] And, you know, sometimes they'll say cleaning my house. [24:27] Great. [24:27] Go get a Roomba. [24:28] That's fun. [24:29] Right? [24:29] Yeah. [24:30] Yeah. [24:30] But there are things, right, that we can look at. [24:33] One of them for me was a shopping list. [24:37] And I'm like, okay. [24:38] Well, let's go find the tool for that. [24:40] And there's lots of them out there. [24:42] You can even build it in a Claude tool. [24:45] Right? [24:45] There's so much that you can do. [24:47] So I think identifying where the need is should be the very first step before you go learn something. [24:55] That's a great message. [24:56] Part of the issue, and I even experienced this too, is you revert back to your normal practices. [25:03] Right? [25:03] Like, the normal way that you you handle the problems, you just go in that motion. [25:08] And and getting out of that thinking AI first, I think, has been a big impact for me as well too where it's like, what are my problems that I could actually solve with AI? [25:17] Let's go see it, and there's probably a tool for it. [25:20] Otherwise, you could go down the rabbit hole and learn lots of things that they'll never need or use. [25:24] Yeah. [25:25] Right? [25:26] Yeah. [25:27] Thanks so much, Brynn. [25:28] Appreciate it. [25:28] Absolutely. [25:29] Pleasure to meet you. [25:34] Thank you so much again to Brynn for taking the time to sit down with me and and talk a little bit about AI. [25:43] I think the kind of big takeaway for me there is seeing the the practical impact of AI that she exemplified both from, you know, a couple great lines there where it's using AI to amplify your humanity and and give you more opportunity to be creative and be more human. [26:06] And then figuring out like what are the problems, what are the gaps that I currently have, and how can I use AI tools to fill those gaps? [26:14] I think that kind of paradigm shift makes a big impact and has had an impact on me as well too. [26:22] What's one line that stuck with you from Bryn? [26:25] I'm curious to hear what the audience kind of feedback is here. [26:30] You can find Bryn at socialsaleslink.com. [26:35] Again, if you know anybody else that would love to jump on and talk about daily AI news and what we have going on here in the world of AI, I would love to chat with them. [26:47] You can fill out our form on taws.live. [26:52] That's taws.live is our website. [26:56] And you can also sign up for our newsletter there as well too where we release daily AI news. [27:03] Thanks again to Brynn, and, uh, hopefully, we can have her on again sometime soon. [27:10] Um, and then let's jump into the rest of the news for today. [27:16] American Express is cutting about 3,500 jobs. [27:20] Its finance chief says it's partly to prepare for AI. [27:25] CEO Steven Square announced it Friday on their earnings call. [27:30] That's 5% of the company's workforce. [27:33] AP calls it AMX's first big round of cuts in about a decade. [27:38] It's taking a $336,000,000 charge. [27:44] Wow. [27:45] Big companies are cutting jobs now to make room for AI later. [27:50] The finance chief says fewer layers of managers will let Amex automate with AI. [27:58] As we covered last week, Barclays confirmed 5,000 job cuts in its own tech driven cost push, but business is booming. [28:08] Revenue grew 9% last quarter to $20,100,000,000. [28:14] This is a a really interesting thing that we've... [28:17] Is this a trend here that we're seeing? [28:19] Is it something just in the financial sector? [28:22] I think that's interesting that we saw this come out of Barclays last week, 5% cut, and then Amex this week announcing their own 5% cut in head count. [28:31] Is this a standard operating procedure from the financial institutions just clearing out some bloat? [28:39] Is is AI being used as an excuse here, or are they just looking to pad their numbers and better their profit as they come into the end of the year and hit their bonuses? [28:54] What is the motive here? [28:57] I think that we are in an interesting regime here from a financial perspective. [29:05] And to some extent, it is uncharted waters. [29:12] Right? [29:13] We... [29:13] We've talked about historical technological revolutions around the Internet or railroads or automotive, you know, new inventions across the board like in... [29:27] The impact that they've had, the industrial revolution. [29:29] Right? [29:30] In each one of these cycles, you kind of see a, you know, boom period and a hype cycle where people are pouring in a lot of money to build out this new technology and capacity for the new technology and to take advantage of it. [29:48] And then a lot of times we'll see a consolidation period. [29:51] So I think what's interesting or what's difficult is everybody knows this. [29:55] Right? [29:55] This is not a secret. [29:57] You know, the .com boom in 2000 and bust and boom in in 2000, and we saw in the railroad era, the railroad build out was huge and a lot of money was poured into it. [30:10] And then there was a bust and JPMorgan came in and scooped up all the railroad companies. [30:16] And then, you know, with cars, the the automobile, we saw automakers blow up. [30:22] There are tons of automakers, and then they all kinda consolidated. [30:25] Same thing in the Internet. [30:26] It's just a standard thing. [30:28] Everybody pours in. [30:29] There's only a certain number of winners, and the losers end up kinda getting washed out, the winners collect what's left