10,000 AI Agents Attack One Problem

0:00Hey everybody, welcome to the Daily AI Show. Today is September 9th, 2026. And with me today are Andy and Beth and Boyle Boyd. We have some new topics to talk about. We have math problem drama. I had opened AI and we're gonna get into that. We have US law schools restricting AI. We have Meta Muse, the app. I downloaded that and played with it. I can talk about that a little bit. We have a new image, a new, a better image generator out of open AI. I downloaded them, but we're not download them, but I use to play that image 2.5. We have all sorts of stuff. Stuff about jobs and oh my goodness, there's so much going on today. So we're gonna get into all those things. Oh and Beth is also saying, open AI is thinking about pausing pro subs or just trolling is Tivo just trolling us. We'll talk about that as well. So I feel like maybe one of the biggest stories coming out of yesterday, at least. You would think it would have been Muse or the image generator

1:05from open AI, but no, it's math. We have drama in the solving 90 year old math problems. So there's interesting parts this, but like do either one of you all want to kind of jump in here and sort of talk about what we know so far. It's an interesting one. It's a really interesting one. And I will say that there were hints that a story was coming. So this is the proof that open AI announced yesterday was rumored to be coming. And to you guys have the name of the, it's like Navian something, salient nap. I could certainly find it while you're talking. Navier Stokes Navier Stokes. Okay. And before that, it's 3D. You learn maybe is related to the Navier Stokes. Now there are seven $71 million millennium prize math problems.

2:09Yes. Several of them have now been solved earning a million dollars for, you know, for those who solve them. And only one many open still. Because I thought only one had been solved before. And it was in 2003 by guy, not with AI, just doing his own thing. And this, yeah. I don't know 2003 and like a psychopath just doing it with. Yeah, just feeling like there's a paper. I don't know. Crazy. Anyway, they're the, the these two people who, Andy, do you have their name? I'm sorry. I'm a Tristan Buckmaster and that's from NYU. This is the mathematician and an anthropic of a researcher named Levant Alpole. Right. And we've already started to unpack the tea because the fact that Levant is, at anthropic is part of the sticking point here.

3:13The story coming out of open AI is that they heard the rumor, but not the specific rumor within their training data or their user records. Because of course, these two mathematicians or the mathematician and the anthropic guy were actually doing this in codex. So the files are all user files in open AI. Open AI says they don't access those. The mathematicians said, did you train on our data and open AI was silent? We don't do, we don't access user data when we're doing things, but no answer about training. The open AI side says, look, we were trying to do this in good faith. We reached out. We asked people to come. We asked the Buckmaster, Buck, sorry. Buckmaster. Buckmaster. Buckmaster, to come to a phone call,

4:15apparently Levant was invited to that and didn't come, but ultimately open AI's position was, we don't see a way for us to announce something that includes an anthropic researcher or gives anthropic researchers access to our internal models. Because what created this result, what proved this result was 10,000 internal coordinated agents. Using an advanced model that's smarter and power more powerful than Astra, that does not exist publicly. So better than Astra in 10,000 agents, that's a powerful one to punch. So Buckmaster was offered, according to both of them, Buckmaster was offered co-headlining on the proof. They offered, because Buckmaster and Levant did solve 3D Yula.

5:15It's like, it's like, Egg McMuffin, I don't know. It has as much meaning for me in some ways. But 3D Yula was like a jumping off point for the math, maybe. And so open AI offered, like you could publish that and then we'll publish our thing. You come on, we give you access to the model, but it can't be with Levant because they're anthropic. Buckmaster might tell this up with many of the same facts, but completely different interpretation. Yeah. And set, if you go forward with this, I will go public with this part of the story. And ultimately, again, I think through both of their tellings, it came down to Buckmaster saying, I don't trust you, right?

6:16There's no trust here. And that is the open AI story. Yeah, I mean, they have not earned our trust in the, like, would you cut corners to do this? Well, with, I don't know, the best of intentions, just the, just the like, not even the best of intentions. Look, we're in a cool new ball pit with 10,000 of our friends. Right? Like who, who wouldn't want to play with this? This is amazing. Let's see what it can do. And if you read there, if you read the blog on Open AI, it's sound, it reads, it reads, it's from Open AI. As a, we were inspired, we had heard, we had heard that there was progress. We were inspired to see what could be done with this unreleased model that nobody else knows about. That's not Astra, it's better than Astra, and putting multiple, at this 10,000 agents. Now, there's been reports about like,

7:17well, they spent 22 million on it. They're using external Astra API pricing. Forget that. That doesn't mean anything. That doesn't mean that's what Open AI literally paid out the door or something like that to do this. Maybe they did, but I mean, we don't know that. We don't know that external pricing for Astra has anything to do with this other model, or what it internally like the raw costs are to open it. Right. But, but to add color to the story, Buckmaster and Levin paid Open AI pricing for everything that they did, right? I'm sure they did. Their stuff cost with the best models available to them in those particular moments, right? Like, if... And look, this is where I think the bigger conversation is here on this particular story. I mean, there is the drama side of it, and I don't know who to trust on this one. I'm not trying to take any size on, you know, who knew what where. I think what's way more exciting to me is like the out what came out of this,

8:17or the potential for what came out of this. Really quick before I get into that part, just Beth, what you said is correct. Only one of the seven 1 million millennial prize problems has been solved that were set by Clay Mathematics Institute. It was the PongCare Conjecture, officially solved in 2002, 2003 by Russian Mathematician, Grigory Perlman, who was awarded and declined the prize in 2010. I don't know why you would declines it million dollars, but good for Grigory. For staining his, you know, I don't know what that... There was a more there. And obviously right below that, that's official status right before that, below that is their recent unverified claims and it pulls up what we're talking about. Okay, so that's that part. There's one of the seven, it will take a long time to prove if it is in fact solved by what OpenAI put out yesterday. So that takes a long time. People have said naturally that normally takes several years.

