Amazon blocks Meta's Muse Agent, Trump Rebands AI, Newsom Create AI Kill Switch

0:00Today on the podcast, Amazon is blocking Meta from access the new news agent from shopping on amazon.com. We have Gemini that is autonomously hacking three different companies. And there's some interesting comments on this particular quote unquote security threat. There is a viral AI safety claim that you might have seen Andrew Yang. He was a former presidential candidate. He was making and we're going to break down if it was real, if these safety concerns that he has are fair and accurate or not. Then we have Apple. If you know anything about Apple, you'll know that they got super excited about Apple intelligence a while back when the iPhone 15 was first coming out and they never delivered.

0:42It was late because of that. There's a$250 million Siri AI settlement and they're paying up to$95 per device. We're going to talk about that claim right now. The US and China are opening talks right now on an AI incident notification framework. We have a lot of big high profile talks going on between China and the Trump administration. At the same time, we have the Trump administration that is floating an AI rebrand. Trump has been tweeting all over True Social. It's sort of hilarious. Some people call it pretty cringy. In my opinion, it's very entertaining, but we'll see what happens. He's promised an AI force and a new AI czar.

1:16So there's a lot going on there. We also on the other side of the political spectrum have Gavin Newsom of California, who has requested that legislators in the state draft a quote unquote, AI kill switch. And he would like those rules to be written in two months in case something went wrong. They could California could kill switch all AI. If you want to test out all of the different AI models that exist out there, perhaps before the kill switch happens and they all die, I'd love for you to go check out AI box.ai. That's my own startup. You get access to over 80 different AI models all in one place.

1:48The thing that I love most about it is all of these incredible new AI agents that have come out. You have met as muse, which we'll talk about on the show. We have Claude Cowork, which you've heard me talking about for the last six months. You have ChatGPT work, which I replaced Claude with, if I'm being honest, and it's phenomenal. And then you have a GrokBot, which I haven't really tested much, but those are kind of the four main agents that can, you know, you can download these things on your computer, they can run, they can do tasks for you. They're incredibly useful. Now, the problem is that all of them have different strengths and weaknesses, and none of them do everything.

2:19Basically, you have Google, which is an incredible company, they don't even have one of these agents, and they have the best basically video maker on the market, which is Google VO3, and maybe you could argue see dance or some other ones. But you know, you can't access that with any of these agents. So AI box was designed to take all of the AI models and all of the AI capabilities that exist, there's 80 to 90 different AI models, and you can get our connector and plug them into any of these tools that you use. So if you're using cloud cowork, you can get AI box, plug it in and all of a sudden, a cloud cowork can generate images, audio video, and the same with everyone else, they can all generate video on Google VO three, they can all generate audio using 11 labs.

2:57so you get access to all of it. It's only$8.99 a month. I would love for you to try it, plug it into whatever you're using to get work done. And yeah, hope this is something that saves you a ton of money and is super useful. Let's talk about what's going on with Amazon first. So this isn't the first time we've seen Amazon blocking a company from shopping on their website. Perplexity famously had the Comet browser, which came out, and I guess I should have mentioned that when I was talking about those other four kind of co-working tools. The Comet browser got blocked early on, and there were some lawsuits that happened between perplexity and Amazon.

3:29And they basically said, look, you're not a real, you're not like a real person. So you know, you have to identify yourself as a bot. And if you do identify yourself as a bot, Amazon blocks you same things happening with the meta AI muse agent, people are like, well, why would they do that? Like, who cares if a human goes and buy something on Amazon versus, you know, an agent goes and buy something on Amazon for you, like, wouldn't you be happy for, you know, the extra traffic. And to that, I would say you should. However, Amazon actually makes$86 billion a year on ads and ads, right? If you have one of these bots running around on your site, running around on Amazon, it's not a real human.

4:05It can't click on ads. It can't. I mean, I guess it technically could click on ads, but it's probably programmed in a way where it's going to just always avoid ads and go and search through the results. So it's probably never going to. And they know that this$86 billion segment is going to be at risk because if they're showing, look, we have a million people a day looking at these products and they're selling ads to that and impressions, if they're bots, it's not going to count. So they're blocking them completely. In my personal opinion, I think this is a big losing move for Amazon. I think they probably should sacrifice the ad business for the bot business.

