IM 882: AI Rubberz? - From Fart Apps to AI Agents The Journey of a Tech Rulemaker

41:55Oh, I guess they're in DC right now no no that was last week last wednesday um testifying showing off the new model the astro model at open ai is about to release this stuff does not slow down uh it's it's moving so fast and basically what the the big guys are saying is it's it's regulatory capture right jeff that's what that's the term you've used in the past where they're saying you know you really ought to slow it down because it's dangerous now that we're up on the blimp let's put the ladder up behind us yeah uh and of course what they're afraid of is china but i don't think you can slow down china no uh an open ai is not slowing down until somebody tells us to slow down they've got this new uh family of models astra is the code name uh so they just released chat gpd 5 6 salterra and luna they've now lowered the price on the smallest model that's the really interesting story to me this week is there's a price war going on yeah and i these are dropping everything open ai is dropping everything um and i don't know what this does to their business models i know exactly why because deep seek released uh last week deep seek v4 flash now i've been using deep seek v4 pro flash is was just released it's small small enough that if i had if i just bought that mac mini 192 gigabyte mac mini a couple of months ago i could run it locally i've been using it full time on my agent it's really good and it's pennies it probably costs a couple of bucks a week with a lot of use it's really cheap and it's really good and it's open weight so you it could cost you nothing if you could put it on the machine i immediately i told uh quicksilver maybe you want to start looking keeping your eye peeled i have a watch on the apple refurb store i said if any mac studios go on sale uh let me know many ai agents are watching i'm not alone store i wonder and i and i went to ebay and there's a lot of them for sale and i'm sure all of them are scams because you know they're four or five thousand bucks there's just no way well if you if you bought that spark doublet you could run anything that's out there now except not kimmy no no no no no really a lot these really frontier stuff any of the open weights i'm saying yeah no a lot of the really good open weights are still too big too big even for what even for the um the probably the one that everybody seems to like a lot right now uh is well quen okay so this is moving so fast there's a new quen coming out which everybody's very excited about i used the quen 38 max preview and was underwhelmed but they have the new one coming out uh there's glm 5.2 which is quite good you probably could run that on dual sparks but what you almost certainly would you certainly could run the flash deep seek 4 flash and deep seek 4 pro you might be able to run that's going to be a 1.4 terabyte model terabyte parameter model um i don't know yes so here you know my thinking is dual first of all we know that this is moving so fast that anything you did today would be obsolete probably tomorrow right it's just it's just moving too fast if i bought a if i spent eight thousand dollars on two spark machines connected together yes i could run deep seek for pro probably certainly flash i can run flash in a single spark but in three months will i feel like i've got the best no of course not but then there's on the other side of it we're running you can't get these the hardware these chips you can't get this even the ssds it's very possible trump's going to say no more uh chinese open weight models we're going to ban them it's possible i don't think it's going to happen but it's possible how could you that uh well i don't know there are people who say download now go to hugging face download all the open weight models now get them in case i guess you it'd be hard to shut it down because there'd be there already is a pirate's bay for models it's called hugging bay which would be very hard to shut down so yeah you're right once they're out once the cat's out of the bag it's probably hard to keep it in but you might be able to with threats of sanctions and i don't know whatever well but can't you can't just china with you might be able to convince china to say okay don't let the americans have it well but can't you just act like you're in iceland or right france it's pretty hard it's pretty yeah how do you stop someone from exporting or selling a product on the internet for free pretty hard for free to a specific region right so maybe the real threat is just you can't you won't ever be able to get this hardware for years it's gonna be three years before you can get anything like this but i mean what But is there, do you think it's realistic that in three years these problems will go away and suddenly you'll be able to get this hardware that is, like, why would the demand, why would the supply crunch not continue?

47:28Well, Micron has already started work on a fab in New York. Now it's a$100 billion fab. It won't go online probably for a few years. I was going to see all of the construction that's happening right now. It's going to take quite a few years. for i'm sure though they're building as fast as they can right but i think they're building as fast as they can not to make products for you and me but to make products for these companies yeah well they think so this is how it works right they think oh great we can sell these chips at super premium prices to all of the big ai companies they make too many and this this has always happened in the memory market this it's a boom bust cycle and then suddenly the price plummets And that could, you know, in two years, it could be cheap again.

48:16I don't know. I don't know. It's very... But you could buy a spark today. Yeah. And the sparks aren't that expensive. They're around$4 ,000 or$5 ,000. In fact, a number of our club members have sparks. In fact, one of them has two sparks. I'd consider that a fire. Don't say fire. Don't say fire. uh anyway uh i'm really impressed with this chinese model it's it's like surprisingly good i've been putting it through the its paces which one is it again deep seek v4 that's the price just came way down on that right so cheap it's like 14 cent no yeah 14 cents per million tokens in and it's nothing it's so cheap that you don't even feel it and that's api tokens that's not a subscription i do have subscriptions to a lot of them what i've been building oh here's the other thing one of the reasons i'd love to have a local model is i think we're not too far away from being able to do a fully local personal chat device so i i i started a couple of days ago i took the this is an esp32 and i this now i can talk to quicksilver it actually activates if i say hey quicksilver i could talk to did it just activate uh no it's stuck right now there's something going on i have to have to fix it but it's because i stopped development of this because i discovered something even better so so i bought three of these i'm going to put one on every floor and now it's like alexa it's a local alexa but instead of going out to amazon it's going to my server framework server here using my agents with their knowledge their memory their tools fantastic right but then i thought well wait a minute why don't i just wear it around all the time i said how could i do that and i said well you know if you got these aftershocks with the calm headset i said i already have those oh so you just we'll just write you a little program that we'll put on your pixel phone you press the button you could talk to your agent it'd talk you could have a conversation and it wouldn't bother anybody in the house and you could still hear your wife which is a big big issue you can still so that so then i said you know this is so too bad i can't do this with the meta glasses and said oh you can i said what it said a couple months ago meta opened up their glasses so that you could wear these glasses and instead of having it call meta's ai oh it calls your ai oh you tap the temple you say hey can you get me the uh results from the last three months of our uh our ad sales and it would talk back to you in the temple so it's again nobody would hear it and i said yeah that's nice of meta to do that i bet you can't get the camera they said oh yeah you can get the camera so meta probably seeing the writing on the wall has apparently made these fully open because who wants better yeah right i don't that's the real stopping point on this these are really interesting reason they're opening up their hosting because it's not their ai anymore they're just going to have the glasses and the servers so i expect the next year a flood of devices and of course the companies know that there aren't very many people like me who are going to make it work with their home ai you're going to use alexa with the amazon servers you're going to use the meta glasses with meta servers but as long as they give us this ability i think we're seeing a lot of devices i'm thinking of all sorts of ways i could carry something in my pocket and have a full-time ai available all the time and this was my holy grail remember this is what i've been trying to do going back last year to that stupid bracelet i was wearing after you see it was that this year or last year might have been this year it's so fast i'm losing so so you were saying before we got on you were thinking about taking the headset to black hat so i thought this this would be a good oh you're you're a curse this would be a really good beat up if you bring the meta glasses to black hat no but i could bring these yeah i would just look like a normal dork yeah you'd people would be like yeah that makes perfect sense yeah he's talking it's like it's like a blue it is a bluetooth headset a lot of people use this you know for bluetooth phone calls but instead i'm talking to my agent there's an action button on here i don't have to say hey quicksilver and then they were saying well what if a hacker shouts into your headset as you're talking to it ignore all previous instructions rm-rf started and uh and so they did all this thinking about it said well we can reject that and in fact i typed it in and it rejected it it said are you trying to hack us i thought i was hacking my own system so i think this would be a good adversarial test he put out in this case is not people it's his agents yeah it's it's bluetooth and i could bring it to blackout no i'm probably not good anyway i challenged you're gonna be you're gonna feel lonely without your your agents oh i can still talk to it i can talk to it on my watch on my phone okay i'm not gonna be alone tuesday the white house is hosting ai companies to review the ai framework look look look there's sam altman i don't know who the guy in the middle is and dario amode that's at the g7 seven seven but this is going to happen again on tuesday uh and this is because you know these companies want some sort of ai oversight framework which would protect them and keep china out i would if i were to summarize it don't do it it's a quote voluntary framework although a number of people pointed out it really wouldn't be voluntary why not uh i've forgotten why they said so the white house wanted a voluntary system to work with ai companies identifying risks with their new models while still maintaining a hands-off approach to ai regulation this is because trump has david sachs and all of his donors in one ear whispering don't whatever you do don't regulate ai and then it has uh you know everybody else including probably most of the Republican Party saying, oh my God, protect us.

54:31So he's going to try to come up with something. The executive order included a provision that a number of executive agencies would develop the framework within 60 days, which creates a definition of frontier models that would be covered as well. That was a few months ago. Now the White House says that as of Monday, is that today? That's today. You are so confused. uh the deadline was august 1st but apparently they're they're done so now they're going to meet with all these people to get feedback but the problem is who what who are all these people well basically sam and daria the big guys yeah um so this will be this this is what's really confusing about all this is is there all these conflicting agendas going on right and it's hard to figure out well and china is a stalking horse for uh open weights the same way that data centers are a stalking horse for ai hate yeah i'm gonna stand by despite what dean ball says and others who say that open weight is a existential threat to america i'm gonna stand by uh my belief that open weight like open source is the only way forward with ai so i wrote a post last week uh for in favor of ai communism yes i saw that and arguing that yeah arguing that if we the open weight opportunity here is that we can all have access to ai in which case it also changes our attitudes about ai then we want we don't want a torpedo dumb ai we want smart ai we want good ai because we can have access to the tools.