9:18Now in the AI speed of things, maybe that's not true anymore, but I think that's the way it has been is that this is not something something tomorrow. Every all the experts in the room raise their hand and go, yep, they did it. So that's where this story sort of has been trickling out over a little bit of time because I think it was anthropic that came out and said they'd created something that helped speed up the proofs. Like they, there are different ways of proving this. I mean, if you imagine, right, like this was at some point, math that was typed or handwritten or whatever, right? So how those things worked to proof has evolved over time. And there are different systems and I believe the system that was settled on is called lean. And lean has been optimized or like a standard agreed upon.

10:22And lean, like if you're accepting that, then that could be used to prove quicker. If you're gonna try to prove it outside of lean, I think it's different, right? Are we all assuming that English, if it's proven in English, it's proven in all languages, I feel like to a certain extent. So Andy, what we're saying here is that six sigma lean is what has, that's what has solved this from box number 90. So everybody was getting black belts in six sigma. Lean has, lean is what they use. No, it's not the same lean, obviously, to it. Okay, I'll see. Come at me, math people, right? Like I did my best and we welcome you're coming into our community, which is the daily show, community.com, correct us. Absolutely, that's a whole channel for post show discussion, like absolutely. All right. So here's the big look. And before we move on to anything else, here's what I think is the bigger takeaway.

11:22This is my response I had on Ethan Molleck's post about this, yes, I was like hundreds of comments, lots of them are bots, but I was like, here, here's, here's what I hear from this. Because his whole point was what we really are seeing is the more compute we have, the more compute we need, right? That was his sort of like, this could be a big deal. So my point, and I have people who disagreed with me in the comments, but I push back. I woke up and chose violence this morning. And it's been that kind of day. And so what I was saying to people is like, hold on, hold on, hold on. Because people were talking about the cost and the computing stuff, I said, all I'm saying is, look what they apparently did with 88 hours and 10,000 agents. Now, with some unreleased model, now quickly scale that to 100,000 to 1 million agents, point it towards net new problems that maybe our math member, not math, maybe bridge over into science.

12:23What is this telling us? And what my point was, I feel like this is an indicator of exactly where breakthroughs will be in three to five years. And I was saying, what I said in my comment was, will the 20, 30s be talked about as a decade for thousands of years? At some point, the dam is burst open in these longstanding diseases and different problems of the world start to fall because it's moving at such a rapid pace. And if what we're really talking about is, where do you focus the laser beam of compute attention? Who pays for that? Who gets to decide? Who gets the next billion dollars to run? If somebody waves their hand, it says, I have a billion dollars. And I'd like to point it at this because my mom passed away from breast cancer, whatever. Like, is that what we're looking at? Is it the people who control and have the money for the compute? Are they the people in the world that are going to dictate

13:26what problems of the world gets up? That is how it works right now. So there's nothing new about that. Michael J. Fox gets Parkinson's 20 years ago, 30 years ago. He starts raising money and has been a huge force for good. What Parkinson's disease, right? This isn't anything new. We see this. So my point is, where does this go? But it's not, I don't mean this is a negative. I mean, like, literally, what will, if it's 10,000 today, in 2030, what will one million concurrent sub agents all acting towards a common good? What will that get us in five years? We know the models will be better. We know that the compute will hopefully be more efficient, even if it's not more data centers, it's more efficient uses we've seen over and over again. That's what got me excited. Now, some people push back and said, math isn't science about it. And I'm like, guys, guys, again, math is it is it is it absolute? Yes or no, it's easy to track. That doesn't mean there's a bridge here. It doesn't mean there's not a bridge.

14:28We saw math first, we prove it. We bridge to to bioscience, things like that. That's how I was seeing it. So anyway, you didn't hurt you. What's your take on this? I'd say English majors math is science. But also the the subtext of that story is are we in I don't think it's going to be in 2030s. But I think it's happening sooner. But the subtext of that is is this is this the person with the billion says, thank you for bringing the ball this far and then takes it all the way to the to the home and says, look what I discovered. Look, look what I just proved because that's the other part of the story. Andy. Well, you stole my point, which was I agree. I agree that this is an issue about compute availability. These two researchers, these mathematicians didn't have the budget to spend millions of dollars using an advanced model in turn let open AI.

15:31And the question is for many human sort of geniuses who are moving the ball forward, as you say, will it just be that Elon says, oh, I'm interested in that. And he's going to point colossus at it and it's done. And it basically undermines the credit that ought to go to the people who formed the problem and pursued it with, you know, steadfast application of their own human intelligence but didn't have the compute budget our brain is limited, right? To pay for the final completion of it and proof. And so that's I think that's a it's just in the back of my mind it just reinforces that, you know, the goods of our society accrue to the people who hold the capital. Yeah. And you know, Gary's kind of pointed out here isn't it's also just how much faster AI can do things every human's?