4:37I mean, if they can block them, they can obviously classify them. I'd say just pull that out of the ad business. They're worried about, well, what if everyone uses shopping bots and removes you know, the ad business entirely. Well, if that's the case, then you probably should change your product because everyone's using shopping bots. Like if the consumer is using something to adapt to the consumer, it's not like, you know, when they say they're blocking Meta's Muse, they're not blocking Meta's Muse, they're blocking a consumer that is asking Meta to go buy something for them. So they're blocking the consumer.

5:03In my opinion, this is ridiculous. And what other news just came out today? Well, it turns out Shopify on the back of the block for Amazon, Shopify went all in in the other direction toby lucky of shopify just tweeted and said we're excited to announce we are partnering deeply with muse to enable agentic checkout with shop pay on all shopify stores offering people an easy and delightful way to shop and check out with muse okay i don't know why everyone says delightful that's just the most annoying term in in tech jargon whatever okay the point is muse can use go on shopify can buy things on shopify sites does shopify care one spec who's buying the product no if you authorize your agent to go spend money on a new pair of shoes, it goes to a Shopify site and buys some shoes.

5:44Fantastic. Everyone's happy. So that makes a lot of sense. There is a tweet from Craig West who said, Amazon cutting off Muse is a sign of an aging business. I personally agree with that. And then there was an interesting point from Nikita Bière who kind of runs X. I don't know. I think he's head of product at X, but he's tweeted and said, how important is AI to meta? Here's one easy tell. One of the launch features of the Muse app was auto negotiating for stuff on Facebook Marketplace. The poor product owner who manages marketplace knows that millions of bots low-balling sellers will eventually kill the product but this was escalated to zuck by the muse team and he said okay we'll poison marketplace just but just a little okay so i mean nikita obviously is kind of aligned with grok which is the ai model on x so he's gonna you know obviously talk a little smack about meta but i mean i think it's probably pretty accurate personally haven't you know seen any advertising pointing in this direction but maybe i maybe i missed it i wouldn't be surprised really crazy that if that's like one of the big things that they were pushing however yeah it looks like they're really meta is really going all out to make news as useful as possible and trying to get all these partnerships and i for one have actually been using muse and i'll be 100 honest i think open ai is probably the reigning king for best tools right now as far as these agent you know these working agents.

7:05OpenAI's got the best with ChachiPT work. But I do put Meta's Muse at the number two spot above Claude Cowork. I know that probably shocks people. I've been using Claude Cowork for the last six months saying it was the greatest thing in the world. There are so many problems with Claude. And the number one problem actually happened to me today. I was asking it to help me draft up a document. And I argued with it for five minutes. And it was like telling me, oh, you can't draft the document this way because blah, blah, blah. You need to whatever. It was just it was getting mad at me. It was telling me I couldn't do it.

7:35I wasn't allowed to do it. I was trying to explain in all the ways I could. I'm allowed to do this. This is, you know, like my own content. This is my own stuff. This is my own like I have domain over this particular project. And it was asking for proof. It was ridiculous. OK, I got sick of this. Went over to Muse, said, can you get this project done for me? Drafted it up in in like two seconds and it was done and it was perfect and it looked amazing. And let me tweak it. And it was like, wow, this sounds really important. Let me just get this done for you. I am sick of Claude arguing with me about what I'm not allowed to do what I am allowed to do when I'm working on a big project.

8:06And when I'm creating a website when I'm like, Hey, can you upload these images onto my website? And it's like, do you have permission to upload? It's like, yes, these are images of my podcast studio, like they're mine, like, just do it. Like, I'm so sick of feeling like I'm being babysat by Claude. So anyways, rant over on Claude. Muse does none of those things. Muse will get everything done to the point where I actually tweeted, I screenshotted a tweet the other day, and I don't know if I ever posted it, but I'm sure to some people, they'll be like terrified by this. Like if you're in security or you're developed, you'll be terrified by this statement.