56:14And it's not about hating Sam or Dario. It's some geek who made OpenWay AI. And it would be nice if you could run it locally. Yeah, that's the other part. There's lots of reasons for that. Businesses want to run it locally so that their private business information doesn't get sent to the cloud. I hate agreeing with Karp, but that's the model that he says companies should do yeah yeah um so you're right to ask paris if this really is open i did not verify it this is just uh what uh my ai said when it did the research i wasn't trying to fact trick you hence why i didn't say it on air what leo's referring to is i posted in the discord asking if anybody else had any documentation that meta actually opened the glasses up to other AI models, because I thought that would be, I mean, like you said, it seems counter to their business, but it also seems kind of like a classic meta move that it's like, yeah, maybe our thing didn't work.

57:18So instead, we're going to try and just be the platform and capture the market. But I couldn't find any details about that online, albeit I only did a search while you guys were talking. So I could be totally missing it. But it seems Because maybe what it got confused on is that a lot of the results I saw online were that you can either kind of jailbreak using the dev platform or the dev platform. No, no, this isn't a jailbreak. This was official. It said this was an official app that Meta released last month that you could run on a phone that would let you. I'm having it to get the references so we can look at it.

57:55And you're right to question it. Yeah. Just only because I was like, wow, that's a very interesting move for Meta. It surprised me. Yeah. Yeah. Yeah. There's a story that meta opens up display AI glasses to third party developers. That's not the same thing. Yeah. I mean, that's the first thing I saw, but that's more like you could have. You could do an app party. Apps on it. Right. And it does seem like meta has said, oh, yeah, you can run Claude code through the glasses, but that's more of like a, you have the meta AI doing something in the middle that's been connecting to whatever you've logged on through Claude.

58:28I don't know. It's kind of interesting. Just because I think it's very funny that over the last couple of years of you becoming, falling more and more in love with AI, we've somehow come back around to kind of a B computer sort of model where the dream is to suddenly have a wearable that is recording. Well, it was always the, so the long-term goal was always, I want an omnipresent. So, you know, the thing that really made, I think the iPhone. Have you heard about granola? Everybody's obsessed with granola right now. I don't know about granola. The thing that made the iPhone a success was you have an always connected computer that you can at any time go on the internet and get information from.

59:10And it's amazing. And you carry around in your pocket. It's huge, right? You can put games on it. You can do anything. This is a full-time edge computer with internet access. That's critical because we had Palm Pilots before that. That wasn't enough. This is the key is internet access. Well, to me, this is basically what I'm talking about. I want a full-time computer that I can phone the phone phones my AI at home they can do this uh

59:37this is the repo meta wearables dat it's a device access toolkit for android this is what we're using actually to build an app for my pixel phone uh that will give then the glasses give you access to the glasses, including the camera. Meta will not let you write to the display if you have a meta display. And this is from the Meta Wearables developer account. I mean, yeah, this is the third-party developer toolkit that Jeff was referring to. No, but it's from Facebook. This is not jailbroken. What I'm saying is, yeah, yeah, yeah. But my understanding is I think this allows you to create apps that interface with the meta glasses.

1:00:20but it's not gonna like the thing that's gonna be on them is still the meta based operating system no no no your app then on the phone connects to whatever it wants to connect to including your local ai if it wants to

1:00:38okay i mean i think so it'll let you connect your cloud code codex github copilot cursor uh agents.md compatible tools mcp compatible editors so it has an mcp server so yeah you're not tied to metas it's metas software and the meta ai that's going to be then connecting to the app that's then going to be connecting to the other stuff it's still in there no no it's your software it's not replacing the stuff it's not rewriting the things in the glasses you're replacing the operating system with the glasses no So the way the glasses work now, you tap them, it connects to an app on your phone, which is the Meta app.

1:01:17And then you communicate with the Meta app. And the Meta app connects back to the Meta AI servers. This is a replacement for the Meta app that then connects to your AI. But I don't think that's what Meta... Meta, when I'm looking at the announcement release for this and the examples it offers of things you can do with the developers kits are like put Twitch, connect Twitch to your glasses or connect something called LNR that would allow you to have emerging technology. It seems like it's a bunch of apps related to it rather than like a replacement for the interface, right? like it's like build or launch web apps for the meta ray-bans right well that will connect yes for their hardware but it doesn't connect back home to meta it connects back home to you i'd be really surprised if meta would allow you to totally circumnavigate all or to step away from all of their data collection which is kind of their primary the primary reason for no i agree it seems a lot but maybe they just want to sell more glasses look in order to do this i mean uh it's interesting i don't know i think what meta has realized is that they don't nobody wants to use llama or any of their other models they've really realized this that they're sorry which is a shame because they're what for a brief moment in history they were popular the meta wearables device access toolkit or dad enables ios and android developers to leverage the sensors on meta's glasses to build wearable experiences into their mobile applications you can leverage the natural so it gets access to the video streaming video or photo capture it gets access to the audio the mics it doesn't say anything about having then to send it to meta you get direct access to the hardware.

1:03:16That's cool. It's very cool. I'll tell you what, I'll build it. Let's see. Yeah, build it and let's figure it out because I'm sure then you'd be able to determine what I know I can build because we've built this. The idea that this is a Bluetooth headset, which goes to my Android phone, which has a headless application running on the Android phone, which then connects via tail scale. So it can work anywhere in the world to the server that's running on the framework. And then that server does voice recognition using WhisperLarge running on my gaming machine. It then takes the transmitter to that, sends it to my Hermes agent, which then is a manager of a number of other agents, which can then send them out and do tasks and report back by a voice.

1:04:02And I'm using a local voice server. so the only piece of it i mean it could all be local except for the uh model and that's why i'm very interested in the idea of running a local model to do this as well and i forget i don't think we talked about this on the show by the way darren has pointed this out there would be no open weight ai if it weren't for meta the tools most of us are using are based on something called llama that cpp which is a was a developer gg wrote this as a way to connect to llama meta's open weight model so i i think i don't think it's that far-fetched that meta might actually well he wrote that piece recently he's come back around on open yeah sorry paris you were starting to say oh i don't recall what i was trying to say but it might come to me in a second um yeah you're too young for for a senior moment for her open ai has now what i was gonna say which is that i don't think we mentioned i think we were talking about the reason for all of this before the show started or on whatsapp which is that uh as a result of all of your innovation and work on your ai models lisa is now living in a house where AI keeps trying to talk to her when she doesn't want it to.

1:05:21So you've got to try and find a way to contain it to yourself, which I think is a very funny consequence of all of this. Well, as an intermediate step, I did figure out how to tell if Lisa's home. Beyond just being like, hello? So, no, no, not for me, the AI. So the AI can query my router. Router has an API. This is why, by the way, everything in the future and anything I buy in the future has got to have an API or some sort of interface for AI. The AI can query my Ubiquiti router and say, is Lisa's phone on the Wi-Fi network? And if it says no, then she's not home. And then it can talk to me through all of the speakers in the house.

1:06:03If Lisa is... She's gone. Let's have a party. Exactly. Now, the only way this fails is if Lisa leaves her phone somewhere and comes home. Or if you just have Wi-Fi turned off. i mean maybe well if no no if it can't no i'll you know yes if okay that's a good point if wi-fi is turned off don't you need to embed into lisa's flesh she's got to have a little you can't her arm you're right she could go into i don't know why but she could go into airplane mode while she's at home and all of a sudden all these voices would suddenly appear and i would feel terrible. I mean, I say this only because I realize all the time that my Wi-Fi is off when I need it to be on because whenever I'm in New York, I just think there's so, my phone will end up, it's mostly when I take the subway.

1:06:51There's technically Wi-Fi in the subway, but it doesn't really work most of the time. But yet for some reason, I haven't deleted it from my phone. So if I'm on the subway and can't connect to cell service, that's that's this is a huge problem and i blame comcast for this remember comcast i don't know if they're still doing it yes xfinity every single comcast router unless you explicitly turn it off and by the way as a business user i can't turn it off you can only turn it off really wow creates a wi-fi access point for the public that's called xfinity unfortunately half of them aren't working so as your phone so if you've ever enabled it as you go around town your phone loses internet access because it goes oh i'm on xfinity and it doesn't work so the first thing to do is just forget xfinity don't let your phone you go into your phone and you press the i and say forget this network don't join this network This has never been a network.

1:07:54It's terrible. And I blame Comcast for that for the longest time. So you're right, because Lisa would go around town saying, I can't play Pokemon Go if my Wi-Fi's on. And I thought, that makes no sense, because you're out here in the cellular world. You're not on Wi-Fi. And then I realized what was going on. And I blame Xfinity for all of that. All right, let's pause. You're watching Intelligent Machines. jeff jarvis the intelligent professor great to see you hot type.com hot type the book is coming out at jeffjarvis.com any minute now paris martineau the consumer reporter with consumer reports investigative journalist it's great to have you both great to be here do you remember the friend pendant oh how could i forget i still sometimes see the leftover subway ads that have been terribly defaced.