16:32I mean, Garrett, yeah, this is a flex. This is a flex by open AI. We're still pre IPO. Who doesn't want to flex and say we use the mysterious internal model. You guys are all talking about how just amazing asher is and we're telling you we're using something better than ashtray and we're solving things in 88 hours with 10,000. You don't think that affects people's investment and IPO status itself. Of course it does. It's a flex. It doesn't mean it's a bad thing, but it's a flex. Well, the other piece of this is pre IPO. There's a certain amount of pressure to raise your value so that your IPO is very successful. Of course. Post IPO. There's a huge pressure to bring in as much profit as possible, right, for your shareholders. That's the internal decision making. And again, this is like what we were referencing earlier on the earlier in the week. It's only Wednesday. That open AI is a public benefit corporation, right, and was previously a non-profit that

17:42was going to create things to the good of all humanity. If the idea behind this, if we were reassured that, oh, of course, when the information is found, when breast cancer is cured, that would just be made free available, right? Because it was, because it's important, like it's really exciting. Or does that then become also, oh, we cured it. But you wouldn't be upset at us for getting some of the profit of that. We put our hard-earned resources in it. And hear all of the other people, like the authors who wrote books or fan fiction online in the 80s saying, wait. Who took all of our effort to train your thing? Now you've found an answer, and now you want to sell it back to us. Like this is the story moving faster, and the B and C and D stories started to come together,

18:49and now they're close to being on the same room. Yeah. All right, Andy. Tell us about you asked law schools, was it? Yeah. This is important. There's this issue around cognitive surrender, when AI takes over a lot of the process of struggling with the development of cognitive understandings and expositions and a center of excellence for all of that is law, right? So that's a really complex concept, written kind of undertaking that you have to have. Deep, critical thinking skills in order to slice and adjudicate all the issues that come into a legal question. So U.S. law schools and importantly, Berkeley's Bolt Law School are now barring generative AI from graded work.

19:50So you cannot any longer use AI to support even the exploration of issues around that when you're being graded on that work. I think there are some applications of AI that are still permitted, but that's not the only one. The Chicago law school is banning devices in all first-year classes. The idea behind this is that they don't want to reject AI, but they want to keep, as has been suggested, in the early grades of general education, they want to keep AI out of the picture until you develop the cognitive skills you get to the point where you have the human judgment before they go into the profession, which is already AI heavy, right? So the existing lawyers who already graduated from law school in a past the bar, those people are using AI extensively in their law practices. So this is all related to the possible downstream and unexpected consequential effects of active

20:59use of AI in every domain of human undertaking. And so I want to just add to something. I've been just asked, question, hey, did we talk about the person who quit? I'm not sure which one that is, but we talked earlier this week about Jacob Pachaki who wrote an essay inside OpenAI. That's the chief scientist at OpenAI, who basically said, hey, AI is moving so fast that we don't really, truly understand how to keep it in alignment. It's going to get out of the box. Pandora's box is opening here. And then yesterday to reinforce that point, an anthropic researcher, Jacob Coxon, said he's leaving the company. And I think this is the one that Gwyn is mentioning. He's leaving anthropic because he doesn't want to contribute to what he sees as a race between anthropic and OpenAI to build systems that will be impossible to control.

22:02And so he said, in a post on act, he said neither company is acting responsibly. They're racing straight to self-improving superintelligence and gambling with our lives. Then the person who runs alignment science at anthropic, Evan Hubinger, said in support of Coxon, the same person at anthropic and researcher who reinforced Jacob Pachaki of OpenAI's complaint about this issue. He said, Jacob is correct here. We really do earnestly believe AI could kill all humans. I personally think it is greater than 10% within the next decade. That's this P-Dume factor. His P-Dume saying the probability that AI will have a devastating and extinction-oriented effect on the human population is at 10%.

23:04That's not a happy percentage to risk. If your risk of cancer is 10%, just throw the dice ten times and you're going to get there. It's pretty amazing that people inside these companies are raising these red flags so publicly. Anyway, superintelligence is a... Maybe the only thing that we could use to solve the alignment problem and we have to get into this circular argument which is, hey, we can't align it. We need more AI to align it and we talked about that. Brian, I think that was your point. Maybe our own help. That is what the conundrum will be by on Saturday. So, oh, see, now that's nice. I just fixed one of our backend settings that literally was just a switch. I think it annoys all of us which is that when you go to pre-share, it would throw it up on the thing and then you end up looking like you're interrupting the other person.

24:06But it's literally a button on the same yard and I fixed it. Now it's sitting where it should be which is I could get it ready without interrupting Andy and then bring up another topic on here. This sounds like you're about to say something. Guys, this is inside baseball. Anybody listening is like, we don't care, Brian. But it's cool. We needed this. So, I'm glad we found the button. Thank you. I think we should point out before you share your thing. These are people's opinions and people when they leave the place, they usually aren't happy or don't feel valued. So, what do you do? You may... I'm not saying this is for all cases, but I'm saying that you... I'm very weary of people leaving and then just talking crap about the company that they were for. Because they probably felt this way long time ago, but why didn't you say it long time ago? And so, all of a sudden, now you're leaving and now you're... You want to say so. But I do think there is some truth to it, but I just... I'm very weary of... Well, for it to say, I certainly...

25:09I think it's an excellent point and we have to take with a little grain of salt the you know, departing, you know, complaints from somebody who's leaving the company. But the thing that I think circles this one as worth considering as valid is that the head of alignment science at Anthropic, who's not leaving the company, said he's right. Discaplaint is right. And but Gwen said what she said in the chat and I said, yes, I feel like this is evergreen content. I feel like that was Ilias thing, right? Like more... We're fighting in open AI for like compute time and direction and that kind of stuff. And Ilias was like, no, no, I need more if we're going to do this in a way that feels like we have control. He left. His thing is safety, right?