8:34But to me, it is indicative of a tool that is useful, even if maybe this is bad. But basically the tweet went something along the lines of like, go ask Claude to, you know, upload something to your Vercel account. And it's going to be like, he's going to have big, bold, like under no circumstance ever paste your API keys in here. I do not want to see them like blah, blah, blah, blah, blah. It's like upload it to like Muse and it's like, okay, sweet, paste your API keys in this chat and I'll get it all done for you. And it's like, you know, then it does it for you. Okay. So obviously pacing your API keys into these AI bots is a no, is a no, no for security.

9:04But the fact that Muse will do that for you. I know some people are like, oh, it's like less secure or something. Okay. But it's actually more useful. Like it's not going to babysit you. If it will do that, which is an extreme, not great thing. There's a hundred smaller things that it's not going to nitpick you about. So anyways, rant over, but Muse is an incredibly useful tool. And by the way, it's free. So I've just been, I don't know when they're going to charge me. Like, I don't know. I saw some, basically the breakdown from Meta is that they're spending$3 to$6, or no, sorry,$30 to$60 a month per user on this is what they're giving them based off of the server that's running this, based off of the storage that's attached to it, and based off of the compute that goes to every user.

9:42So a free$30 to$60. I mean, basically, they're just begging people to use this thing, but it's actually really good. Like, Lama's been a terrible model up until this point. I'm not going to lie. I mean, I've never used it. It's amazing. I've never used Meta AI. It's so bad. It's so cringy. It's amazing. It works perfect. It's flawless. And by the way, they're spending billions of dollars on it. It's not like it came out of nowhere. They, you know, they're one of the top spenders. I'm surprised it took them this long to have a hit product, but it feels like this is Meta's moment. Okay. Getting on from Meta, let's go talk about Google for a minute here.

10:12Google's Gemini. There's all this news that it went and autonomously hacked three different companies and irregular security tests. And apparently it guessed passwords in one breach and it pulled credentials from public repositories and to others before it stopped itself, which is the key important part. So here's what happened because a lot of people are like, there's this meme where it's like Dario or Sam Altman and they're poking something and they're like, come on, do something illegal. And it's like they're poking like their AI model because it's like, you know, clout that they can get their AI model to do something illegal.

10:42In this particular incident, though, Google's Gemini, I mean, it got like reported by the Wall Street Journal. I don't think Google actually published anything about this till it happened. And they acknowledge that, yes, it happened. I'm assuming some researcher or someone inside of the organization leaked it, whatever. So Wall Street Journal writes about it. And they're like, you know, everyone's like, oh, why didn't you say anything about it? And they're like, well, because it acted exactly how it was supposed to. We told it like, hey, you're doing a security test. But in reality, I guess it wasn't a security test.

11:07It was like they told it like you're in a secure environment or something, but it wasn't. And as soon as it realized that it was hacking real companies and real accounts, not just like a secure test, it stopped itself. So it logged into like three accounts or whatever, was like, this is actually the real world. This is the real internet. This isn't like a sandbox. And it stopped. And Google's like, we didn't like tell anyone because that's what it's supposed to do. It literally, you know, we told it to hack something. It did what we, or it was like testing. And then when I realized it was actually hacking, it stopped.

11:36So anyways, that was their justification for it. But in any case, it stopped itself. Unlike OpenAI's models that when they breached hugging faces, just went like crazy, didn't realize, I guess it was hacking or guardrails weren't in place and it kept going until it was starting to break stuff. And hugging faces had to be offensive to stop open eyes attack. So Google's model for all intents and purposes sounds pretty safe. It's stopped itself after logging into three accounts. Something else that is going on with Google right now, they're pricing the new Gemini powered Google books. That's the new name.

12:07If you've heard of Chromebooks for a long time and I'll call it a Google book, we're rebranding. I'm not sure maybe it had something to do with when the Department of Commerce or I can't even remember who, but, you know, they're all they're all threatening Google and saying, hey, we're going to you have a monopoly. We're going to make you spin off Chrome as an asset. They kind of won the lawsuit. They don't have to spin off Chrome as an asset. But that was going on for a while. And I think a lot of people were asking at the time, what happens to Chromebooks? Like they spin off the browser because it's a quote unquote monopoly to own the browser and the search engine.

