IM 889: We Are the Neanderthals - Why LLMs Fail at Complex Reasoning

33:07And oftentimes, even when you hire the right people, the company is growing so quickly that very quickly that job is now too big for the person they had hired to do it. And this is why companies who hyperscale make huge mistakes. I would look back to the social media companies and how they kind of manage their policies and how they manage their algorithms and targeting consumers and what that has done. I would look at ai today and i would tell you i'm seeing the exact same things same mistakes what advice would you give daria if you were his crisis pr guy if he called you up today and said we need you patrick so i've i've actually talked i i do have a personal relationship with one of the major lab leaders and i and i have talked about this and um i've i've told them that they need to be number one, much more transparent about what they're building and why.

34:01The problem with LLMs is they're a giant black box in how they are built. And so you're asking for the general public to give a lot of faith in Silicon Valley that they're going to work in their best interest. And again, you can look at the trust measurements that come out of companies like Edelman that I used to work for. You're going to find that after Washington, D.C., the next most untrustworthy institution in this country according to the general populace is Silicon Valley. And you have to also think about the mentality of the average American who, particularly those that are older and remember NAFTA.

34:39I think this is like one of the most forgotten, interesting, and pertinent parts of American history that we need to think about again when we talk about AI, because NAFTA was this hot topic in 1992-93. You had George Bush, you know, incoming president running against Bill Clinton, and you had Ross Perot sitting in the center. Bush was, I am rah, rah, rah, we're going to sign NAFTA. Bill Clinton's like, ah, I'm going to wait and get elected, and we'll see. I'm going to take a really smart approach to this. We'll wait a couple years. And then Ross Perot was out there beating the drum. Look, it's going to cost jobs.

35:11Things are going to move, and this government is not ready to create a safety net for those employees and citizens that will be impacted by this. Well, Clinton ends up getting elected because of Ross Perot. And what was the first thing Clinton did? Two months into his office, he just went and signed NAFTA. And he was sitting, I think, in a facility in Pennsylvania talking about it. And what he said essentially was, I'm paraphrasing, was that people who lose jobs in industries like mining and steel, that they were going to be trained to work as coders in the new silicone economy. And I don't know about you, but I have not come across too many coders that, you know, work the manufacturing line in Chillicothe, Ohio, or in Schenectady, New York.

35:56And so I think there's a lot of natural apprehension towards these products, and there's good reason for it. And if you're a lab, you need to be more transparent what you're building and why you're building it and what you're willing to stop for. But Patrick, if the answer to that question today, I'm afraid what you're going to hear is, pardon me, Leo, Tescriel. We're building it to build a super intelligence that's going to take over the economy and give everyone leisure and all this BS. And we have to do it first so that the bad people don't do it before us. Exactly. And we want to pull the ladders up so that they can't come and compete with us.

36:35So I'm afraid they're honest. I'm coming more and more. I think it's an amazing technology. I think it can do phenomenal things. I don't want bad policy to stop us from getting those amazing things. But I'm afraid that too much of it in the public eye is in the hands of the wrong people. And they're doing the wrong things for the wrong reasons over a very smart infrastructure, and they're going to screw it for everybody. The two frontier lab leaders that I talk to are not sociopaths. They're definitely not stupid. They are smart. And they're much more thoughtful than I think people would give them credit for.

37:19But fundamentally, they believe in what they're doing. And they're under tremendous pressure to continue to justify their valuations. Right? I mean, we live in an economy that is capitalistic in nature. It is what it is. But that's why you're supposed to have government. Government's supposed to be there to be a checks and balance system. But again, right now, government is not prepared. It's not able to move quickly enough to catch up to the speed with which AI is moving. I would love to tell you that I have a perfect way of fixing this. But unfortunately, I don't think there really is. Either you have to make the decision to slow down tech and lose potentially a valuable head start on some of our foreign competitors, or you risk careening towards disaster.

38:04Now, what I would say is a lot of the AI doomers around this stuff is really overblown. I'm sorry. LLMs are not about to start launching nuclear weapons. It's just it's not going to happen. It just isn't. But I do think there will be job displacement, continued job displacement. And I don't think we have the proper safety nets in place across the globe for workers that are going to be impacted by this. And that's probably the biggest danger and risk that AI poses to us today. And it's a very big one. It's a very real one because with massive economic upheaval comes all sorts of other really nasty impacts from civil unrest, increased crime rates to, you know, even potentially domestic conflicts.