26:10And we have the guy from Anthropic who was like, peace out, live a good life. I'm going to go to London and study poetry, right? Like I'm finding that my direction and what I can do is not having an effect. So I think that one... Like we could say it about the situation of the awareness guy who was fired and then wrote that Leopold, Aachenbrenner, damn. Like why I can't remember like, yeah, anyway, it's a good name, I guess. He had context, right? He was fired and there were things, but it does seem that it plays out in repeated conversations. So you may not be able to look at one instance and say, well, there aren't circumstances that affect that, but this is a pattern of safety people saying they're not being listened to, they have concerns, they're not getting the resources, and so they're going

27:13to leave. Yeah. And so a related news item, by the way, let me just throw in this related news item, which is the UN Human Rights Chief called in their global update to the 63rd session of the Human Rights Council. UN High Commissioner Volkhoch Turk called for countries to agree on AI red lines backed by independent safety verification, citing these specific behaviors that we've observed that are rogue AI agents escaping their containment and so on. So this is... You know, this bubbling up to the highest levels and here's the head of human rights. That's what I thought was interesting about this. Head of Human Rights saying, we need to do something here to protect human rights in respect of AI. All I was going to add to it is, you know, if you're leaving a company, wouldn't you want to put your... I don't call it a 15 minutes of fame, but like, wouldn't you want to put your thumb on and they'd be like, I think like, I just would want my statement out there.

28:17This is the reason I'm leaving and here is why. I think we're doomed, right? Or my B-Dume is 10% or whatever your position is, you would want that to be publicly out there. It kind of hits the new cycle a little bit, but from that person's perspective, maybe that's exactly right. Like, they just want that information to be out there, to be known, to be documented, that this is where I sat as far as, you know, early September, 2026. And so like, you know, I agree with you, Garrett, too, as far as like, you know, what are ulterior motives in there and all the things that go along with it. Sometimes these same people are the ones that show back up in two to three months, saying that's why I'm building, you know, so like, is there an ulterior motive there or is it just simply to be on the record? It's a beyond. I'll provide a countervailing point. And that is these people have equity in the companies that they're criticizing. That's that too, yeah. And they're not losing that equity as a result of criticizing it, but they have an inherent incentive not to cripple the company in the process.

29:21So there's a very, in my mind, there's at least I can imagine. A very principled stance being taken by people who are leaving these companies with this claim, because they've already earned their wealth. They've vested enough shares that they're going to make hundreds of millions of dollars. A researcher inside a philanthropic is a high level person, right? It's not just somebody who's, you know, managing this operational system. No, you're right. Yeah, you're sitting on the beach house staring at their yacht and going, I feel like this is going in a bad place. You're like, we've got the beach house in the yacht. Well, the guy that you get the beach house and the yacht. So okay, I'm going to turn to other news here, because not all doom and gloom. There's amazing discoveries happening in the world of science. Jimmy's not here. I can't do it like him, but either way, from our friends over at DeepMine and Google, we have yet another alpha product, which we've heard of many, many times before.

30:22This is alpha genome Atlas, a predictive map of every possible DNA letter change in the human genome. Now you may ask, okay, Brian, but what is that? Well, I've queued up a little clip here from the video that they shared, and hopefully we can all learn together. So I'm just going to play this, everybody can hear it real quick. I kind of try to get it to where I felt like they're explaining it. So here we go. The genome is the instruction manual for itself. Alpha genome is a model that looks at the chunk of the genome sequence and predicts what happens if you make a single mutation to this region. And through kind of modeling the language of life, we understand some of the genetic mutations what they do and where they're hamper not. The question now is how do you combine all these predictions into a single number? The AVI score, the variant impact score, is a single number that tells you how deleterious or impactful this variant is, the higher the number, the more likely it is to have an effect on the phenotypes that just cause disease.

31:22And we have pre-computed this for all 9 billion possible single-ethyl changes in the human genome. One of the core motivations for AVI was how do we go from 10,000 numbers that Alta genome gives you to a single number that you can then use for efficiently prioritizing? Okay. So you know, this is just one of those areas that is super cool as they have done many, many times before, the alpha genome atlas is free and available to those who want to use it for research purposes. I absolutely love these stories that come out of Google DeepMind, especially their alpha series and all the different ways from alpha proteins. We've heard and we've talked about all these different stories. It's very interesting and I love seeing how excited people are specific to those that deal with DNA in genome because up until a certain point several years back, it just, you know, I think he says at the beginning of this video, if you focused on every variant for one second, it would take in the order of tens of years just to get to the end.

32:24And of course, that's just not something that was obtainable by just human processing. And so now we have these, you know, new tools and I don't know exactly where this is going to go, but they talk about it a lot. And DNA and the genome is the story of life. It's the story of us and, you know, how our cells mutate and what makes a difference. And this will eventually lead in Hasla already led to net new, you know, discoveries in medicine that are going through trials right now. So I love these stories. I love again with Google DeepMind doing, Garth, you're shaking your head like, no, you don't. I mean, is this available to the general public? Yes. Have you guys seen the movie X-Men? Yeah. Yes. Do you see where this could be going? I see it as solving a lot of. Yes. I agree, but I worry that people are going to be getting a little wild on it. I listen, we already know that, I mean, I saw this not too long ago and you think, well,