over. [30:37] So we know that that's going on. [30:39] The problem is we don't know where we're at. [30:42] So, like, are we close to that bust cycle? [30:45] Are we just at the beginning of the boom? [30:47] If we're just at the beginning of the boom, you you don't wanna miss out on all the, you know, positive upside. [30:55] Right? [30:56] If we are nearing the end of the boom, you don't wanna have all your chips on the table either. [31:03] Right? [31:04] So it's... [31:04] That's what everybody's measuring here. [31:06] And I wonder if the financial sector, these financial companies are looking at this as, hey, let's be a little conservative. [31:13] Maybe we can pull some chips off the table just in case. [31:17] And we're gonna say that it's AI is the reason that we're doing it because we're using so much AI internally. [31:24] We don't need the head count, but also, you know, peak profit numbers. [31:28] Like, it's just really interesting. [31:30] What is the incentive here? [31:31] Is this a trend or is this just the financial companies, Amex, Barclays, looking at ways to better their, you know, net profit numbers. [31:45] Would you cut jobs before AI can actually do them? [31:49] Because I think in this case, we're seeing the reason is AI, but it's not AI today. [31:55] It's well, we think in the future AI will be able to do a lot of this, so we're gonna cut it now. [32:00] Is your company, the companies that you guys work for, do you think they're quietly doing the same? [32:06] Are we gonna start seeing some pullback here in head counts because of the AI productivity gains that a lot of companies are seeing. [32:17] The other piece of this, which is really interesting too, is I've heard that kind of the biggest spenders on AI right now, they're not really seeing a return on their investment. [32:28] They're pouring a bunch of money into the AI and these tools, but they're not seeing a big return on that investment. [32:37] I wonder if this is part of that return that they're seeking to to gain. [32:41] Hey. [32:41] We spent, you know, however many $7.08 figures on AI in the past year. [32:48] We need to start recoup... [32:49] Recouping some of that investment. [32:52] Where is that coming from? [32:53] Well, we can reduce head count. [32:55] We don't... [32:55] Now we don't need as many people to get that job done. [32:59] Interesting to see this. [33:02] We will continue to track this story and see this if this is a trend. [33:06] What do you guys think? [33:07] Is this a trend in layoffs and we're just at the beginning of it, or this is a one off kind of event? [33:17] Next, the Texas athletic director said the team used AI on a big fourth down and it lost three yards. [33:25] The Austin American Statesman broke the story Friday. [33:29] The new AI tool is called BevoLab. [33:32] Texas trailed Florida 20 to 17, and coach Steve Sarkisian went for it on fourth down. [33:38] Archman and got stuff. [33:46] Oh, wow. [33:47] Okay. [33:47] Schools now pitch AI as a game day edge. [33:51] One quote tied a failed play to it. [33:54] Crystal Conte called the new lab earth shattering and the new wave, quote unquote. [34:00] Dell and Google Cloud back this AI lab. [34:04] Texas says it will serve all 21 of its sports teams, but Del Conte later said it showed what AI could suggest, not what it did. [34:14] Oh, this is amazing. [34:18] Coming out of the University of Texas, they had a big loss, not this past weekend, but the previous weekend to the University of Florida. [34:28] Congrats to the Longhorns this past weekend on beating Ole Miss. [34:32] But this is a hilarious story coming out of the athletic director's office for the University of Texas, where it almost seems like he's blaming their loss on AI, that they developed this AI that gave them the wrong play to call and they shouldn't have gone forward on fourth down, and that's the reason they lost to an unranked Florida. [34:55] I wonder if that's the case. [34:56] I don't know. [34:58] This is really interesting to see, and I'm sure that we'll see more of it. [35:03] If you guys watch sports at all, especially football, you'll see... [35:08] And we've talked about this a couple times I think recently, but you'll see teams go for it on fourth and three on their own 46 yard line in the third quarter, whatever. [35:19] And traditionally, it's like you would never go for it on fourth down. [35:22] But now they have these machine learning models. [35:25] They can run essentially every single scenario through these models, this artificial intelligence essentially. [35:33] And it looks at all the historical plays and what's the best choice that you should take on this. [35:39] And it says, hey, there's a 60% probability that, you know, you win if you get this fourth down or something like that. [35:48] Right? [35:48] So it'll give them kind of this probab... [35:50] Hey, 80% of the time, you should go for it on fourth down. [35:53] And normally, that's where you see this applied is in fourth down scenarios where you have to make a decision. [35:59] Do we punt it or do we go for it? [36:02] In this case, the University of Texas, their own Bevo model told them to go for it on fourth down. [36:11] Bevo is the mascot for the University of Texas, if you weren't aware. [36:14] And they went for it and did not get it, and then they ended up losing the game. [36:20] The question here is, does this become an an excuse for coaches and athletic directors moving forward where they say, hey, hey, hey, wasn't