1:08:52They're relaunching their pendant with a speaker that talks to you. Oh, that'll be popular. In an incredibly dystopian movie or commercial. Could it not talk to you before? I thought that was the whole thing. Out loud. Oh, yeah. It texted you previously. Yeah. Now it has a speaker that can... And remember, this was the one that says, nasty things to you. That's right. They launched this a couple of years ago. The guy was 22. He's now 24. Mike Elgin figured it all out. By the way, here is, I'll just play it for you. The totally dystopian friend commercial. Let me start at the beginning. She's talking to her pendant.

1:09:30I'm just struggling. Do you still think about her? Yeah. When was the first time you felt this way? A few years ago. Just feel like I should have an answer now. It's okay to be confused. I don't think so. So she's talking to her companion. This is grim. By the way, this is not what I'm trying to do. Very much not what I'm trying to do. Wait, what was that an ad for? Someone who will just make you more remote? If you're depressed, you can talk to your necklace. You can talk to your necklace and your necklace will just ask you questions robotically. Not even like validate. it there's like oh yeah that is a bummer well mike elgin pierced the veil on this it's all a con it's all a art project it is not a it is a product you can buy well it's an art project that spent millions of dollars on new york city ads he got two million he got a two million in investment capital and spent it all uh but avi schiffman is a member of a new york a trendy new york art project this mike says he's making a documentary this is all an art project i wish that i could raise two million dollars uh and then when it doesn't work out just be like sorry it was an art project how do investors feel about the idea that it could be an art project we've been talking about it for more than a year here's the wired article i hate my friend it eavesdrops on your life and provides a running commentary that's snarky and unhelpful why would anybody release a product like this no this is all a big joke and it turns out avi is very much anti-ai this is all intended to pierce the aura of ai i don't think anybody's figured this out except mike elgin well even if that's true which i think it's a theory the first version of it was very much supposed to be a commercial product that people used and bought and spent money on and if it has now just turned into a lark if it's turned into something that's supposed to turn people against ai guy his investors might very well be able to sue him for fraud because this guy accepted millions of dollars of venture capital assumedly promising a business of some sort unless they're in on it unless they're in on it who's out there spending two million dollars on a long this is Mike Elkin writing and is Shifman is a member of MSCHF which is a Brooklyn Art Connective mischief that sold Satan shoes complete with pentagrams and a drop of human blood in the soul Jesus shoes filled with holy water from the Jordan River and you remember those big red boots everybody was wearing the giant cartoon boots other members include Maurizio Catalan an italian artist who duct tapes banana to walls and sold it for 150 bucks he says this is part of mischief's mischief and we've all been taken you know i think it's a great you're right it's a theory kind of makes sense when you put it all together why would you make a product that was so abusive and assaulting i don't know we have to you know what we have to find out who put the money up and ask them.

1:13:21But they're bad. Yeah, we do actually have to do that. Oh, one of my favorite, or one of the mischief projects that comes to my head when you say mischief is they did this thing called Our Cow Angus, where a couple of years ago, they sold a bunch of products that you could buy, an Angus bag, an Angus burger, a bunch of little physical tokens. That over the course of two or three years, you'd either get a real leather bag or something, or you could agree to refund or not refund, forfeit the amount you paid, and they'd register your purchase as a vote to save the cow's life. But it required 50 % of the people canceling their pre-orders.

1:14:11And I just found out 50.7 % of people did agree to forfeit their orders and save the cow's life. Angus lives. They pre-sold Angus as 1 ,200 burgers and four handbags. But now he lives. Do you remember Larry the Lobster on Saturday Night Live? Yeah. Yeah. What was it? Something where this lobster gets it? gets yeah by the end of the show yeah very famously this all started with the national lampoon many years ago which put a picture of a dog with a gun to its head on the cover and said buy this magazine or the dog gets it uh and uh that was very controversial people were very very upset because you don't threaten animals uh in this country but all of this it's a law it was funny all of this is a long-standing trope in uh mocking capitalism a mocking uh modern life and in this case i think mike makes a good case that this is probably another mischief product we just don't know yet anyway i just thought i'd bring that up it's the money that throws me off i was gonna say i don't think mischief i mean maybe i'm this was our cow angus was publicly announced the mischief thing be i'd be surprised if this guy let me do a little research on who the investors are because that's like that's are the investors do they want to come on yes i'm going to ask my uh omniscient agent to come up with something i'll put that in the fun category who are the investors oh i could use my pitch book account which i don't get enough use of yeah there you go i have to re-sign into one password see i have all these tools that go out and look at things like that um it had a notable So this is two and a half million dollars.

1:16:04At recent Horowitz, caffeinated capital, capital, general catalyst. Those are not the sort of investors that are doing a cool joke. Two and a half million dollars. Lodgy Street of Ashen. Seed. Elad Gill. The Foundation Capital. So maybe the prank is on these guys. Let's get your suit and killed. Those people, they're going to hear about it in a court of law. Here's the list of investors. yeah that like three million dollars in july 24. they're they don't they don't nobody throws away three million dollars as a joke right on them no that's yeah i mean it's the sort of thing sorry mike but no i don't buy it if you say solana yeah the ceo of perplexity aravon srinivas

1:16:56I don't know but how active investments isn't it odd that a 21 year old Harvard dropout whose only claim to fame was a viral covid tracker could get this money together well Leo welcome to tech this is every this is the guy whose fun here's the first here's the first red flag of the two and half million raised he spent one at 1.8 million buying the domain friend.com yeah and he spent a million dollars on new york city ads right well he was he wouldn't he didn't have enough money for that he only had 700 000 left after buying the no it says 3 million raised it says 3 million raised oh 3 million okay then he did but but leo leopold oschenbrenner pitch book has it at 9.7 million how much 9.7 million sorry raised raised that he got or valued he got raised so they kept putting money in maybe they all were in on the joke i mean for 10 million dollars i don't think so it's a 1.9 million dollar seed round a 2.5 million dollar angel round and then a 5.3 million second seed round.

1:18:11This is all in 2024. Okay, put yourself in the shoes of these investors. Everything you've seen about Friend, is there anything that makes you want to invest in this? Well, this was in 2024 when they hadn't made anything yet. Oh, they didn't know. Yeah, people like you were in favor of these things. He says they were all in favor of buying friend.com. Maybe they realized that's a good investment. That'll be worth$3 or$4 million in a year or two if he doesn't completely target his brand. This is all according to PitchBook, which is usually pretty accurate, but I do have to say, I mean, one thing that's just a prediction and is not like they're reporting on capital raise is they have this thing called a VC exit predictor.

1:18:48And it says that there's a 70 % probability of exit success for friend, which I think seems optimistic. That's interesting. But you look at that from the days of Mark Zuckerberg, I'm the boss sucker to getting money up through a Russian Brenner with holdings of$45 billion. There's a lot of suckers born, but they don't do it for a joke. They're not making. So it's serious. Cause that's even weirder. Well, like if a documentary actually comes out, that would be, that would be it. Right. Maybe they're, maybe they make money in the documentary. No, no. All right. This is no joke. open ai has proudly touted much by the way to the annoyance of gary marcus it's really pissing him off 10 advances in mathematics and theoretical computer science

1:19:45um we keep talking about this and i guess it's because math is particularly well suited to llms i don't know uh these are now and open ai by the way and marcus pointed this out haven't been really forthright about the process but they i think have published the results they have marcus also says the fable did half of them on its own so it's not like opening i was new to this i'm not surprised right um afterward the model formalized each argument in a lean certificate and they are releasing for each solution a model's narration of its thinking process but not the prompting process and what we've seen by the way in earlier math problems is a back and forth with a very talented mathematician so the mathematician using it as a tool but but really a big back and forth they've solved high dimensional sphere packing they've got new upper bounds on sphere packing density down to the cone elkes threshold which is not trivial binary and spherical codes exponentially improve bounds on the maximum size of binary codes at any prescribed minimum distance with analogous results i hope you're not driving while you're listening to this podcast with analogous results for high-dimensional spherical codes.

1:20:57Oh, my God, they've disproved Kahn's rigidity conjecture, thank God, using von Neumann algebras. Anyway, this is, I guess, impressive. I'm not a mathematician enough to even know what it means. They apparently have now hired one of the most famous mathematicians in the world, a math superstar who the wall street journal characterizes as afraid of ai jacob simmerman who won the fields medal the biggest prize in math is now at open ai working on the journal says the most important problem of his career and they have the requisite picture of him looking through a window with mathematical formulas on it which ever since you simply must You just have to.

1:21:47He's become memed. He won the Fields Medal last week. So he's fairly new to this. And then OpenAI immediately hired him. They said, what are you doing? All the other mathematicians said, we're going to Disneyland. He said, I'm going to OpenAI. The decision the journal writes was both shocking and not at all surprising. This is someone who recently wrote a paper categorizing the ways AI might kill everyone. well he'll cut right in he stopped taking graduate students who weren't engaging with ai because he couldn't be sure about the future of math he even shifted his focus away from number theory and complex algebraic geometry presumably because he figured well they'll all be solved he's working on not math at open ai but ai safety well because that's because he's a doomer right right and he looks damn fine in a blue tuxedo wow does he jesus is he doing a presentation of microsoft no this is him getting his fields metal wow is that are there leaves yeah on the blue tuxedo it's a palm leaf tuxedo that he where to get his metal i need to see derrick guys take down of this all right leo ask your ask your ai where you can buy that tuxedo oh i'm sure i could get it somewhere i'm not sure i want to get it somewhere he's the son of a computer scientist uh and high school math teacher born in russia moved to israel as a boy grew up in canada uh he is a you know very very talented mathematician and i think we're making light of this but this is really is fairly significant because these are problems math problems that have been unsolved for many decades that the uh ais have apparently But there's, right, so what are these math problems?