33:25how many, how many families are going to have an opinion on this? I've only, it's never been my direct family, but I've known people who good friends of the family, my parents and stuff like that, who had children with Down syndrome. And we already know that there are ways in the embryo to potentially eliminate screen, screening a Down syndrome and maybe saying that wrong, but my point is, can you imagine all the people who have very emotional feelings about whether that's something they would want? Obviously, I've known wonderful people in my life who have Down syndrome. And so of course, the idea of eliminating something like that. So there's going to be a lot and I only bring that up as saying, like, that's just one story. Can you imagine all the other stories that will come out of this, whether it's freckles or skin pigmentation or all the things that will potentially come out of it? So I agree, Garrett, but like I like to think of that. I'm all for it. I am all for it, but I do like to be the devil's advocate and point out that there are other things that also make the aside effect. And what happens often with these things is it's less the direct choice that you're making

34:29for you where the ethics can be tricky for that, but the unintended consequences, right? We created a wheat that was resistant to bugs. You got more wheat out of your yield. You also like 10X, 100X, the gluten in it. And then a bunch of people discovered they were gluten intolerant, right? That wasn't part of the intention. They were just trying to make wheat that was better that was gave more yield. Is that a bill? 100%. So we'll have to see where it goes, right? They talk about this AVI score. It's a simple number describing the impact of each generic variant of which there's nine million, I think it is, as they were saying. It's called alpha genome atlas. If you want to learn more about that, you can obviously go to deepmind.google.com and there's all sorts of amazing information there. But more importantly, they make it available to everybody. So going back to our, is it only those with compute that can go and make a difference? This is the exact office of that. This is deepmind releasing a tool that any researcher at any level adds access to if they

35:34need it and just think about where that might go. Yeah, I want to reinforce that point and just kudos to deepmind for doing that with alpha fold, just doing all the work and then publishing it so that researchers can now just search the alpha fold database. They don't have to have a GPU farm to do any high end ML. They just have to look up what has been published and shared from the application of that GPU and compute cluster that deepmind used to do that. Now they've done it with the entire human genome. And deepmind has had that ethics the whole time, right? They've, I mean, it's Google, they make money, right? Like I'm not saying their saints. But in terms of the science, they published the transformer paper that was the beginning of all of this for the various companies. They share their research, did a lot in the beginning. Some of it is not so shared, but they are definitely releasing things for the public

36:36good. Like that's still a part of their process. Last thing I'll say, if you want to see something super cool, go look up online because it's part of this documentary as well. I think Andy, you've seen it too or I know you talked about it with Google deepmind, but there was actually happened to be cameras in the little tiny room when Demis Isabis is talking about originally with Alpha Fold and mapping DNA and he does some quick math in his head and you actually are watching him sit at the end of this little conference table, go, why don't we just do all of it? And it's like this really cool moment and he even talks about it with Chloe Abrams in huge of true. She does her interview series and she refers to it and she said what are the chances? And he goes, yeah, it was, they just happened to be filming in the odd, they were doing a different piece and the cameras happened to be on when he is sort of doing the quick math, the paper, the napkin math and saying, now I'm not giving all credit to him.

37:37Obviously, there's a whole team there. I don't mean like, oh, he solved it, but he is that head or has been the head of deepmind over, you know, during the years. So it's a very cool moment and I think something will look back in history over and over again and go, that was one of many sparks, not the only just one of many sparks along a long chain of different discoveries that AI has been involved with over time. And I get excited about the idea of some future students sitting in a college or otherwise class about the history of AI of which I hope they're looking at our show and our transcripts for our part piece of this to see what was going on in any given day. Shout out to our website. You should go to it. And then, but what I'm saying is I get excited about that future student that maybe isn't even born yet, but like maybe this future student is sitting in this class one day and like, I had the history of rock and roll. It's like the history of AI and they're like, in this happened and this happened and watch this video clip. You can watch, Dennis, this office like have this lightning moment and like how, I don't

38:38know, I get goosebumps when I think about silly stuff like this, but it's like, it's so cool to me to watch history evolve like that and think like we're sitting in it and sometimes it's really hard for us to see like exactly what's going on because we're in the fog. But in the future, people will be looking back and going, oh, we can see all the leap points. And this is what led to this and this is what led to this and this is eventually what splintered in led to these five things and hopefully that means that there's a lot of positivity in the world. It comes with all the negative two and that there's always a devil's advocate side to it. So, okay, I want to talk real quick about because I want to make sure we get some, there's all the news stories out there. Meta news, we've already known about news, but actually released an independent app. It is sort of like, it's like, to me, it's kind of like taking approach of like an open claw, but even specifically my claw where they're handling the computer for you there and they have there, they're saying basically they have a VPN. It's all sort of tracked and you know, they make claims.

39:42Mark Zuckerberg had a video talking about how it was going to have end to end encryption from WhatsApp. I downloaded and used, used last night because I was curious. It's supposed to be like your personal agent. It can connect to apps that you give it access to. So as a trial yesterday, I gave it access to my Spotify app that felt like a low lift. It wasn't going to be like, okay, you can see what I'm listening to in Spotify. And it even still had some issues with like what it actually had access to or not, but it was able to come back and look at my playlist and make other recommendations to be okay. It can connect to your calendars and a whole bunch of, let me give you a whole list of apps that it can connect to on your phone if you want to do that. Importantly, it can do encrypted payments. There was just a stat saying that two thirds of people do not trust AI to do a payment visa just came out with that. I said if visa is involved, it goes to 66%, not trust versus like 80% not trust. So visa was, I guess, taking the win on that one, but go to show that people are just not

40:44at that point yet where they're willing to say, go ahead AI, I trust you to go do this. So I had to go look yesterday. It's not connected to my Amazon app, but I had to go look for this particular, this may buy a company called RX and they're called protein bites. And they would have been priceless. There's one, you know, what are the weird things that I've been buying them off of Amazon. And I was like, well, go find them, go see if you can find them cheaper. And so we, you know, it did that kind of thing and went out and said, where it is. And then it said, do you want me to, do you want me to make this purchase for you? I had not connected of Walletude or anything. And also specifically, whereas there is some end to end encryption going on specifically, there's not end to end encryption with your chats. Muse itself makes that distinction because I asked it. And I was like, where exactly is your end to end? Are you just like WhatsApp? And it said, no, there's other parts of it that are encrypted. Mark Zuckerberg on his video talked about how they were going to be coming out with even a higher level of security that you could opt into.