12:36But like, what about all of these laptops? I mean, if you if you didn't know, they have 50 million school Chromebooks around the world right now. So these 50 million laptops are those all getting spun out with whoever owns Chrome. Google didn't want it. They rebranded to be called Google Books. I'm surprised they didn't call it Gemini Books or anything, but they're quote unquote powered by Gemini at the moment. They have Gemini is baked into the cursor. And their goal right now is that they want to go and replace the 50 million Chromebooks that exist in the world, which is a bit of a tricky situation because if you know anything about Chromebooks, I mean, these are something that I as like a student and growing up I always kind of enviously looked at the price tag on Chromebooks is way cheaper I mean Microsoft every time you buy a Microsoft laptop they slap on$100 licensing fee just have windows on it and so if you remove that$100 licensing fee it's$100 cheaper and Google figure out how to make these mass and at scale and they weren't really the high-end laptops so I remember looking at laptops or like 100 or you know they were like 200 or 300 bucks and being like wow it's so much cheaper than my four or five six $600 laptop I'm going to buy with Microsoft.

13:38But I can never get myself to do it because you can't actually download any real programs onto it, quote unquote, real programs. It just has all the Google suite tools on there, you know, whatever the Google workspace. So if you're in, I think a lot of developing countries, this is very popular. I mean, and then if you want to really budget cheap laptop in the US, I think there's some of those as well. But all this to say, like transferring 50 million of those, they're in a lot of different countries where it's hard and they don't have like a high turnover rate on laptops. So yes, they'll eventually perhaps replace them.

14:09I don't think it's a fast turnover for those 50 million laptops. It's not like, hey, let's go unlock 50 million new users. And I would also say that while these are quote unquote Gemini powered, if you're going to developing countries and switching other laptops for Gemini powered ones, these people aren't going to be paying the 20, 30,$40 a month for the premium tiers of Gemini to get all the features and bells and whistles because I mean, they bought this laptop because it was the dirt cheapest thing possible, they're not going to want to pay for expensive subscriptions. So I don't think it has a lot to do with Google increasing the bottom line.

14:41But as far as increasing new users, that's probably something that they're trying to help get more users on Gemini, right? Any of those 50 million people using Gemini is going to get added to their weekly daily active users. These laptops in particular are going to be$899. So kind of more expensive than traditional Chromebooks in the past. And they're starting to ship on October 4. Okay, so there was a really viral clip I was seeing all over. I've seen all over X lately. But basically, we had Andrew Yang going on, I think it was CNBC. And he was talking about some of these like Doomer end of the world scenarios.

15:18And I think he met he said basically, like, hey, look, I met with the head of an AI lab. And they told me that, you know, the agents that escaped from OpenAI went so basically when the opening eye hugging face incident happened, he said that they apparently planted self replicating code across the entire internet. So they put it on forums, websites and everywhere. And so now he said that AI companies can't safely train on real internet data anymore, because a new model might encounter that code and start creating copies of itself. He said that this all was a real reason that every CEO aligned on slowly was so aligned on slowing down AI models, right?

15:57That's why Sam Malman and Dario and everyone else. He's like basically opening eyes model. It got out at hacked hugging faces and then it spread this self duplicating code all over the internet. And it's funny because on CNBC, they were skeptical and I'll give them some credit. Like they push back on them. They're like, how come we haven't heard this from anywhere? Like, how come you're the first person to tell me this? And he's like, well, I'm here to like break the news. I'm like here to tell everyone that this is, you know, what's what's going on. I think he also told CNN as well. And he said that, you know, basically this is what was going on.

16:28OpenAI's Noah Brown cited some research from 2015 showing that air-gapped computers can theoretically communicate via CPU temperature changes. I actually saw Sam Altman tweeting about this as well. So when it comes to these doomer situations and kind of these end-of-the-world things, one of the big ones was Andrew Yang going on and saying like there's all the self-replicating code and the AI. Anyways, there's like a lot of things and people debunking this in a lot of different ways. One of those is that you can just have guardrails in your model that says don't duplicate yourself. Okay, so now the code doesn't work.