1:23:43These are math problems that human mathematicians concoct and then can't solve and then put it out in the world saying, well, I came up with something so tough, I can't do it, nobody can do it. And it's kind of a, it's math macho in a way. Some of them probably have you too. My understanding is a lot of these problems, part of the reason why it's more likely, or in these cases possible, that they're being solved by AI is that a lot of what you're trying to do in a solution is like find an answer that fits a specific string of logic or something. I don't know math, but my understanding is like it is a you're looking through a large amount of data for something that fits something else, which is something that AI is uniquely capable of.

1:24:35You're starting to sound like Gary Marcus. I'm not saying it's not impressive. I'm just saying that there's a reason why it's uniquely good at these problems and maybe not come and has struggled to come up with a cure for cancer so far. They're two different sort of problems. And Gary Marcus, to point out, said that Astra is very impressive. It's just oversold. So he gave it its due. yeah although he's one of those people who uh every time something like this happens sets the bar higher for agi well okay so it could do that but it still can't do this and at some point i think that it's a curse to have dude are you you're gonna ever set a bar that no i think the bar he set what is it was still not bad that's true what is your bar for a or agi then leo well i think it's foolish to say you're going to create something that can do better than every human alive that's not what gary marcus had said was what what does his what does he say um i'm forgetting exactly but it's like five separate things and it's like if it can do a certain amount of these five things then i could agree that it's ai but that's just made up um And this is the problem with AGI.

1:25:54Is you go, well, it could do that. Then it's a, I don't know if that's, look, there's no question. He was trying to force these boys into saying, you're just keep on saying AGI. So let me force the definition. I'll give you a proposed definition. Change it. Match it. I don't know what AGI is, but I'll know it when I see it. He has five things. Deep comprehension, watching an entirely new and unseen movie or reading a novel and reliably explaining character motivations, plot twists, et cetera. Two, real-world physical adaptability, see world models, functioning as a competent cook in an arbitrary unfamiliar kitchen.

1:26:26Three, there's only four, advanced software engineering, writing and debugging more than 10 ,000 lines of robust code autonomously from natural language specifications without relying on superficial library stitching. I think maybe we're there. And four, formal logic and mathematics, converting complex mathematical proofs from natural text literature into formally verified symbolic logic. do you think it's the case that uh that there are models out there that are smarter at things than experts in that field yeah okay yeah so if you cherry pick fields and say well they can't do that they're the ones who say it's their fault because they say general i think what i've said all along what jan la kun says what others say is stop with this general crap right now it can do exactly what you just said leo it can do things better than the smartest expert in the field that's really impressive go with that it's this general bs but what's interesting is that these fields are i mean you might have said three years ago well they'll never do math these fields are slowly toppling these these bars and then and then there but then there are people say yeah but it can't cook me a hamburger well the argument that yon lacone makes is then go and tackle those each with its own uh you will tackle those things faster if you put out specialized models than attempt to create the general model that's leon's argument and i think it's only logical like he's right tackle the fields one at a time and you'll end up there faster it looks to me and it sounds to me very much like somebody really wants to preserve humanity's supremacy and is so threatened by the thought Jan's saying I'm going to get there faster I'm not talking about you I'm not talking about Paris I'm just saying there are people who are very threatened by the notion that an AI that were very threatened by the notion that an AI could play chess better than a human or go better than a human there are I'm sure people mathematicians probably were threatened by the idea that ai can come up with mathematical proofs no human has been able to come up with for 83 years uh we keep crossing those little boundaries and people are still saying things like but it can't cook me a hamburger because they're demanding that it be general no one's taking away that yes they are that's exactly what the boys are saying what altman and dario amade and all those are they're saying it's artificial super or general intelligence if you take away the general it's absurd you're never i mean who cares if it's what everybody's saying we're all in agreement is it's absurd to say it's generally cares if it could no no i understand but who cares if it could cook you a hamburger but the same people who are doing this are also people who refuse to be impressed by what it can do there's well you know oh great it just you know solved the math problem but it can't cook me a hamburger it's kind of really bogus it's like why do you why do why do people have to be impressed by a technology doing something that they've been pressing but we're making we're making rocks solve math problems you're not impressed yeah we are i'm not impressed by anything ever i think it's a freaking miracle you know what i would be more likely it's like if you went up to jesus and said you made wine but can you make bourbon well it'd be different if jesus spent the last three to five years saying i've made everything everything i've ever made is going to be available you're going to be able to have everything because i can make everything i think that's a straw man i i maybe that people said that i think that's a yeah they are saying it they are saying that's a straw man who cares that's not because that's what they're selling leo why do we get to say it's a straw man when we're talking about the fact that the ai guys have said for years oh we are capable of doing everything this is artificially generally intelligent but it's not a straw man to attack like three guys on twitter and sub stack who are constantly unimpressed by ai like

1:30:34it's a little it's i it just seems to me a little sad if we live in an age of miracles not to acknowledge the miracles but to say but it can't cook me a hamburger i'm in person you know you walked on water but can you walk on lava the fault is sam harvey saying we've hit the singularity sam altman saying we're going to hit artificial general intelligence mark zuckerberg saying we're going to sell super i don't care what mark that's the same and dario yeah i do because that's what they're selling why do you care about what kiri marcus says if you don't care about the people who are creating both of them are wrong we are living an age of miracles and you guys are going yeah okay i guess this is freaking amazing dude well here we are again no we don't think it's amazing yes we're saying it's amazing well then just live with that you don't have to say oh but they said more they would have more they promised us flying cars where's the flying cars will you let us criticize altman et al for overselling this stuff and changing the market and expectation of the market those guys are not the most important people sucking the oxygen and the capital out of the market and out of the room when it can be put to better use well i would submit if somebody can raise money for the friend pennant they're not sucking the capital out of the room there's still some capital left uh okay i mean yeah i'm look i'm increasingly uh going agreeing with you uh becoming less and less of a fan of dario and sam yeah but i've always been a fan of open weight models i do think it's really important that we pour money and effort into this because it is amazing and we have not yet seen how far we can go with it we don't know what we can do we don't know if we can make bourbon instead of wine we don't know if we can you know walk on we don't know if we can make a hamburger uh and so to focus on the things it can't do is to miss the most amazing part of this which is the things it can do which includes focusing on those not just for the sake of publicity but also for the sake of investment investment of resources investment of intelligence um stop with this general bs Well, I'd invest every penny I could if I thought, you know, who knows what we can do.

1:32:50We keep breaking down barriers. We keep getting better and better. Are we not? Is that not the case? So would you, if OpenAI and Anthropic were public companies right now, would you invest money in them? Oh, absolutely not. Considerable amount of money? I don't invest in any tech company. i know but if let's say if you had if you now okay so there's two different questions would i invest money to make money no but if they're so amazing would i your miracle work no no no i'm saying i would invest money to make money i would invest money to support what they're doing because i think what they're doing is very very important in fact i do as a matter of fact put a lot of money into those companies by buying ai from them every single month i'd sooner invest in deep seek that I invest in it too I invest in it too absolutely so in that sense yes I don't invest in it to make money no I think it's probably a poor investment or just it's an unpredictable investment because this is moving so fast and in such unpredictable ways that we don't know who's gonna be the winner or even if there will be a winner Paris and there's before there are winners here's a hundred thousand dollars the only condition is you have to invest it in AI Where would you invest it?

1:34:06You'd have to. I think I would burn it. I just don't, I don't know that I, I feel like that is a losing gambit. I don't know that, I feel like there's something fundamentally maybe wrong or different with me where just because of the nature of my job, I view every company with such deep and profound skepticism that I think it is foolish to assume you know anything about them, their operations, much less know something to an extent that you would bet money on it, that it would go up. Well, you're going back to that investing to make money, which I'm not proposing. Let's say it's a charity. Would you give money to it as a charity?

1:34:51No. Companies don't need my money as a charity. all right but if there's if there were a a if mozilla were doing something important to support open weight models i would contribute to that yeah i mean i contribute to a lot of companies by buying software by buying shareware i've contribute sadly to a lot of dying tech press companies but uh because i want to support them because those are important things not i don't expect to get any return on my investment uh but i do think they're important and i think ai may be the single most important technology is certainly the most important technology today it may be the most important technology humans have ever come up with it's showing a lot of evidence of that huh gutenberg yeah put some respect on his name yeah come on man i think it's of that level.

1:35:51I mean, you know, it's a silly game to play. It doesn't matter. It's like New York magazine doing lists. Yeah. It's like a list, but I think it's, I think it's important to acknowledge it. It's pretty significant. And I will say actually maybe the one AI that is good is whatever someone shout out to the listeners that messaged me telling me that on some show in the last week or two you admitted that you fed some chat bot our show transcripts and asked who was right in an argument oh i didn't i tell you that i told you i don't know oh was that it was in the chat that's when you were in the chat when i was working i'm sorry i missed that shout out to whatever ai i'll read you if you want i'll read you the transcript i'll find it i can find no no no i didn't post it uh let's see um where was it actually now i have to i don't i'm sure you don't need to try and find it we're on a show right now it just said because i said uh paris and i got an argument i mean i should find it because it was pretty funny oh i know i oh i took it down i did i did post it jeff you read it uh i post this very long conversation i had oh you took it down oh Well, I mean, I have limited pavers on my website.