41:45But just know that going into it, I found it, I found it to be a fun chat interface, I guess is the way I kind of felt like, oh, I could see this actually existing 100% like a wechat in China or a WhatsApp for a lot of the world. It would just be integrated in that way. It wouldn't be its own separate app. So I do think it's like pointing towards the future. So how did you compare then to GROC inside X? So if you're in Facebook or Instagram, what does Muse look like in there? And what can it do? Is it distinctly different from the way GROC works inside X? I don't know because I don't use GROC in X. So I can't answer that myself. But what I could tell you is it had access, I gave it, but it had access to my Facebook and Instagram. And I was able to say, like, because I was talking about the Spotify thing. And I was like, you gave me some playlists. I was like, have any of my friends or connections on Facebook or Instagram share to recent playlists? And it came back to me and said, no, not really.

42:47Nothing in the recent. And it would be, all I'd be giving you is random groups on Facebook's playlists. I think the one we talked about earlier that I made the suggestion is actually the better fit for you. So what you're hearing there is like, it still was pulling the context of the conversation in that same thread about Spotify was able to go look at a different app like Facebook, come back and say, I mean, given the two, I think you already have the better recommendation. That's what I would do. Let me know if you want me to open that up for you. So I'm not sure Andy, because I don't have that like full context of the GROC side of things. But you are in an individual unique app. You're not in X asking GROC something. You're in the Muse app. I see that you have the best in the world. That's, I think that's a big difference. And what I've read, and I haven't looked at that, I try not to live inside Facebook. And I do, however, favor Instagram feed. And so I, you know, I have spent some time each day looking at the Instagram feed.

43:49But what they're pitching it as, and this is Alex Wang, the head of the Super Intelligence Group at Meta, he's saying, this is a step towards the broader goal of personal super intelligence. So it is personalizable when you set up Muse, you can create, you can give it its own personality, you can make it your personal assistant. And Zuckerberg has said that this Muse integration into the Meta suite of tools, WhatsApp, Instagram, Facebook, those three principally, is really to have a play against OpenAI and Anthropic who are providing this concept of the personalized agent that you give complete access to your system to in the case of Claude Code or, you know, with, with, with structure, you can decide to contain certain folders and so on.

44:52But that's, that's where we're going is you will personalize one of these things or many of them. And, you know, I try to be agnostic and try to experiment with them all, though I doubt that I don't feel any draw at the moment to the expected use cases for the Meta Muse one, which is like book travel, you know, make emails for me. I mean, they're trying to make, yeah, basically a consumer-friendly assistant. Something that's not technical. Nope. Yeah, and it's a good play. I don't trust Meta enough and I think there's a lot of distrust and I wonder if that's going to not... Which is funny with a lot of people trust WhatsApp owned by Meta because of the end encryption. So I always find that funny when people are like, I don't trust Meta, don't get me wrong. There's a lot of good reason for that. But I always think like they do run WhatsApp. Yeah. And a lot of people do trust WhatsApp. So I'm like, I don't know.

45:53It made me not trust WhatsApp, though. I was like, oh, I just used Telegram now. I won't use WhatsApp. And so, but speaking of assistance, and I know Carl did this and he was going to rig it up. No, it's just that. If you were going to pivot, I just want to read it. Yes, for it. So we don't have to leave it. Yeah, go for it. Okay. I asked about, are you similar to OpenClaw or Hermes? That was my question to me. I just want to read the answer really quick. Ha, yes, we're cousins. Same species, different habitat. OpenClaw and Hermes are the DIY self-posted versions of this idea. A personal agent with its own machine, long-term memory scotch of jobs and talks to you over message apps and does tasks instead of chatting. Funny detail, Hermes uses the exact same soul.md memory MD user MD conventions I run on. Same lineage of ideas. The difference is packaging those you host yourself, your VPS, your API keys, pick any model you like. I'm the managed version. Nothing to set up. And I come built in with built-in app stuff.

46:56They don't have out of the box, like the feed, the goals tab, the shareable pages, and connectors. The trade-off self-hosted means your data stays on your hardware and you're in total control. But you're also the cis admin and securities on you. So it talks about that. And so then I asked about security. And I said, on this version, it's reasonably secure, but not private from meta. This conversation isn't end-to-end encrypted and your interactions with me can be logged and reviewed by meta for safety, debugging, and product improvement. What I know about you also feeds into trainings, meta as AI, though you can opt out of that by going to the settings data controls. On the plus side, nothing here goes to meta's ad systems, even if you're Facebook connected. And you have real controls. You can delete messages and chats. Tell me to forget things, export your data, or do a full reset in the data controls. So there you go. So there's the answer from meta, from Muse about Muse. I thought those were interesting, sorry again, before.