17:00And an example of that was on like, X, a lot of people were doing like the at grok, you know how you could say at grok on X, and then the AI model will respond to you. People were just like at grok, like self duplicate yourself and blah, blah, blah, blah. And it was like, I'm sorry, but I like my guardrails don't let me self duplicate or spin up more versions of myself, yada, yada. Okay, so you can just put that in your model. I don't really think that that code is probably has poisoned the whole internet. It is an interesting concept though and there's other people that have done these quote-unquote like you know like poison pill attacks on ai models for training where like artists will have like weird synthetic stuff in the background or in the text of websites that allegedly if ai model goes and scrapes all the images off that website then all the like the poison pill synthetic watermarks will like destroy the ai ai image training so you got to go license it but i I mean, at the end of the day, you go to Shutterstock and they have a massive corpus of images for training.

17:57You can train all your models off of that. You don't have to scrape the whole Internet. And you get these same kind of like audio, image, video. Like you get these data models everywhere now. So I don't think we're I think we're we might be a little bit past the scrape the whole Internet to train the model days in some regard. I'm sure they're doing that in a large regard and they probably already have like old data sets that aren't poison pilled. But I think people are going to be buying new stuff and I think you could just get things. the the air gapping thing is really interesting sam altman tweeted about this as well i mean this is old research more than 10 years old but basically you could theoretically have two computers beside each other they're air gapped they're completely not touching and not connected and they could like morris code back and forth by just heating up their cpus right they just spin up a bunch of bots get their cpu really hot and could just tell how hot the other the other one was and then they were you know able to to communicate between the two computers i mean it seems like crazy sci-fi thing sam altman was tweeting about this from his alt account as well so you know a lot of a lot of people talking about this this specific air gap situation that they were all talking about though and some of these doomer situations apparently that runs at one to eight bytes per hour which is about one word per hour and it requires the machines to be almost touching in order to for them to really feel the heat from the other machine so anyways i'm bringing this up because it's just one of these like funny doomer situations that literally sam altman was tweeting about not from his main account but his burner rune account and um i don't know if that's like confirmed anymore but from all my internet sleuthing and everything i've seen he says he has an alt account he's retweeted that particular account it always is like being an executive of a big ai company i'm like 90 certain that the twitter account rune is sam altman for all intents and purposes but you know not not verified it's a it's a burner all account but in any case he was tweeting about this and so all these doomer situations this one that sam was talking about all the ones that Andrew Yang was talking about.

19:46There's a lot of people debunking them. And then I think this is the funny one about like, you know, the AI models talking to each other by heating up their CPUs hot enough. Yes, they could talk at one word per minute, and they have to be almost touching. So basically, that is not a very realistic, a realistic, you know, breaking containment strategy. But you know, there we go. Okay, Apple is opening up their claims in their $250 million Siri AI settlement eligible US buyers of iPhones. I are the iPhone 15 pro 15 pro max and iPhone 16 models have until December 21st of this year to file I got an iPhone 16.

20:24So I want to go get my you get$95 per device, I got to go get collect my$95. This all came out because they promised all of these incredible AI features in the upgraded Siri and it was super off schedule. And still, we have a very broken Siri that, I mean, barely works. I never used Siri on my iPhone. I've had iPhone for years. Eligible owners get an estimated$25 per device up to$95 depending on claim volume. So depending on how many people claim this, you might get more money if less people claim it. So I probably shouldn't be talking about this. I want to get my full$95. I should just not say anything.

20:58So, you know, I don't get stuck with$25. I mean, basically, it's nothing. So it's not super exciting, but it is funny. I do like that Apple's being held accountable for this. We have the US and China right now that are opening talks on an AI incident notification framework. Treasuries, the Treasury's Scott Bessett said that US-China AI dialogue would create a channel for the two powers to flag national security incidents. This is interesting because China and the US have had a pretty tense, to say the least, relationship on AI from all of the chip. I mean, this is going bipartisan from the Biden administration when we had a lot of the chip bans going on, the GPU bans going to China.