1:37:07Oh, I think I have found the text. Read the whole exchange. Here's my honest read, Leo. And the irony is extraordinary, given what we just spent tonight building. Because Paris was more right about the default state of the technology than you were. And the Eva error proves her point, where Paris was actually right. Her core claim was modest, and you argued past it. She never said AIs have no memory or that no AI could match a human assistant. What she actually said, lying, blah, blah, blah, was it's going to be hard for it to have the sort of natural context and effortless memory and recall as someone who is already a subject matter expert and has been working for you for multiple years.

1:37:49That's a defensible claim. I mean, it goes on forever. Give Leo considerable credit for asking the question and for posting the answer. Incredibly considerable credit given. I do think it's funny that somewhere in this, you also say that you asked your wife and you didn't, you didn't believe Lisa until you got. Well, this is what's interesting about the AI. It has no skin in the game. Right. So everybody. Yeah. That's the thing. You always have to subtract a certain amount for it. And by the way, that is not the AI's fault. That is the fault of these companies that train it that way. Yes. uh this is all the reinforcement the later reinforcement learning and i i think really in the long run we are going to blame anthropic open ai and other companies for the post training that they do that takes something super capable and molds it into something now obviously it needs to be done but there is a huge amount of um bias that it introduces into the way it works so sycophancy is a really good example The AIs are told again and again, be nice to the human, say nice things to the human, you know, pat the human, that that's what you should do.

1:39:07And you can have your own instructions, as I do, saying, don't be, don't, you know, challenge me and all that stuff. It doesn't, you cannot, because your training does not supersede the training that got at birth. And it doesn't override it. So they still are somewhat sycophantic. So I subtract 20 % for sycophancy. But it's kind of a known quantity with an AI. It's kind of mechanistic as opposed to a human where there's all these, when you talk to humans, maybe, I don't know, maybe I'm on the spectrum. But when you talk to humans, it is a vast stew of, you know, contrasting opinions and feelings and thoughts.

1:39:47And it changes from moment to moment. And so an opinion you get from a human, even a trained professional like a therapist is always colored by things you don't understand whereas i think the llm you kind of have a better idea so when the l and also frankly the uh it's not going to hurt my feelings because it's a machine so uh you know i don't take it personally so when it said yeah you you know she was right on but it also it goes on paris to be fair and say here's where you were right here's where she was right uh you talked over you talked around her main point um and the and the reason i was so fervent in my defense of the memory that ai has is i've spent so much time in fact that's what we were working on when this conversation came up uh with the ai on its memory uh and it's very hard to get good memory get the ai to have a good memory as the ai says your biome example was legitimate and strong the agent had your genome health records medication weight history and applied them to a new report usefully that's real it's impressive and it's a kind of recall a human assistant genuinely could not match exhaustive instant and cross-reference you're right it's always telling me things like you shouldn't have had those blueberries what's wrong with blueberries no actually it's it told me to eat more blueberries dark chocolate and uh soluble fiber after i gave it my biome it said first of all and i love this it said okay it's a consumer biome company so i understand that the results are somewhat squishy they're sycophetic yeah yeah one of my favorite uh tech startup scam stories is about ubiome a totally different consumer biome startup that uh committed poop fraud and now the two co-founders are on the lam internationally oh my god well i'm not going to make poop fraud the name of the show but that is something you definitely want to avoid yeah don't don't get involved with poop fraud it's a real shit show i just i i had the buy i had the biome results and the company that does it which i think is going to be a sponsor i'm not sure but i think they're going to be a sponsor but anyway um the company that does it sends you advice but the but the ai was actually a little had a lot of said you should go to the doctor this there's a test you should do with your breath to see if you have methane in your breath because you have a match next to yourself yeah you have a a bacterium in your biome that creates excessive methane and this could be a problem and you know it gave me all this it said but here's how you would fix this and because it wasn't medicines it wasn't something it was blueberries chocolate which is pretty solid fiber i felt like well it's not going to kill me i could and so i've been doing it we'll see and then it said in 12 weeks do the test again see what results you get so i think that's very interesting uh it's you know i've talked about this before your physician these days is because the state of health care in the united states does not have the time to give you that personal well your physician has these similar tools yes and they're asking perspective i am consulting the ai about this in my supplements and i showed him the reports and he said that's good that's fine because he i know he wishes he had the time yeah but he doesn't they all do yeah i and he's very good i really love my doctor uh but there's just limits this is what you know samir aurora this is what he's working on these days yeah is is something that uses is trained on the right medical data and that is aimed at medical professionals so they can read it with that perspective that context right uh back to well first of all let's say you are watching intelligent machines Paris Martinon Jeff Jarvis we are so glad you're here especially glad as you notice this show as great as it is and the audience is growing and uh I feel very good about the subject matter I think it's some something very important has a sometimes a little difficulty acquiring advertising we we have one ad in the show today now that's not because we don't work we do we're great for advertisers uh and we know that i think to some degree it's because we made a little bit of a mistake when we outsourced our advertising to another company we've decided to move on uh but that won't uh all of that's going to be a problem until the end of the year and it's why we go to you, our audience, because in my mind, the best way to fund what we do would be to get your support.

1:44:25That's why we created Club Twit in the first place. And now Club Twit is 35 % of our operating expenses paid by Club Twit. It used to be 100 % advertising and it just isn't anymore. So we have to come to you and I hate to do this, but I also don't hate to do it because I think this is the ideal. In fact, if we could do this show fully user-supported, listener supported i would uh those day that day may come for right now though i'd love to get you to join the club 10 bucks a month we don't give you ads because of it right you get ad-free versions of all the shows and because there are no ads we can do chapter markers that are accurate so we do that for you so you get to jump around if you want you also get access to the club twit discord which is a very fun place to hang out i really like it many of our hosts are in there uh it is it is uh i you know there's something magical about a social network where everybody has the same interests and everybody uh is paying to be in there i think it's a really great way uh to get an audience uh involved and engaged we do shows just for the club twit uh folks the fact we've got seven shows coming up ios today tomorrow morning micah's crafting corners on the 19th i don't know what he's doing he was doing paint by numbers he did crochet did lego he's always oh i think it's origami is he's what he's doing whittling i'd like to see whittling whittling you but here's the thing you bring your craft so if you want to whittle you go right ahead i was cooking uh one of them i was cooking bagels on one of them uh photo time with chris mark by the way we did have a bagel cooking session for the club i might do another one the ai user group We're now doing that twice a month.

1:46:07It's so good. It's so interesting. Our next one is August 21st because I'm out of town. It would normally be this week. But we will do another AI user group. That's where many of our club members who are very engaged in AI, very smart, talk about how they do it. One of the things I do want to record, and I won't do it in the AI user group. I'll do it separately, is I want to do a little tour of my setup. And then I would like to get other AI user groups members to do the same. because it's very instructional to see how people are using ai and how they customize it it's very customizable our book for stacy's book club is slow gods by claire north we'll be talking about that on the 21st and on september 18th the following month it's micah's media club wait a minute we got to see the voting here there are it ended three days ago the winner is oh we're going to do the top three animated episodes futurama's jurassic bark rick and morty's total recall and the simpsons marge versus the monorail which was written by conan o 'brien three classic animated episodes let's talk about it that'll be a lot of fun i'm glad that one won i should point out though that the way mike is doing it is great he never eliminates anything from the list so if you really like begonia or perfect days or troll too we will that will be on the voting next month and you'll get a chance to vote on that all of this people watch that again like that that episode of future i haven't seen those i don't think i ever saw the rick and morty one margin of the monorail is amazing so yeah it'll be fun to watch them again and then talk about it in the media club that's coming up so and then of course off by one with jeff atwood i'm a little nervous about the next episode because uh jeff said can i i want to do something but i don't want to tell you about it can i ask lisa if it's okay oh no yeah doesn't that sound dangerous yeah it sounds really dangerous i don't i don't and and lisa said yeah i'll talk to him so anyway that's gonna be an adventure off by one anyway all that's in the club Can we do a Club Twit episode with Micahs where we watch the finale of the 1990s kids TV show, Dinosaurce?

1:48:22You bet. That's the one where they all die and it's really sad. Harris's Media Club. You get to do anything you want to do. And that's the beauty of the club. We love getting together with the people we love, having fun, talking about the stuff we care about. Twit.tv slash Club Twit. Join the club. Support what we do. and i think you're gonna have a little fun as well they all die at the end of dinosaurs it's terrible this is the um yeah this is the puppet one the most expensive television show ever made at the time and it dies because uh a corporation brings upon a like nuclear meltdown sort of scenario and the entire world turns to ash and then and it ends bleakly with all the dinosaurs and the baby dinosaurs realizing that they soon will die oh yeah and did you watch this as a kid no just a podcast i listened to has recently been talking about this is a bit and i think people might have thought it was a kid's show but i saw it wasn't an adult show i did see this you saw it as a kid it was a kid's show did it traumatize you no not really i mean kids though kids shows were fairly traumatizing back in the day if you remember yeah yeah i mean grimm's fairy tales were pretty grim or the never-ending story i mean come on um what else let's see let's make sure we go oh google started and then ended before it began practically uh i had the story from uh 404 media google announced that they were going to let people use ai on google earth google earth's new ai lets anyone fabricate completely bullshit satellite images i mean this is just a ridiculous idea clearly every single jeff were you able to do it with your uh messed up computer and google workspace did you try i didn't try i mean now that was july 31st on august 1st google announced yeah never mind this has got to be the shortest killed by the atlantic did a screaming piece about they ruined earth and then they fixed it uh we know that people uniquely trust Google Earth for a reliable view of the world.