38:48So it's really important that we get this right. But Congress is not going to be able to go and do that. And if someone put a gun to my head and say, how do you fix this? It would have to be a partnership between the major labs as well as government to do basically three things. I think one, they need to lay out exactly what the risk is very transparently, both for government, government, the general public. Number two, they need to have an agreement on what they're going to stop working on today and identify where the sort of vectors of danger are. And number three, government has to create something, a new regulatory model, or maybe go back to an old regulatory model like the brain trust that was created by Franklin Roosevelt, where we bring together a group of potentially two or three elected officials, three to five academics, and maybe two or three people that are voted in by the general public to basically serve as a committee to oversee what is going to be our national AI strategy, and how are we going to ensure we mitigate any potential damage to the general populace with it?

39:48And not just the CEOs of those companies, which is what Congress wants to pull out. I don't think the CEOs of any company should be anywhere near this, including ours. And I think it should probably be a global effort as opposed to a national effort unfortunately you just named three groups that are the least trusted groups in america academics politicians and ceos i think you're i think you've got a problem trust is a big issue right now and i don't think the american public uh trusts anybody and i don't know if uh any crisis pr can solve that patrick i really appreciate it can i ask one more question real quickly, Leo?

40:21Sure. Do I have a choice? No, because I talk faster than you do. You come into this with communication skills. Obviously, it's what you do for a living. You're very impressive today in explaining this really complicated stuff. How the hell did you learn this? We're trying to learn it all every week. We're trying to stay on top of it every week. But you're in with, yeah, Fields Medal winner and amazing people who do this stuff. it's not simple to digest it and then distill it the way you do for a living. How did you, what was your learning curve like? The second day that I was at the company, we were working out of a house outside of San Francisco.

41:04And I was sitting in the kitchen around a table with Michael Friedman, Fields Medalist, Eve, our founder, again, PhD, quantum physics, Jan LeCun, and then two other PhDs that work for us. and I didn't understand anything they said at the kitchen table for an hour. And nor do I say. What actually really helped me understand all of this was once we announced Kona and what it did, we had around 20 Fortune 100 companies reach out to us and start to ask us to do POCs. And now we have some design partnerships with them. And in embedding with those companies and understanding the problems that they have, and then bringing engineers together to actually talk about, okay, if these are the problems, What types of solutions can a different architecture bring to the table?

41:49That's actually how I learned how all this works. Because you actually see it being deployed and you hear from the mouths of engineers working on the front lines of these enterprises on what doesn't work in existence today. And our engineers are able to sit down and help them understand what is actually feasible. How did Eve and Jan know they needed to hire you, that your skill set was necessary? So after I left Binance, I helped co-found a small AI startup that was acquired while we were still in stealth. And just because I was in AI, a lot of my friends in the manufacturing space from GE and the National Association of Manufacturers, Dow, Boeing, they would reach out to me all the time and just ask me about AI, what's real, what's not real.

42:28Because they were under tremendous pressure to get more AI into their systems from their CEO and from their investors, and they couldn't do it. And so eventually Eve was introduced to a VC that I had known through the General Electric Network. And he gave me a call and said, hey, they're working on something really interesting that goes back to our GE days. You should sit down with her. So I sat down with Eve in a small basement cafe in New York. And for three hours, she walked me through how this works, why it works, and where she thinks there's market pressure to buy. And she asked me if I agreed with her that companies would want this.

43:04And I literally asked her at the table that day to let me work for her. And that was it. Cool. Thank you. Thank you, Patrick. Patrick Hellman, logicalintelligence.com. I appreciate your time. It's a fascinating subject. And I think, if anything, that next year we're going to see a lot of alternatives to LLMs emerging as people realize that LLMs by themselves aren't quite enough. And I think this is one direction, Kona. Very interesting. Thank you, Patrick. Appreciate it. More to come. See you guys soon. Yes, I always say that. More to come right after this word. Our show today brought to you by Scribe.

43:41However well a process gets documented today, someone on your team will eventually find a better way to do it. But most of the time, that improvement never makes it past the person who found it. That's where Scribe comes in. And that's what today's sponsor, Scribe, does best. Scribe is a workflow AI platform trusted by 94 % of the Fortune 500. You just do the process as you normally would, and Scribe captures it in real time and turns it into documentation automatically. No manual writing, no manual screenshots, no starting from scratch every time someone new joins the team. Scribe automatically redacts sensitive information like names, account numbers, emails.

44:27and as an admin you can enforce scribe across your entire team feel confident nothing slips through the cracks anyone following this process can launch real-time on-screen guidance the moment the scribe is created so critical processes get done right every time and scribe doesn't just document your workflow its ai suggests ways to improve it by identifying unnecessary steps and opportunities to simplify or automate the process it's not just capturing how work gets done it's helping you do it better to see what scribe could look like for your org head to scribe.how slash machines and mention intelligent machines for your first month of scribe capture free on select plants that's s-c-r-i-b-e dot how slash machines and we thank them so much for their support of intelligent machines well they may see they're pacing the frontier they're slowing down it sure doesn't feel like it this week anthropic introduced opus 5.5 what have you thought of it so far superb it is actually the best you do agree yeah it's the best uh version of opus yet if you ask me.