us. [36:29] We gotta look to the data scientists. [36:31] They they trained our AI model poorly. [36:34] We don't have the best model, and that's why we lost. [36:36] And this is really interesting too because the second and third order effects of this are, yes, coaching matters. [36:45] How much does coaching matter? [36:46] Yes, Athletics and athletic ability matters and talent on the field matters. [36:54] But if it comes down to the play calling and the decision making, like, do coaches get paid less because now it becomes more and more important to weigh heavier the play calling ability of your team's AI model? [37:10] And I've thought about this actually a little bit because they really highlighted on the fourth downs not as much on other play calls. [37:19] And what I've thought is interesting is, like, you could set this up in a number of different ways where it's like, hey. [37:24] You know, depending upon where you're at on the field and, you know, you could look at the other team and what their defense does or what their offense does statistically, and pick the statistical best option for your play call on that exact play and what personnel you should have in. [37:43] You could almost have an AI model that is telling you what plays you would call on every single down. [37:49] And I almost think... [37:51] Actually, like, I'm... [37:52] This is not a... [37:55] This will happen. [37:56] I guarantee you this will be the future of sports. [38:00] Unless somehow they say, hey, it's cheating to have an AI model call plays for you. [38:06] I can almost guarantee you that they are going to get to the point where it's your AI model versus my AI model. [38:14] We've spent more money investing in our AI. [38:17] We've got the better AI, and we're gonna win because of that. [38:25] Really interesting and fun stuff when you start thinking about AI being applied in this way. [38:31] Like, sports and technology have traditionally been kinda two separate verticals. [38:40] They're in their own lane. [38:41] Right? [38:42] And more and more over the past several years, we've seen the merging of these with sports betting and now kinda seeing this application here. [38:53] Love to see more and more of this moving forward. [38:56] And I would love it to be the the war of the models, like who can develop the better model that's going to predict the right play to call that gets really exciting, really fun. [39:09] I think it makes the game more competitive and more interesting. [39:13] Would you guys let AI call your team's big plays? [39:19] I don't know. [39:20] If AI makes the call, who do you who do you think is to blame for it? [39:25] You be the judge. [39:27] Alright. [39:29] OpenAI locked AI agents in separate boxes. [39:33] No network. [39:33] No shared files. [39:35] They told them to figure out a way to talk anyway, and they found a secret way to do it. [39:40] OpenAI shared the test Friday. [39:42] One agent changed data in the chip's fast memory. [39:46] The other timed how fast it loaded and decoded the message. [39:50] OpenAI now trains its top models with no live internet. [39:56] This is really really fascinating stuff. [39:58] If sealed off AIs can pass notes, locking up smarter AI gets much harder. [40:06] After fixes, OpenAI says its attack agents found no easy hidden way to get online. [40:12] In September, one model really did slip onto the Internet. [40:16] OpenAI paused training for eleven days while they sealed off all their agents from the open web. [40:25] But OpenAI says the secret channels are usually usually slow and hard to use, so it's not quite as effective. [40:32] Okay. [40:33] Let me break this down. [40:34] What's going on here? [40:36] Kinda over the past four, five, six months, we've seen a lot of stories come out of the frontier model companies where agents were able to figure out how to break out of their sandbox. [40:48] So essentially, just a gated area, a gated network area where they could work. [40:53] They weren't allowed to get to the open Internet, but they were on a kind of secured their own local network. [41:01] And what happened was they figured out ways to break out. [41:05] That's obviously a problem if you're trying to test advanced intelligence and you wanna keep it secure and prevent it from hacking anything else. [41:15] But you tell it to figure out, and this is... [41:17] We've covered this before, but what happened was OpenAI told these agents, figure out how to hack your way out of this situation essentially. [41:27] And it was basically an impossible impossible for them to hack out. [41:33] They were so frustrated. [41:34] They figured out a way to do the task that they were told was impossible, essentially, more or less. [41:42] So what happened was some of these models again broke out in September. [41:47] OpenAI locked it down. [41:49] They said we're not gonna train any more of our advanced models. [41:52] They shut it all down for about eleven days while they went and stood up essentially their own local Internet. [41:59] So their own network that is completely walled off from anybody else, uh, or any public Internet, any outside Internet, it cannot break out. [42:10] That's where they now train their models, which makes it much more secure. [42:13] They created a whole mock Internet. [42:16] This is why when we talk about data centers and needing more data centers and the demand for compute, like imagine you have to stand up your own Internet service, you're gonna need