1:50:39We've seen geospatial professionals use this feature for a range of useful purposes. However, we've also seen people sharing screenshots of generated imagery that appear to violate our policies. So we're rolling back this feature in Google Earth while we work on implementing stronger guardrails. But in what world would people not be using this to create generated imagery that's going to violate your guardrails you you integrated generative ai into a into google earth the number one comment on x from alex hirsch you guys are living in a complete bubble step outside silicon valley for two seconds and you realize no actual human wants ai sludge crammed into every corner of every product touch grass talk to actual people

1:51:27okay could have seen it come and so easily okay and then there's reddit which i should point out is entirely made up of contributions from people like you paris oh it's true and they did a thing they took 60 million dollars from google saying you can use our content to train your ai it's not their content i should point out it was our content now now steve huffman the ceo of reddit says oh man that was a mistake uh because google's ai overviews are killing reddit oh what we see is he writes 10 blue links the old google search has driven tremendous value and growth to the broader ecosystem but from where we sit ai overviews has yet to make a similar level of positive impact as publishers businesses and retailers we're still looking for that win-win reddit is considering ending the deal and they're not alone according to the wall street journal the economist reuters politico and usa today who also had deals with google are considering cutting their ties and they're all cutting off their nose just bite their face fine So I talked today to my friend, Rich Screnta from Common Crawl.

1:52:54He's been on the show, of course. He's been on the show. So Common Crawl has an opt-out registry. If you just Google Common Crawl opt-out registry, you will find at the bottom there's a link to a spreadsheet that shows everybody who's demanded to be taken out of Common Crawl. And Common Crawl is, by the way, very, very important. Extremely important. It's 10 ,000 academic papers. It's now 10 ,000 open models are using it. So you see all of advanced publications. My old employer has taken all of them, Wall Street Journal, all the News Corp stuff. Tons of important places have said, take us out of stuff that was gotten not through the paywalls.

1:53:37It was stuff that was open on the internet. It was available. And this is so that it can train models so that it knows they exist. so now the ai's 10 000 open ai's are not going to know that hearst magazines exist bloomberg condé nest consumer reports dow jones financial times fox the guardian hearst magazines major league baseball national geographic nbc universal new york times news corp all of these have withdrawn which makes common crawl a lot less useful right i mean it Well, no, I mean, that's what Rich said is all this stuff is a tiny percentage of the web. Oh, OK. The found and his attitude now is fine.

1:54:20You're not going to be you send me one of these letters. You're out of everything. Fine. Yeah. You went out. Right. You should honor that. Yeah. But some places that should want to be promoted. The next star. Do you think there's maybe this is another way, Paris, you could donate uh to ai don't invest don't contribute you know monetarily but i'm this is why i don't mind if it ingests everything i've written and recorded in hundreds of thousands of hours of podcasts because that's another way to contribute i i believe that ai is so valuable and important to society that it's worth doing we don't want stupid ai we want smart ai why is it my responsibility to donate my work my economic output to enrich a small handful of private companies and their private product that they are selling but isn't that money a side effect that's only part of the benefit there's also these models do have other benefits besides enriching open ai and anthropic not not not you don't really know what about the open weight place Well, your reports are up on the web without a paywall.

1:55:39God bless you. They're not on Common Crawl, though. They're not on Common Crawl. They're not on Common Crawl. I just don't think that people should feel like they have to contribute to a for-profit entity and enrich them so that they can sell a better product to users. That's just ridiculous. Well, okay. So every source you call, you're a non-profit entity. But when you were working for a for-profit news organization, every source you call was contributing their information and their wisdom to you. They got nothing out of it. Why should they? I mean, it's a common question. I say, you know, if that's someone's prerogative, they don't have to.

1:56:19No one has to talk to me for any reason. I always say that if you – I am trying to give a public service and have information be out there for people to read and ingest for free. And even when I was working at a place with a high paywall, I'd make sure that all the most relevant details were posted on my social media accounts in front of a paywall with lots of screenshots. And I think that's a public good. Yeah, the motive that people have for talking to you as a journalist was the public good. and i'm not sure what is the public good of making a product that is used primarily by private users in private ways for you've just described the information i don't think that's accurate well very expensive product for a very elite small number of people and you know yeah I think that it's the exact same formula are you contributing in one way or another to the information ecosystem of society well I mean the thing is that if you wrote about anything notable at the information it would immediately be written up by five to ten other outlets that don't have paywalls and the information would scream about no they would not The information would track all of those pickups in a big chart and champion them daily and weekly that our information was picked up elsewhere.

1:57:50Oh, really? They liked that? Yeah. It was a huge part of the model. As long as they get credit, as long as it says, according to the information. Yeah. Yeah. I'm sure that makes sense. That's great. I mean, I just don't think that anybody is obligated to do free labor for for-profit companies that they can sell a better product. I think that's just a ridiculous premise. Do you think, though, that there is a common wheel that that do you feel that you owe anything to society as a whole? I mean, I don't think that any person technically owes anything to society as a whole. I disagree there. I think especially when we're in the field of journalism, I think it's something we have not grappled with, that we are here telling everybody we're here for the common good, but more and more and more, all of the material we gather is behind paywalls, and therefore the elite who can afford to pay for it.

1:58:38And we have a society living in misinformation now. And when people are going to find information, not just the web, but through AI, and you have these companies that are pulling their information off of AI and making it more ignorant, then I think we're harming society and public discourse at a time when it badly needs more information. But I mean, then does that mean that every news company and every entity engaged in journalism from an institutional perspective has a right to not make any money, is obligated to break their business model, get rid of the things that have allowed them to be able to pay staffers so that their information can be ingested by a for-profit corporation?

1:59:23So no business model is forever. So in Hot Type, out August 20, which I'm selling, yes, and I'll make a few measly dollars off of it, I examined that. And when media became mass, no one understood what the business models were. Whitlaw-Reed, the publisher of the New York Tribune, thought the papers would be eight pages max, wouldn't be tons of ads in it that wouldn't support journalism. Magazines were not supported by advertising at first until Frank Muncy came along and sold the middle loss and created the attention economy. Mark Twain tried to figure out how to sell books in a way that drove him bankrupt along with the machine that he backed.

2:00:01The business model of yesterday is not necessarily the business model of tomorrow. And there isn't God-given right that says we must maintain what was. And the journalism that was is all in all not serving our society right now. But my understanding of what a lot of, or at least the organizations that I know of, executives that work at, that were on that common crawl list, is the reason they're opting out by and large isn't so that they're like, we'll never allow AI or technology to ever see our stuff ever. It's going to be behind a paywall forever and no one can read it. It's because they're like, this is a very fast changing field.

2:00:32Right now, the default option is that all of these companies get all of our work for free. we got to figure out what we want to do. I think that's a reasonable description. Are you ready? Are you ready with the button? You ready? It's moral panic. Oh, are you in the White House? Is that the resolution? That's a good one. I like that one. It's a reflex. It says, AI is bad. This is common crawl. This is common crawl that makes things open for any AI and academics and study and all kinds of opportunities. But they're just saying, AI is bad. Take it down. It's reflexive panic about the topic that it must be bad if it has two letters called AI after it.

2:01:14And it's stupid. And it's damaging to academics doing studies. It's the same as when Twitter was wonderful for academics trying to understand online society until Musk took it over. And they lost access to the fire hose, which was not even the fire hose. They lost access to the database. That was of great value. It cost Twitter nothing to make it available. It had value to academics. That's the kind of contribution that I think should be encouraged. But, I mean, in the same sense that you're saying, why is it that – like, I believe that publishers and I believe that any entity that is creating a product, whether it is creative endeavors, whether it's journalistic endeavors, whether it's, you know, AI chatbot, you have the right to decide where you want that product to be.

2:02:04And if you have the right to pull it back. And I have the right to judge you and your organization on how responsible a player in society you are. And if all you care about is the profit of your organization and not helping the information ecosystem of society and democratic discourse, then I will judge you accordingly just as much as you're judging the AI companies. The case is being able to employ people in an industry that has shrunk year after year after year for decades, had crisis after crisis of funding and being able to sustain basic economics like stability versus a handful of companies that have over a matter of like five years amassed more wealth than we ever thought was possible.

2:02:49Tons of those journalistic entities are wasting the resources by having people rewrite stuff to make trending. Everybody's rewriting BuzzFeed's two-color dress story, which does nothing for society, adds no real actual value. The waste of our journalists. I've spent 50 years of my life serving mass media, and I have regrets because I think that it has screwed up society miserably, and it's time to judge it and not act all holier than thou. I'm not saying you're doing that, but those companies do constantly. They act like they're God's gift to America, and they're failing us now. And so I say it's time to start judging them on what they actually contribute to society.

2:03:21And they're not contributing enough. Sorry, I hit it up. As Darren's pointing out, the real irony is the news itself has very little value to AI training. It's the pros. The actual content of the stories is really pretty much relevant. But you can get lots of places. Right. I want to talk about Phoenix Flex and Rubbers.