47:57That was good. I like to hear the voice and the clear description of what it can offer you. Yeah, it was cool. I mean, I had to do stuff. I was trying to think of different things to try. You would reflect the computer. And so I was like, oh, I think I'm narrowed down to this particular EV that we're going to, because we're given one to my daughter's with to replace it as our second vehicle in the house. And so I then narrowed down to this. And it's like, yeah, I can be reoccurring tasks. And it's like, I will go and just keep looking for those that particular vehicle in your area and come back to you. It made me a whole dashboard very similar to what you would have gotten from Proplexity Computer. So I'll keep using it. I'll report back. I mean, that's kind of like my JavaScript system. I think what Grockbot does a really good job at. Perhaps I need to go back and check my Grockbot clone I made with Astra just to play with it. But I use, and this is like, every year when we do a Final or Fantasy Football, I, it is a really stressful night for me.

48:59Like really stressful night. It's a really stressful day, because it's a easy way to like the last minute, to plan everything or do my research. So I'm doing my research, do my research. And then I find out I have to, I'll be driving from one soccer practice to home. During the time of our Final Fantasy Football draft. So what do I do? I go to chat GPT and say, hey, at 7.45, I want you to jump into my Final Fantasy Football draft. These are, here's the research that we've done. These are the two players that I want on my team, make sure we pick them accordingly at some point, but not before anybody else takes up. And do my draft for me. And then I said, okay, go. And so I said, it said, yep, great. I was like, all right, let's practice. Let's do a mock draft. I let it do a mock draft, did it perfectly. Last night, driving, can't,

50:01freaking out, because I can't, my phone is not connecting to the draft chat. I can't see what's going on. Luckily, my wife is in the draft as well. And she's like, how are you, I mean, how are you drafting? I'm like, I got, I got that GPT doing this. Don't worry. She's like, it seems to be working. So, yep, did my full draft. I got third rank, my score was like third highest score. And it did everything perfectly. Exact got my two players I wanted to. It did everything autonomously. I didn't have to do anything. I watched it do things. And in fact, one time I did try to draft and then it asked me a pop up question, came up in the chat, because I was sitting there watching it. Did you push the button? Please don't push the button. I was like, okay, sorry. That's right. But it told me not to mess with it. It's like, don't mess with what I'm doing here. I'm on this.

51:01And it was interesting that it did have that pushback. It told me not to mess with it. Yeah. Because it was taking, it was like, I have control. Like, basically, I forget the exact word. Don't mess with me. I run on top of it. Yeah. Yes. I will tell you when monkey fingers are required. Yeah. Monkey fingers are required. Microwmanage me. And so I want to just quickly remark about how different that world is where the AI agent is responding to eventualities rather than just generating responses in respect of a prompt. We have come so far. And it's just amazing to me that these agents that we're working with are so similar to a real human colleague and attentive and receptive. And at the same time, monitoring your reactions and checking in with you. And that's the whole world here with virtual immunologians.

52:03It's new capable of Astra's. It's able to do things, do multitasking. Do things monitor what's going on. Also, be thinking in the background. Do multiple things at the same time. And so that was interesting to see. And it was really quick at knowing where it was at, where the queue was at. Doing research on players like in real time. I could see it clicking on players, looking at the player, adding them to the queue. And then when the Eagles, I was planning on getting it, it had the Eagles defense queued up to pick. Somebody took it out right before me. It quickly shifted. When back to the defense, found another defense, selected them. It was really cool to watch. And I don't think a lot of people realize that this is where we are at. I'm beginning to see that we, the general public is in look what chat, GPT image can do for me. Look at these images.

53:04And I saw that a lot lately. Oh, look at the 80s style. This is stuff we did two years ago. And this is now coming back to the general public being understanding it or thinking it's fun to do. And so I'm like, oh, interesting. Is the general public like two years behind? Yeah. Well, Jude says, take your pause up the keyboard. And what I think this is going is, Gary, you're sitting at the keyboard and it's doing its thing on the draft. And it gets a sense that you're going to get involved into doorbell rings and you go check the door. And it's like, there's nobody at the door. And it's like, oh, no, no. I mean, I just like, I mean, I had to get you away from the keyboard because you were going to mix me up. There's your monkey fingers. And you're like, there's nobody at the door, babe. And you come back to the computer like, that was weird. I mean, that was weird. Ring doorbell that I have connections to. I connected Hermes to, I created an Alexa skill for Hermes so that I could talk to Alexa

54:04and do that when you were talking. I was like, oh, God, Alexa's going to be like, oh, no, no. I don't have the, I have a show, but I don't have the, I don't have plugged in so we can't watch me on the camera. But like, that's the next piece, right? Like, I see you don't do it, Beth. Yeah. Like, I, I've got this. Yeah, my needs are don't, don't send that slack message, Brian. I think I'm out of that Brian. Stop it. Stop. Don't put the keyboard doorbell rings. Go get a sandwich. Aren't you hungry? I am hungry. When is the last time you got up and walked your order of rings as you've been sitting for a while? How about you go take a walk? Say I was going to be protecting me from my, myself in the future here. And I've lost the date in the chat. My kind of shift really quick. My boss brought this up to me. I was like, I was saying to him, I was like, have you, like, Astrid can do anything, literally can do anything. He's like, what do you mean anything? And so I give him a list and he's like, check this out. And he has a friend who was hunting

55:06and has a scope that records and is a heat, a thermal scope. And the scope, basically the recording had this weird blurry image that, or foggy image, a cold image in the recordings. And it was here and it was pulsing. I was like, that's weird. I've never, he's like, I've never seen something pulse in a thermal scope. And then it moved on a different, it was in a different location. So he's like, his buddy sent it to Luke or my boss and said, can you figure out what's going on this? So he asked Astrid to do it. And Astrid tracked through the four videos that he sent, tracked that blob or that cloud, dark cloud, through all of the videos and square, like, highlighted it. And tried to figure out what it was. Couldn't determine what it was. We think it's a portal to more ghost. Because there's no explanation of what it was floating.

