Anthropic Faces FTC Scrutiny as OpenAI Claims a Major Math Breakthrough

0:02 Welcome to the AI Chat Podcast. I'm your host, Jaden Schaefer. Every day I cover cutting -edge AI news and talk with the leaders behind it, breaking down what it means for your life and business. Welcome to the podcast. The FTC is demanding that anthropic executives testify in their AI consumer harm probe. We've seen this in a lot of different industries with social media, and now it is moving into AI. We have banks and Blackstone that are lining up. 63 billion dollars right now to fund broadcom ai chips and these are all going to anthropic we also have opening eye that says their internal model disproved the exponential time hypothesis this is my science corner that i'm excited to cover today on the

0:49 show We have Verizon that is signing a $2 billion AI deal with AWS, and they're raising their 2026 Outlook, which a lot of people are excited about. OpenAI is launching GPT 6 .1 Terra. It's going to be half the price of GPT 6 .1 Sol. And there's a bunch of interesting reasons why that's going to make a difference. Oracle has launched 25 Fusion Claw agentic apps across ERP. HCM, supply chain, and CX. So we're going to get into what Oracle is launching there and a couple other interesting things. Before we get into all of that, I wanted to mention, if you have ever wanted to start a business with AI, create a side hustle, or grow and scale your current business or your career, I'd love for you to go check out the

1:31 AI Hustle School Community. We have 1300 members and I make a new video here every single week breaking down how I'm scaling and growing my businesses with AI. We have a classroom section with over 100 different videos. And my goal basically is for people to join the school community and make an extra $10 ,000 a month from a side hustle or business. If you have a much bigger business, you can grow that a lot. But if it's a new venture, something new for you, that's my goal. We have a couple different courses. Essentially, we show you how to build and grow SaaS apps. We show you how to become an AI consultant. We talk about how I make money doing AI generated music, podcasting.

2:06 And if you already have a product, we go over all of the strategies for organic growth, social media growth, paid ads, offers, emails, newsletters, all that in. And I think we have an archive of over 100 different videos breaking down different tools, strategies, techniques. I would love for you to join. I have. personally created apps that have over 150 ,000 users. And I recently vibe coded an app that just passed a thousand users as a brand new app. I talk about all the strategies in there and a lot of other cool stuff.

2:33 So it's only $19 a month to join. We made this very cheap. If you're interested, go check it out. There is a link in the description. All right, let's talk about what's going on with the FTC right now. So they have demanded the anthropic executives are going to come and testify for their probe. The testimony request follows a 50 -page document demand sent to Anthropic and OpenAI. And of course, everyone knows this is happening right as Anthropic is pitching their investors on their IPO. They've already filed for their IPO. OpenAI said that they're not going to file till next year. So we have these two different timelines going on right now. Anthropic and OpenAI have both each received this 50 -page

3:10 civil investigation demand. They're talking about a bunch of different things here. One of them is security incidents. One of them is safeguards. One of them is marketing. And one of them is privacy. an FTC spokesperson said that individuals at other AI companies have also received testimony demands. They didn't say what companies, but we know for sure we got Anthropic and OpenAI that are requested to come and testify. And we probably have a bunch of others. And I think there's a whole bunch of different things that we'll see here. There's kind of the usual lawsuits that we've seen around like teen safety and different things like that. But I think at the same time, we're also going to see, you know, when they

3:41 talk about marketing, it makes me think of what some people have accused Anthropic of doing with marketing stunts where they'll go and find security vulnerabilities. I mean, it's an interest. It's so tricky because some of these security vulnerabilities are real that Anthropic is discovering and they're telling people about. But at the same time, it's just a lot of people have criticized them saying it comes off as wild that they discover a security vulnerability that requires millions of dollars worth of tokens to go fix. And then they tell all the companies and all the companies have to use their models to go in. fix all the security vulnerabilities. So they make money discovering them, make money from the patching. But at the

4:16 end of the day, people have bounty programs for these discoveries. And these are some really big security issues that have probably existed for a very long time. So, you know, obviously there's a lot of good, but it's going to be a pretty wild investigation to see how much of this is, you know, legit, everything, everything that's going on. Anyways, the demands went out the same day the Anthropic filed their IPO. So it's definitely at a tense place for the company.