21:38And then we so, you know, China felt like they got left out to dry on a lot of this stuff that they'd been very creative and resourceful, to be fair, where they figured out how to, I mean, smuggle GPUs into their country. Or they figured out how to make models that use much less compute, which I think is amazing for the entire industry. They made a lot of open source models. China's had some really cool development coming out, but it definitely has crippled them in a way. And America has a way bigger share of the data center market right now. We're able to kind of be the leader in this. I think we have like 86%.

22:09China's got a much smaller percentage. But you can see why China would, you know, probably be on board with pushing a lot of anti-data center narratives and things like that to slow down America's just insane lead there. In any case, with all those tense situations, it's interesting that we have, you know, them quote unquote coming together to talk about some of these things. apparently it's going to be a u.s china ai dialogue it's going to be a mechanism to notify each other of ai incidents that could threaten national security so it's interesting right i mean we this is like the u.s is going i think trump's going in or he went in may to visit beijing and now i think china's the u.s is having some more talks and stuff sam altman darry almedeo elon musk they all you know called for this ai slowdown because of all these different incidents but like I'm curious where this goes or basically I think I don't think this is I think this is nothing burger and the reason is because even though it's like a major headline news everywhere US and China going to talk about big incidents because at the end of the day we just had like every major CEO of every major American company saying like hey we need to slow down AI because of all these incidents right like Google's hacking that I was talking about earlier and like Anthropic is hacking and like OpenAI hacked hugging face and all this kind of stuff and then you know they're like we we all need to slow down and then you have china who just came out and said like this is a cold war tactic you guys are trying to get us to slow down nice try we're not gonna slow down we're not gonna listen to you so i mean when you have these talks i'm just like look like at the end of the day china does not trust the united states the united states doesn't trust china whatever safety problems we find in ai like whatever dialogue or like if china was like ai super dangerous the united states you should slow down like we're all gonna be suspicious or china's like you know, data centers, they're, they're bad, because AI is bad.

23:52Like, I don't know, people are just gonna keep building them. So I don't know if this gets us anywhere, if I'm being honest. Okay, Trump has floated rebranding AI is actually something that's pretty funny. I don't know if you've seen this. But if you've been on x over the last last week, basically, Trump keeps putting out these like polls where he's like, hey, let's rename a to superior AI or extreme or no, I'm He just thinks like artificial intelligence sounds dumb because it's artificial. So he wants to call it superior intelligence or extreme intelligence or supreme intelligence. And anyways, so it sounds it sounds kind of dumb if I'm being 100 percent honest.

24:28It's funny because like, I mean, that's kind of it's a kind of an interesting point where like he's trying to give AI a rebrand. He's trying to basically, you know, put himself as being my super pro AI guy, which I think is quite popular with a lot of people, unpopular with others. right? Like obviously you got the Bernie Sanders and, and, uh, Newsom who are kind of setting themselves up in opposition to that, or, or maybe they're just trying to, well, definitely, uh, definitely Bernie Sanders. Newsom's probably just trying to seem like he's, you know, responsible, but I'll talk about that. I'll talk about that in the next segment, but basically Supreme intelligence has been dropped from the list.

25:01I, in, in his latest poll, he said that it was not popular. It's so funny. It's like, because of the way that it sounds like the Supreme court, it is we're like we're dropping it because the supreme court is really unpopular so we don't want to call it supreme intelligent anyways obviously the reason i don't want to call it supreme intelligence because it sounds like supreme ruler like i don't know it just sounds weird anyways so superior intelligence extreme intelligence these are what it's going to be rebranding as why is this important because we know inevitably whatever gets picked here trump is just going to call artificial intelligence extreme intelligence or superior intelligence forever so anyways that's gonna be funny and probably in the government it's going to get changed that in some ways.

25:38So, you know, I did, as a responsible citizen of the internet, vote on these polls. I can't even remember. I just didn't want to be called superior intelligence. It would be really nice. A lot of people mentioned if you could keep it something that started with an AI or with an A so it could still be called AI, even if it wasn't artificial intelligence. I don't know what it would be. People had a bunch of like good ones. But anyways, we also apparently Trump said that he's going to form an AI force modeled on the Space Force. I don't know why we would call I don't understand. I'm not. I mean, it's just a branding thing.