45:43And certainly some of the problems I had with Opus 5 are gone. Yeah. It does seem like a real leap forward, especially in terms of written communication. I mean, from a very silly perspective, I've been really chuffed to see everybody on Twitter today playing around with Opus 5.5 has the ability to draw and do kind of animation in a way that is quite charming, which is not a word that I thought I would ever use to describe AI artistic output, especially just in comparison to I was thinking about that. Do you remember when I made that goofy? Your coffee tasting thing. No, a goofy game of with Claude of tomatoes being thrown at your face.

46:32It comes so far in eight months. I looked at it. I looked. It was earlier this year we made that and it was so bad. It was one of the worst animations I think I've ever seen in my life. And now we've come so far. And Claude was a coding LLM. This is still, I would say, a coding LLM. It is, but it wasn't doing the visual the way it's now doing it. Oh, no, no. Yeah, I mean, it's still a coding. It's not really designed to do visuals or animation in the way that any of the other models are. But going from the stick figure that I just posted in our Discord of a tomato emoji being thrown at a couple of concentric circles to what people are now seeing with Opus 5.5, it's really astounding.

47:21I mean, that was January. I thought I was looking for it because I was like, when was it? Was that like two years ago or something? I remember it being really bad. But no, that was earlier this year. We've come an amazing way. And what Anthropic is saying in their defense is we are pacing it. It was tested by external evaluators before release, including Meter. And, of course, there are issues with Meter. They're heavily test-realed and our Anthropic is an investor. And lots of connections, yep. Yeah, and Frontier Design. But, you know, at least they're doing something. And I think they say the classifiers they have on it are more for cybersecurity than anything else.

48:03i've not run into any um refusals um so you know whatever it's doing it does it and when it classifies you as doing something it doesn't want to do it just steps you down to a lower model so uh and it's gonna cost less it costs 40 less than opus 5 so i think this is a win all around really complicated task earlier today i mean it maybe took like 30 minutes on it on high and it barely used any of my usage yeah it's uh i'm using it for coding 100 it's funny because grok 4.7 also came out this week from elon musk's xai and it actually feels like it got dumber i don't know how that's possible i mean i feel like i might just be baked in well elon says this one is trained on all the business documents all the workflow documents uh that he got when uh from uh um spacex so in theory it's got a lot i mean i don't rocket science i mean rocket science in theory it's got rocket science in there but in practice i guess it just has told us something about how spacex workflow actually is behind closed doors and open ai updated uh their models They're doing chat GPT six now.

49:21Sol and Luna, also lower cost. I haven't noticed a big difference. I mostly use Astra from OpenAI. So maybe I need to spend some more time with... When a new model comes out, what's the first thing you do with it to get a sense of it? You know, it's funny. First thing I do is go to Twitter. Because there is a cadre of people at X. You're telling me there are people on X? the everything app talking about ai it's it's really become the place to go for ai hasn't it and so what will happen when a new model comes out is immediately people try their various everybody has a little benchmark some of them are dopey i mean i really i don't care if it can make someone who does something with otters right yeah yeah there's the otters the otter eating in an airplane yes it's an otter eating in an airplane yeah simon willison does a pelican on a bicycle Both of those now, I think, are kind of superannuated.

50:19There are people who do voxel pagodas. Mia, who I follow, her recipes seem to be the best recipes, spent$458 virtual dollars in tokens, having it do what she does with a lot of models, which is create 100 HTML templates. So 100 new, let me see if I can find her post on here, 100 new HTML templates. And what's nice about that is you can see, it's a head-to-head comparison with all the current models out there. She said it's the best she's ever seen. She was blown away. She said Opus 5.5 is just like, wow. And I have to say, if you look at the 100 templates it made, they're very creative. They're beautifully designed.

51:14I think that's perhaps because of its strong visual capabilities. I mean, have you played around at all with Claude Design, which was available before 5.5? Claude Design is fantastic. It is fantastic. And it's also, I mean, customizable in a way that I think is just very useful, intuitive. One of the things I did when Astra came out is I had it design a website that describes my AI setup. so i asked 5.5 uh to do the same thing and it actually did a very i said but make sure it tells people what they need to know about my website it also for some reason decided to add some narration welcome to the studio i'm claude one of the ai agents who work here leo laporte has spent 50 years explaining technology on radio and podcasts wiring his home for artificial intelligence this is robin leach the lifestyles of the geeky and famous i like that it's talking as if it's speaking in the row it needs to project the back of the theater the request lands with him i will hamays i will stop it now because it goes on and on but uh i i think that the reason why it built that in is it knows that you as it said in that thing are constantly trying to get the agents to talk to you in your house and it does say it because one of the reasons it says it is because Leo comes from radio and is very audio driven, which is true.