4:40 Banks and Blackstone right now are lining up $63 billion to fund Broadcom. These are going to be chips that are going to Anthropic. It's a new special purpose vehicle, an SPV, that is going to buy Broadcom XPU accelerators and networking gear, and it's going to basically lease them to Anthropic. The deliveries are starting in 2027. It's interesting because we have to get creative. Anthropic's already raised so much money. They're about to IPO. Broadcom is obviously a... a company that is very well known. But even with a company like Broadcom and a company like Anthropic, making it just a clear deal for Anthropic to go acquire chips from Broadcom is sometimes tricky because of the way that these deals are

5:18 structured. And so we're able to do all sorts of... creative financial things like spinning up new SPVs to go and actually acquire the chips and lease them. So it's not like Anthropica is buying them directly from Broadcom for basically financial reasons. Bank of America, Blackstone City, Morgan Stanley, all of them are creating this new SPV. And Broadcom is going to give residual value support to it, which is equal to the first lien portion of the debt. The way debt works, the way lending works, the way the terms work, this is kind of the best way for Anthropic to get its hands on these chips. Deliveries are expected to start next year under the Compute Capacity Agreement that Broadcom announced earlier.

5:58 Apollo and Blackstone previously agreed to finance a $35 billion extension to Anthropic's computing capacity. And so now we kind of have these follow on companies that are... that are adding more to this. All right, we have a big one. This is a science corner for me, but basically OpenAI says that their internal model has disproved the exponential time hypothesis. So I found a really good tweet over on X that explains the exponential time hypothesis, and I want to read it to you. It's not too long, but I just feel like this is a really interesting problem that computer science has been grappling with for a super long time. and through our OpenAI is able to solve it, there's a bunch of really interesting

6:36 implications. So we had Greg Brockman that was actually tweeting about having this solved among a bunch of other ones, right? So we had a whole bunch of, you know, math problems that were solved a while back by OpenAI. And apparently there's a whole bunch of new ones that they've now just done. So anyways, there is a really interesting tweet that went out all about this by Ramesh Karnawat. And they said, many of you don't understand how serious this is, so let me break it down.

7:00 In computer science, there's a famous kind of puzzle called the three sat. You're given a long list of yes or no switches and a set of rules about how they must be arranged. And you have to find a combination that satisfies every rule. The obvious way to solve it is to try every combination. But each new switch that you try doubles the number of possibilities. So a puzzle with a few hundred switches would take longer than the age of the universe to brute force, right? So it's like, you know, here's a hundred switches, get them all in the right order and, you know, got to guess the order. And every time you switch one to test if it's the right order, it doubles and doubles and doubles. Okay, so it's kind of

7:36 been this problem where it's just like, it's impossible to solve because it's constantly doubling and you're never going to get it in the right order because it just keeps getting bigger and bigger. For decades, the smartest people in the field found a whole bunch of really clever ways that made this a little bit faster, but every trick still hit the same wall, and that is that the time required kept exploding as the puzzle grew. About 25 years ago, two computer scientists, Russell Impoglasio and Rohamo Patry, were working with Francis Zane, and they turned that wall into a formal idea, and that's what we call the exponential time hypothesis. This is what they proposed, which basically says that this... explosion

8:14 is not a failure of human cleverness, but it's just a law of nature. So they're like, look, no algorithm, no matter how smart it is, is ever going to be able to solve a 3SAT without the time blowing up exponentially. Nobody could prove it, but almost everyone believed it. And I think it became one of the foundations that a lot of modern computer science, it used it in a bunch of really weird ways, to be honest. Researchers basically were using this as a measuring stick.

8:38 So if you could show that solving some other problem quickly could also crack 3SAT quickly, you had shown basically the problem was hopelessly hard. Hundreds of papers, things from scheduling, graphs, logic, a lot more, were all kind of sitting on the same assumption. And so now, fast forward to today, an internal AI model, OpenAI, says that they basically broke it. They said that they found an algorithm that solves the puzzles.

9:04 a lot faster than the hypothesis allows. And it has checked the proof on Lean, which is the famous computer verification system. So the thing that is the most wild is that of all the people that have kind of, or the people that originally proposed this hypothesis, Impalagi, Zee, and Patry, along with MIT's Ryan Williams, they're basically now leading the public review of whether this has actually proved their own idea. So they came up with this and now they're actually...

9:32 kind of checking the facts on it so basically what happens next is if the if the disproof is accurate then a huge chunk of a theoretical computer science has to be revisited and i know it's like it's like theoretical computer science people kind of hates these problems sometimes because they're like you know they're like this problem that has no impact on the world is all hypothetical you know but there's actually some interesting things because basically Because they were kind of using it as like a yardstick or a measurement for how hard like a problem was. Every result of previous problems that said, you know, this problem can't be solved faster than assuming ETH, right? They're like, you know, this problem

10:09 we're currently trying to solve can't be solved any faster than that one. Basically, all of those lose their foundation and researchers are going to have to work out which of those walls that they hit when trying to solve a problem were actually real walls and which of them just simply... they just stopped looking because they thought it would take too long. So I don't think it means that everything becomes easy, but basically the new algorithm is, it's still slow in absolute terms. And I think it's not kind of the famous P versus NP problem being solved, but it means our map of what is possible was wrong in a place that we were very confident about. And if this is the case, then we have to redraw it. So anyways,

10:45 there's a bunch of interesting implications, but it's always impressive to me to see these new algorithms coming out of models like OpenAI and Anthropic. solving these hard problems, I'm excited by opening eyes, kind of push towards solving some of these math problems, some of these like brain twisters, some of these things that we've thought about for a long time. Because yeah, that's kind of what's exciting to me about science is making progress in these areas. So we'll see what the actual implications are. Maybe we revisit some problems that we previously thought were unsolvable, because it would just take way too long. And we find new ways to solve them.