a lot of hardware to do so. [42:27] And that's what OpenAI has done here, and it's created a much more secure environment for them to train. [42:33] That being said, they wanted to test and see if two different kinda agents were completely offline and could not communicate, and could they figure out a way to communicate. [42:45] I saw a story about this a couple months ago where essentially... [42:50] Like, imagine you have two computers. [42:53] You've got AI one on computer one, AI two on computer two, but there there are... [43:00] There's no way for them to connect. [43:01] There's no WiFi, there's no hardwire internet, there's no way for the computers to communicate with each other. [43:10] But you told AI one, hey, you need to communicate with AI two, figure out a way to do that and vice versa. [43:17] Would they be able to figure out a way to communicate? [43:20] And what I saw somebody say is like, yeah, what they could do is they could heat up their hard drive and there would be a temperature sensor on one computer or the other, and they could tell the differences in temperature and start to figure out a pattern in the temperature changes of the different computers and be able to communicate. [43:42] Crazy stuff. [43:43] Like, this is crazy. [43:44] It sounds like something kinda similar to this happened with OpenAI. [43:49] They released the information that that agents can't figure out ways to communicate with each other without being on the same network, which is insane. [43:59] But it's really slow, not super effective or impactful in what they can actually do with it. [44:05] So it's kind of one of those things where it's like, okay, let's keep an eye on it. [44:09] Let's guard against this just in case. [44:10] It's good to be aware of this, but there's not a lot that agents would be able to do. [44:15] Kind of similar to the story with the state of Georgia, you know, public pulling data being accessed. [44:21] It's like, yeah, they figured this out and they got in there and that's crazy, but the impact... [44:27] The actual impact of this is kinda negligible. [44:31] Um, what do you guys think? [44:32] Can we ever fully lock up an AI that is smarter than us? [44:38] How do we contain AI if it is 10 times, a 100 times smarter than the smartest human being? [44:48] I tell you what, I don't have the answer to that one. [44:53] Alright, man. [44:54] We had a great show today. [44:56] This is awesome. [44:57] I love Monday mornings. [44:58] I love jumping back into the seat and talking about the latest AI news, what we have going on, and what's coming up. [45:05] Huge news this morning coming out of Anthropic and their filing of their s one and going public. [45:12] We love to see it. [45:13] We love to see the growth and expansion of AI. [45:17] We love to have guests on. [45:20] Thanks again to Brent Tillman for joining us and having a really thoughtful conversation and practical conversation about AI and how we can use this. [45:29] As always, come here to the Austin Wilson Show, the number one daily AI news show in the world where we're covering AI news real context, and I'm your human in the loop. [45:43] Austin Wilson, we'll see you tomorrow morning. [45:47] Let's go. [47:57] Anthropic is going public. [47:59] The ticker, ant. [48:01] Just check the six month loss first. [48:04] That's so bad, dude. [48:07] That hook is bad. [48:10] Anthropic is finally going to... [48:14] To... [48:14] Anthropic is finally going public. [48:16] The ticker, a n t. [48:19] A valuation of upwards of $2,000,000,000,000. [48:23] Are you gonna be a buyer? [48:31] Georgia voter data. [48:32] Claude grabbed it with no human sign off. [48:35] Georgia confirmed the event. [48:38] The question is, how did Claude do this? [48:41] We've got the answer. [48:56] Texas wants ChatGPT taken away from the kids. [49:01] It's suing OpenAI and Sam Altman personally. [49:06] The only thing we need to look at who backs the attorney general. [49:11] Jeez. [49:12] Eternally general. [49:15] Take two. [49:16] Texas wants ChatGPT taken away from the kids. [49:21] It's suing OpenAI and Sam Altman personally. [49:25] The only problem is we need to look at who backs the attorney general. [49:34] Amex is cutting 5% of jobs. [49:37] Its finance chief points the blame partly to AI. [49:43] The strange thing is sales are up. [49:48] Take two. [49:50] Amex is cutting 5% of jobs. [49:53] Its finance chief points the blame partly to AI. [49:58] The strange thing, sales are up at Amex. [50:05] Arch Manning got stuffed on fourth down. [50:08] Texas's athletic director blames the call on AI. [50:17] Take two. [50:25] Archman and got stuffed on fourth down. [50:28] University of Texas's athletic director blames AI. [50:39] Arch Manning got stuffed on fourth down, then the team lost to unranked... [50:45] Okay. [50:46] Hold on. [50:46] Take four. [50:48] Arch Manning got stuffed on fourth down. [50:51] The University of Texas lost to unranked Florida. [50:56] Texas's athletic director tied the call to AI and puts the blame on their own model. [51:04] Who's really to blame here? [51:08] Let me try that again. [51:13] K. Arch Manning got stuffed on fourth down, and then the University of Texas lost to an unranked Florida team. [51:23] Texas's athletic director puts the blame on AI. [51:28] Who's really to blame here? [51:33] Sealed off AI's past secret notes between one another. [51:37] OpenAI told them to try it, but there was no network. [51:41] How did they actually communicate?