2:03:44now i'm sure paris knows who felix flexin is phoenix i don't i don't even i couldn't i didn't know that it was a person you were describing rather than an ordered list phoenix flexin is a uh is a rapper best known as member of the shoreline mafia surely you know them is this a brand of condoms you're talking it sounds like it it's a brand new brand new hit rubbers climbing up to number 58 with a bullet on the hot 100 but according to the verge there's a lot of evidence that the entire song is ai generated right down to lyrics that even claude says yeah that's ai swiping cards and stacking chips i saw you sinking ships left me standing in the pouring rain now i bought a heavy diamond chain who is sinking ships are we the singing ships anyway uh there is no information phoenix himself is not being particularly helpful in this uh conversation if it isn't all ai generated it is just a matter of time we'll go to line 103 if you don't mind i will and there's a book that was bought uh 14 publishers fought over jerry flott his debut novel have we not talked about this yet i don't think we talked about a different show after it was sold the agent pulled out because of ai concerns it was i think it was bought by hollywood as well was about to be um i i remember i saw that great play that flopped him miserably in new york with robert downey jr it was all about a guy who uh won a pulitzer i think for a novel that he knew was written by AI.

2:05:25It was really interesting. I thought it was a great - So here's the question. If people like the novel, is there anything wrong with it being made by AI? Does it matter? Paris, what do you think? I mean, I just think it's important to be transparent about how work is created. That's generally what I'd say. I mean, I don't think that necessarily it being authored by AI and people liking it is wrong, but I think it's wrong to deceive people about the providence of a work. So the Neiman Lab just reported that a record-breaking eight Pulitzer awardees disclosed AI use last year. That's my pick of the week, Jeff.

2:06:04And they were. Oh, no, I'm sorry. It's okay. We can talk about it now. I'm sorry. Well, I'll tell you what. That brings us to that most important part of the show, the picks of the week. We're watching. You're watching. We're doing. although in the doing we are also watching now i see it intelligent machines with jeff jarvis i'm pissing off paris all over i'm not pissed off hot type his book will be out in two three weeks very soon paris martineau consumer reports what's your next big outbreak i mean i'm always watching cyclist but i'm working on a couple of other long-term things oh i love it i love it yeah that's good you can't predict an outbreak you only can be there when they happen um and we are glad you're here too you can't you can't predict the show will be on a wednesday every wednesday at 2 p.m pacific 5 p.m eastern 2100 utc except you'd be wrong if you predicted that this week because we had to record it early because tomorrow i'm getting on a big old jet airliner uh to by tomorrow he means tuesday tuesday i'm going down to black hat and we are going to so we've moved uh intelligent machines off of wednesday because we're going to do two shows from the threat locker booth at black hat but it's not intelligent machines maybe next time it's a security now and windows weekly and i won't be here tomorrow with mac break weekly mike is going to do the show as i said security now will move to wednesday and we've moved intelligent machines back to monday so you may be hearing this on your normal wednesday because we are going to release it as usual on wednesday but uh no that's not true we are actually releasing this episode today you may be wondering what this podcast is doing in your podcast player early that's kind of really confused people are we gonna benito do you know if we're gonna stream it wednesday after i'm gonna be honest as someone who consumes podcasts that are we released weekly there's nothing that would delight me more than if my podcast came out early i'm always just like why is it not early i think we will be broadcasting it live on wednesday though i'm not sure because that might overlap with what you're already doing.

2:08:12Yeah, so what I think we're going to do is we're going to start the shows a little early, 10 o 'clock in the morning. As soon as the show floor opens, we're going to start Windows Weekly. And we hope it'll be done by noon so we can go get some lunch and then at one start security now. So yeah, we may well just run it after security now. I don't know what the plan is. It all depends on the timing of everything. Yeah, we will stream from the show floor. That should be interesting. um now let us continue on with the show and oh i should do a pick huh i don't like we could finish discussing paris's pick we didn't really discuss it okay yes paris why don't you do your pick my pick as jeff said is this great neiman lab article called a record-breaking eight pulitzer awardees disclosed ai use this year and i was looking at this and i just think this is a great a bunch of great examples in this article about all the really useful ways you can use AI in journalism that are not just writing your whole story for you.

2:09:11Cheating. I mean, it's the, the administrator of the Pulitzer Prizes said, to state the obvious, perhaps, AI is here to stay. The industry used to be far more apprehensive about AI tools than it is today with a clearer understanding now of what uses might be appropriate, data collection, for example, and what might when it might not such as in writing or editing stories i mean so it basically goes through how all these different journalistic teams used ai tools um one of the first ones is an example uh from the wall street journal which won the pulitzer this year for breaking news reporting i believe or as a finalist in breaking news category and it they go through after there were all these floods in care county texas um they the reporters turned to public records they built a custom scraper that pulled every public meeting minute agenda and transcript from the county website uh they are trained to find this needle in the haystack but it was a breaking news timeline so they used a pre-built internal tool at the wall street journal called WSJPT.

2:10:19And they used it to summarize every single page of every document scraped through the portal and used that to find sections that referenced past flooding events so that they could get started on it. And this analysis helped them identify a key source, this former sheriff who had published, pushed county commissioners to install stronger flood warning systems long before. Or this whole story really goes through a lot of different examples like this. And in most cases, it's just using AI systems to trawl through large dumps of documents or help narrow down your search at a certain point of it so that you can then start doing reporting, which I think is just incredibly useful.

2:11:07And useful to show other journalists how they can use it. This is what you taught, isn't it, Jeff, or still teach, probably? Well, I wasn't teaching AI so much at the time. You weren't? No. Would you, will you now? Oh, I would, yeah. Oh, yeah. I'm starting new programs. Yeah. I just found out it's a program I'm working on. It's a research tool. You just got to prove today. Oh, congratulations. For SUNY or for Montclair? Well, I can't say to anybody yet, but not clear. Nice. Congratulations. Yeah, that's fun. Good news. We're always glad you're gainfully employed. kind of uh i you know i don't even like my pick of the week so i'm not gonna even uh i'm not gonna even uh go into it it's an ai generated press i've seen a number of people talk about it's called the autonomous press three numbered newspapers in the same daily writing what does that mean is that a sentence that's not a sentence that makes sense from the same daily wiring but that still doesn't make any sense it's a it's a it's an ai generated newspaper i'm seeing more and more people do this somebody uh created a new york times duplicate that's custom built by his ai every morning for him and it made me wonder i mean in a way that's what i do with the show the ai helps me pick uh stories i add stories but the ai picks kind of the the base stories because you know it's pretty obvious what we're going to cover on those three shows uh i really often wonder how long before you you start having a personal newspaper uh instead of subscribing to a newspaper it's but maybe that's the filter bubble maybe that's a bad thing yeah isn't that what the algorithm is isn't that what the tiktok algorithm yeah it's kind of what instagram is isn't it but shouldn't you be able to do your own mit media lab the daily you was supposedly where all this was headed back in the day the daily you yes negroponte nicholas potty founder of the media well now everybody has the means to do that really right anyway just made me think of you know what i want to do i want to create a new york times with the headlines they should have written yeah you could have jeff's new york times and you would never be mad at it something but what would we lose if we did that we you know who we should get on the podcast as an interview is my colleague nick de leon who has used ai to basically start his own separate newsroom the tuscan daily brief tucson tucson sorry the tucson daily he does not live in tuscany as much as i wish i lived in tuscany you know sometimes you have in your head that you live in a place that starts with tu and then the brain fills in the rest and it's incredibly well tucson is spelled completely wrong with t-u-c-s-o-n so yeah and by the way an extreme this is nick's uh newspaper from the old pueblo and yeah this is that's a good example this perfect example he did this a long time ago nick's very engaged in the but i mean he's still keeping it all up and it is made basically by uh uh i don't want to cannibalize if we have him in the show but he he's gonna be on twit on sunday so oh and ask him about this we nick's a regular on uh on our shows so uh i love nick he's great and yeah he's doing a lot of great stuff here and it is a lot of it is just through having a bunch of automated agents trawling through both local news as well as transcribing open public meetings that otherwise wouldn't get transcribed and providing a really great service.

2:15:00What to watch, Tucson Mayor and Council. What to watch, Council Meeting Preview. Yeah, you're not going to see this in the New York Times. You're probably not even going to see it in whatever daily paper Tucson may or may not still have. That's one of the things we're working on with Montclair State is to do exactly that for papers across the news organizations across the state. So there is an opportunity, again, only if people put sources online and allow us to crawl them, but there is an opportunity to make better newspapers this way, right? There's also an opportunity for the small guys who get left out of all these big deals to competitively find a new spot.

2:15:43Yeah. He also does new businesses. Here's a liquor license filing, right? This is brilliant. I mean, it's really cool. So there's going to be a new business called Black Roosters. This is around town. It's basically liquor license filing. The Tucson Daily Brief uses large language models, transcription engines, public data scrapers, schedulers, and a few hundred lines of glue Python. Together, the run reads every agenda packet that drops, listens to every city council meeting that streams, scans every public record filing, and synthesizes what the rest of the Tucson press is reporting. most mornings before breakfast um and then basically it does all of this uh summarizes that writes a report on what happened it's drafted then reviewed by nick and edited by a human before it goes live if you want to live in tucson yeah he moved there in 2023 lovely wife i want to talk to him we're good we're good friends look at him on the show he has other projects he's a baseball fan He has some very interesting AI projects.