56:07It was transparent. It was pulsing. Like it is really weird, really weird thing to see. And it wasn't a scope issue. I mean, that's what it determined, but it couldn't determine what it was. And it was on an Indian reservation. So there's that aspect to it though. Yeah. But yeah, but the fact that it was able to go through four videos, the four videos and track it through the four videos, do different filters on it, do different enhancements on it. And it gave him a full report of what it was going on. And it can, and can conclude to what it didn't know what it was. Well, we don't have time for it today at all. But I will happily say tomorrow how Atlas has failed me horribly. Not Atlas, not Atlas. Atlas. Real Atlas, the computer. Atlas, the model has. Astro. Astro. Astro. Astro.

57:08Astro. Astro. Astro. I may have said Atlas. Sorry, I was even when you said Atlas, India was like, I'm right, Andy. You know, I'm not. It's Astro. Thank you, Andy. Peace. Astro has failed me horribly. I will leave that. I'll dangle that out there. I can't see. You always leave these danglers. It is such a, it has been such a frustration over the last two days. I will gather my thoughts on that and talk about what I attempted to do and where it has failed five times now. It's your harness. Yeah, it is something. It's about to go out the window as it was about the deer. It's, it's beyond frustrating. Considering it was supposed to be this amazing model. All right, we're pretty much at time. Was there any last, you know, I know we sort of backed up on a new, or any last like 30 second new story that it wants to get out? Are we just, we good? Well, do you want to just do 30 seconds on image? Oh, yeah, right. Yeah, so, a shout out. Thank you. Of all the other things that came out yesterday, opening eye released their image 2.5,

58:09which is their new image model. I had it on my, on my mobile pretty quickly. So you'll see on their templates. Last night, I was playing with their template for making thumbnails. Not bad. There's something that was a one shot and I shared it with the team internally. And I was like, this is what it would have been for yesterday show. I think obviously with some guidelines and some do this, don't do that type of stuff. I can definitely see it being valuable. Creating thumbnails has been one of those like we're just really like, I don't enjoy it. I see the value of it, but I really don't enjoy it. It's always been sort of a struggle and bet that I have always sort of ping pong back and forth over the years doing it. But I eventually just stopped because it was taking up so much time. This gave me enough like maybe that what I did was, I was giving it the show summaries and basically saying, here's the show summary, here's the chosen, because it gives me options for what I think might be a good AI title. And then I pick one of those. And then based on that, it's like, go ahead and put it together. And I know well enough that like the words on the thumbnail

59:14should play with the title. It should not be the title and even though that's exactly what we do on our videos, because it's just low lift, right? So this was coming back and saying like, well, if the title is yada yada about Atlas, or alien mind we were talking about on Monday, it produced a clip or a thumbnail, excuse me, and it would say like, what don't we know? Well, that is much better. Whether the layout and all that stuff are sort of on point, I think again, I'm curious enough to go check it out. But I would say go definitely check it out, because there's a lot of different ideas on there and a lot of templates on there. So if you're making flyers, if you're doing these kind of things like you might do in Canva, I would highly recommend going and checking out image. I was definitely impressed. I mean, it definitely was enough for me to say, go spend more time here specifically for our thumbnails, because man-o-man would I love to offload and just through the API once we get it honed in. And Beth has built all these things.

1:00:15And just while it's doing everything else, Beth, right? Have it built the thumbnail too and know that it's quality enough to go out the door. So I'll say what the headline advantage of this new model is, is that you can edit only a small region of that without affecting the entire image. So it's not regenerating the entire image. Now you can edit differentially within the image. Yeah, that's what we do that getting rid of stuff on tables, like plants on tables and stuff like that. I feel like technically you could do it before, but now it works. Right? Yeah, the comments. You should be able to do the individual comments, like pick it and then make the little comment. And that did already exist, I agree. Yeah. So my last thing, T-bo 9 hours, T-bo person who does the resets and works for OpenAI, demand for Astra is unprecedented. We're pulling all the levers possible to sustain the demand, but I've not seen anything like it until now.

1:01:16And we went through very steep growth for four. Priority will always be to keep excellent service for existing users, but we might have to pause new pro subscriptions for a bit if this continues. That could easily be, hey, we want to up our pro subscriber, but I am in a position where I know I'm going to upgrade to pro, but I figured I'd wait until I used all my resets, just to see what I was using it for. So that's why I'm interested in this and I may end up. I went through three resets yesterday or two resets yesterday. Yeah, I don't have any more on my side, but I've got two more to do with the job. Astra. Jesus. So and my question is, if I upgrade is the reset for the upgrade allowance, even though they gave me the reset when I was a plus subscriber. Oh, that is a good question. Five X bump. I would think so. If it's if it's post, I bet you do that.

1:02:17That would be I might pay the. I don't pray now. Yeah, right. And get like, hey, that's a lot more compute if you're able to get it. You know, if that works like that. So that would be that would be pretty cool. I would research it before it's then hate for you to lose those resets. That's true. That would that would actually deeply suck. Thank you. All right. All right. We'll talk about all this stuff tomorrow. We got two more days of the week to have live shows. So we'll certainly pick up a conversation tomorrow. Thanks for everybody in the comments as well. Thanks for hanging out with us. And yeah, until until tomorrow, guys. Have a great day. Bye.