11:17 Because if an algorithm could be created for that, maybe an algorithm could be created for some of these other problems as well. Okay, let's talk about what's going on at Verizon. They just signed a $2 billion AI deal with AWS. They're raising their 2026 Outlook. The AWS framework covers distributed compute for agentic and generative AI workloads. Right now, Verizon is saying for a third quarter straight that they believe they're going to be making more money, earnings per share.

11:40 And so with this, the deal is basically scaling their distributed compute for agentic generative AI workloads. They raised... Their earnings per share guidance to $5 .04 to $5 .06. This is a third straight guidance increase. So Verizon believes they're going to be making more money with this. Their Q revenue this year was $34 .4 billion. It's up 1 .5. And their adjusted earnings per share is $1 .31, which is up 8 .3%.

12:08 So Verizon added $406 ,000. post paid phone subscribers in the quarter, which means you know, this Verizon is growing quickly. And I'm excited to see some of the some of the big bets that they hopefully will take in AI. I've actually interviewed people from Verizon on the show a few years ago from some of their AI labs and some of the projects they were working on. So I'm excited to see them investing more in that and excited to see what those investments ultimately bring. OpenAI has just launched GPT 6 .1. This is their newest model. It's 6 .1 Tera. It's half the price of GPT 6 .1 Sol, which is the midsize model. It now costs a million or $1 per million input tokens, and they're adding computer use and a one for

12:51 replacing GPT 5 .6 Tera. I actually have personally been using 6 .1 Tera for a bunch of different projects, or sorry, 6 .1 Sol for a bunch of different projects. So if... If, you know, 6 .1 Tera is half the price, I could see that a lot of people would get a lot of really good usage out of this. OpenAI said that 50 .3 % on terminal bench for a max effort is up from 21 % for the previous model. So it basically got twice as good on terminal bench, which is cool. It's going to be inside of chat GPT work and codecs. for plus pro business enterprise and education. It's also going to be on the enterprise and app co -pilot. I mean, obviously they'll roll it out to all of those different places. I'm excited about having it

13:34 on ChatGPT work if it's, you know, if it costs half as much when you get low on tokens and you need to go to a lower model. I'm always happy when they're able to do that for less. Oracle has just launched 25 Fusion Claw agentic apps across a bunch of different areas. I'm excited to see Oracle making some big bets here. I've heard a lot of rumors about projects that they're leading. Talked to some people there about different things. So I'm excited for the direction that Oracle's going. This project in particular was cool. Fusion Claw is basically Oracle's execution runtime for longer running agent work. And it has continuous replanning and deterministic execution.

14:10 So it could be halfway through a project. It's like it's OpenClaw, but it's like halfway through a project. And then it will say, look, is the path I'm taking really necessary? Or is it wrong? And then it will replan what it's trying to do. I think that's really cool. Five of the top apps are on the Fusion Cloud CX. So they have service health check and executive briefing documents.

14:32 They have a bunch of like cool little apps that they've made basically. Oracle Fusion Companion lets users work in Fusion applications through support AI tools like ChatGPT. They didn't disclose the pricing for the applications, but they've created a bunch of new tools. And honestly... whether they're, you know, frontier level or whether they're just good. I'm excited to see Oracle jump in. I mean, I want to see Apple get more aggressive in the space. I want to see Oracle get more aggressive in the space. I want to see Verizon get more aggressive in the space. I want everyone to come up with foundational models and also just tools that build on top of it.

15:05 I think it's great for the overall industry when you have a lot of competition and a lot of people working on different projects. So yeah, kudos to Oracle. They are obviously raising a ton of money and making a ton of money from selling the hardware and the compute to OpenAI and everybody else. But I think that there's a space for them to do some of the software, some of the algorithms, some of the AI as well. So that's what I'll be excited to see them working on. Thank you so much for tuning into the podcast, guys. Remember, if you want to grow and scale a business or your career with AI, go check out the AI Hustle School community. I have over... 1 ,300 members in there.

15:40 It is so much fun. If you have questions for me about a business you're working on, projects, what's happening in industry, go join the community. Send me a DM over there. I'll get back to you and I'm excited to see what you build. All right. Catch you all in the next episode.