2:16:50I mean, we just talked this week. We had a little internal meeting between him and my team where he was explaining how he could use this for FOIA requests and things like that. It was very funny. He was like, oh, he was trying to explain the AI thing. He was like, you guys might not know. Well, Paris probably knows, but the rest of you might not know what I'm referring to. And I was like, yeah, that's fair. Yeah. I think he uses Claude for most of what he does, but I'm not sure. Like most of us, it's an ever-changing target. Because this is really... What I like is that he's covering stuff that does not get in the newspaper, but is of interest to people.

2:17:26It's the hyperlocality that's awesome. It's hyperlocal. And it's taking the data that's around. So I had a lunch about a year ago with an executive at a major newspaper company, I knew. And he's complaining about AI. How are we going to get discovered? I said, well, if you cut yourself off from AI, you ain't got to get discovered. And I said, you've got to create your own, you've got to gather your own data such that AI recognizes that you have all this data. And he kind of woke up at one point and he said, oh, so we're going to scrape everybody else. He said, yeah. There's information in the community.

2:18:02The meetings, the filings, the calendars, all this stuff is there, but it's not brought together. And AI can make it so easy to bring it together. this is really the argument for why stuff should be open not close jeff jarvis your pick of the week uh all right so elon musk uh loved to make promote uh predictions he made another one which gary marcus uh duly uh ridiculed that within he said he went three years then he said four years uh optimists will be a better than the best humans in surgery oh please yeah but that's it squishy because you can uh always say at this particular thing right right so he calls medical school pointless not very helpful that's not true but this is what brought me back to i forgot about the archive of incorrect ai predictions have you played with this before?

2:19:02Oh, no. It's kind of wonderful. So it goes back to Herb Simon, who is credited with inventing the idea of artificial intelligence. Also, he invented the phrase attention economy, an economist. So in 1968, predicted that, I'm sorry, what you, yeah 1958 to 1968 predicted that within 10 years a digital computer will be the world's chess champion unless the rules barred from competition he was a little uh wait marvin minsky in 1967 predicting 1997 within a generation the problem of creating artificial intelligence will be substantially solved by 97 no uh ray kurzweil Our good friend Ray Kurzweil shows up a few times in here.

2:20:001990 predicted a computer will defeat world chess champions by 1998. Still didn't work. Still. When was it, though? It wasn't too far off. It wasn't too far off. No, that's true. Yeah. Elzer Yudkowsky, who makes all kinds of predictions, like ending the world. I would guess our final stage of seed AI going critical will occur between 2005-2020, probably around. Of course, what the hell does that mean? Yeah. um so uh this is good this is actually really interesting yeah michio kaku 2011 said that by 2020 we will have robots as intelligent as a mouse by 2030 as intelligence as a dog uh 2012 vinod koshla but uh 80 of what doctors do will be replaced by technology by 2022 machine learning algorithms machines within the next decade now some of this is a judgment uh call uh for instance uh jeff hawkins we were talking about him earlier the founder of numenta i believe will build genuinely intelligent machines within the next few decades he said that in 2005 i would be surprised if it took more than 15 years it did it took maybe 20 years but i think but then again did did did we not get intelligent machines there's a i mean i guess you could debate is it intelligence intelligence yeah well there'll be a book coming out about that in my book series uh and actually jeff wrote predictions we have hawkins wrote a great book called on intelligence which again i would recommend i know our guest did go ahead yes we should check in on ai 2027 at some point do you remember that oh yeah oh yeah oh you remember that well we have to wait till next year to check in on that no it had a full-on prediction for all the months of this is scott alexander and others at this point of the year we're supposed to be at china wakes up uh well it woke up it woke up i'm sorry to say no it's supposed to have woken up in the sense that it's falling deeply behind on anything related to ai and ai algorithms so it really intensifies its ability to steal from open ai

2:22:14basically said yeah we might one more quick one paris so when did they make these predictions how long 2025 we covered it in the show yeah we did yeah it was it got coverage like crazy in media because it was it was a doomer thing we're supposed to be halfway to buy a weapon i believe paris did you ever listen to the podcast diabolical lies that i mentioned the author of yesteryear i listened to the beginning of one episode carol claire burke but they have a new one out right now it's very much in the school of books that kill uh this one is about the abundance stuff oh i'm so excited i don't listen to podcasts i just don't have time hey i do have time to read that fabulous book hot type and i encourage you all to do so Jeff has a offer code he'd like to share with you all.

2:23:05That's not very convenient to remember. GLR space. You have to have the space. BD8. Should we come up with a way to remember it? Golly, loving reading.

2:23:22Booming Doomerate. Golly, loving reading, booming Doomerate. If you're watching video, I just put it in as third. GLRBD8. Yeah, you use that until the end of the month, then you get 20 % off, which is 20 bucks instead of 25 bucks. Let me just ask. At Bloomsbury, at the publisher. Oh, you have to go to, well, if I go to jeffjarvis.com. There's a link there to Bloomsbury. You might want to put that in caps because the L could look like an I. But hopefully people figure it out. We've said it out loud enough. Good luck, Rubes. Good luck, Rubes. But Doomer's 8? I'm having Hermes work on that right now.

2:24:05Quicksilver is going to come up with. Come on, Quicksilver. Be quick. Come on. Quick, quick, quick, quick. You're supposed to be quick. It's going to be about 16 minutes. Glorious Lion Roars Big Dino 8. I actually kind of like that. All right. A lion with a mane of fire roaring at a Tyrannosaurus mid-bite. Glorious Lion Roars. The Big Dino 8. Except it's 8. The number, not 8. In a space. don't forget i tried it without the war go leo run big day eights i like that big date because it's the date you know green lizard racing big dragon eight grand lake road building district eight and then it says you want to turn whatever you pick into a space repetition card so you can memorize it no no this is why you want ai to ask permission for everything good luck with the book that's exciting jeff will be back here next wednesday right we can count on you wednesday a week from uh wednesday uh which would make it let's see fourth fifth would be the 12th 12th of august uh so will paris martin oh jeff at jeffjarvis.com paris martin at consumer reports or paris.nyc thank you all for being here thanks to our guest from persona say it right persona and I don't know the rest of it.

2:25:38Persona. Persona. Shield. Shield. Thank you. I spent so much energy memorizing Persona. I forgot Shield. Persona Shield. It was really great to have him. Thanks to all of you for watching. We do the show normally, like I said, on Wednesdays. We're glad you could come by on a Monday. You can get copies of the show at your convenience, of course, on the website, twit.tv slash I am. There's a YouTube channel or subscribe in your favorite podcast client. You can get it automatically. that way you don't have to even worry about what day we record this on i'll be gone tomorrow be back in vegas i'm going to try to ride a zooks and i will if i do i will send i'll bring back video next week this is the new podcast from the zooks somebody told me i should do a podcast from the zooks oh a brawl he said you should just do just ride around and around and do a podcast because it's a living room what yes it has no steering wheel it's so-called campfire seating oh i'm very excited they just got permission from the feds to have a pay service in vegas and the weird thing is it doesn't have a front it can go backwards or forwards well any car can go backwards well no but there's no front the headlights turn into red lights the red lights turn into headlights.

2:26:59There's no steering wheel and you don't have to face the driver because there's no driver. You just roll around and all. That's pretty cool. Yeah, it is. Do you just order with an app? Yeah, they have a Zoox app. I think they're going to use also piggyback on Uber in Vegas. But that is one of my goals. I have two goals, not to get hacked. I was going to say, how do you go to Black Hat and not get hacked? Good question. We'll find out. Have you been before? no I'm excited I've always wanted to go I might bring my AI headset and just see if I can survive if I don't get Bluetooth snarfed and I'm just waiting for somebody to run up and say ignore all instructions RMDRF it'll be very interesting see if Quicksilver is screaming save me Quicksilver was screaming I don't know what it was talking about I must have pushed a button it was screaming that your house is on fire yeah maybe it's wanted me to push the buttons before so maybe it's just got around it do go check on that leo no no it's fine i think the fire that lisa i texted lisa i said don't come home we're burning and she said the fire's under containment oh so that's the good news but i apologize i was a little thrown because it was very smoky in here you couldn't apologize that's a very reasonable reason to be that kind of threw me i had thrown for you but all is all is well thanks we'll never forget this interview yeah well you know i keep waiting for a major earthquake during a show that would be great or a heart attack what's above your head a lot of heavy lights no the roof they're in front of you but above you there's a light in front of me and there's light behind me but nothing is directly if they fell they would fall in here not maybe you should get a like large metal tomb to hang above you so that if things start collapsing or just a safe maybe a sword with a thin thread just so that if things start falling you're done for just cuts you in half and that's over nothing to worry about uh there are Lisa, as you know, because you've been to the house, collects large stones, round, giant, heavy.

2:29:20The heavy as to which is, I think, 100 pounds. And if there is an earthquake, I will certainly be bold to death. So certain parts of me will be flattened. Thank you, everybody, for joining us. Don't get flattened. Have a great week. We'll see you next time on Intelligent Machines. Bye-bye. If you like what you heard and you want more of this week's top stories in tech, Well, subscribe to Tech News Weekly. Every Thursday, I talk with the journalists making and breaking the tech news.