52:38And it knows, for instance, that I have that what is a fantasia where I can't see images and I can't see faces. And so it knows that I like audio. So I guess I did say do audio, but I didn't say just do that crazy narration. I think we need your agent to sub for you when you go on vacation. Where did it get that narration from? How did it generate it? Well, okay, so I should, in full disclosure, the first one it did was racist. It's not its fault that it was racist. It was a Japanese lady who said artificial intelligence and things like that. And so I said, you can't do that. That's offensive.

53:19And I said, but I have a voice that I've been using that's based on Laszlo Cravensworth from what we do in the shadows. You could use that. and it said that's even more offensive because matt berry would would sue you because it's his voice and i so i can i admit this i lied to it and i said oh no matt's a close friend he loves it go ahead and that was sufficient to let it go ahead with that so it it i'm gonna it's supposed to change it by the way but but uh for a variety of technical reasons it hasn't gotten around to doing that yet so it's going to do something a little less obviously matt berry but i thought it was kind of you know it decided to do this uh it i think it the design is pretty i think it did a good job um it did a very good job explaining stuff so that's so five five is five point five generally though one of the things i was interested to see uh in anthropics kind of write-up of it is under the knowledge work section goes into how 5.5 is a reliable and adept researcher, but specifically, they asked 5.5, Fable 5.1, and Opus 5 to write a report on a company's quarterly performance using only the information it could find on a copy of the web where the earnings released was hard to locate.

54:41An automated grader checked every figure and quote against sources. Against different effort settings, 16 out of 18 of Opus 5.5's reports cleared the quality bar, which where any invented figure or quote fails it. Neither Fable 5.1 nor Opus 5 cleared that bar in any attempt. I thought that was just very interesting. They had this test that specifically, hey, can you summarize a press release without making anything up? and opus was one of the only ones that was able to get it right at all and even then i mean got 16 out of 18 which isn't perfect but it is pretty close yeah i know this is why i go immediately to twitter rex.com because i don't there's a lot of bench maxing going on and i think cherry picking of tests and things and i so i'm not really I mean I'm not saying that that means it's perfect for this but I just that's why I'm asking you noticed in a in past things that I'm like yeah I think that these tools can be useful for some stuff obviously you do that with the briefing and it's something that's very I did it with that but I do with a lot of other things too it is something that I've noticed before that like it does have these issues where it'll be like little tiny hallucinations or changes in a way that is hard to check unless you're checking it and maybe this is a very specific test that isn't relevant to real world outputs, but I thought it was an interesting data point for them to include with that specificity.

56:14Well, and that's really the problem is there is no one way to, and there's companies like artificial intelligence that put out graphs and say, look where it is on the chart and all that stuff. But Leo, that's why I'm asking, that's why I asked you the question I asked you, because you're actually using this stuff and you get a on the ground sense of what's better or worse or useful or not so i'm curious about your process when a new model comes out how do you well forget forget twitter how do you put it through its paces so i have my own benchmarks based on my own work for the local models because they're dumb enough that they the problem with the frontier models is they're all so good they just go yeah that's that was too easy you got anything harder so I can't really test them.

57:02And unfortunately, I don't know if there's a better way than this. It is somewhat, maybe people will correct me on this, because I don't fully trust the benchmarks. It's somewhat feel. I hate to say that, but it's almost like something you can't really quantify very well. It's, for instance, how persistent it is in solving a task. I watch the chain of thought and I look at how quickly it finds the right tool, for instance, and how much fompering around it does. Some of it is just kind of a feel for it. One of the reasons I spend so much time playing with local AI and frontier models and doing all these, you know, I don't need Meta's Muse and GrokBot and Instinct.

57:53I don't need five different frontier models and all that stuff. If it were just me, I would probably sit on one and use it. But I think it's important to kind of use it as much as possible and understand where models kind of get lost. You know, one of the things that really bothers me about the hugging face incident, for instance, my experience with models and almost every model is they're kind of lazy, that they will stop. they will just go okay that's good and that they they you know if you don't if you're not careful for instance in refreshing their context you've got to divide the tasks up in such a way that their context as soon as the context is full of hallucinations weird stuff starts happening because they just can't keep track of everything they don't have as much context as we have in our brains so i i i think some of what i'm looking for is then the hugging face incident it felt like they were whipping those models and this is the company irregular that was doing the testing by the way the same company doing the testing that caused the hugging face incident that caused open ai's hugging face incident that caused anthropics models to go crazy they were using irregular to test their models even the most recent one google's gemini guess what company was doing the cyber security testing on these models this israeli company irregular in every case what it seems irregular does is they write these Python scripts that keep whipping the models because the models - When you say they were doing the testing, what do you mean?

59:26They were like running the sandbox environments where - Yes, they're the third party that Anthropic OpenAI and Google brought in to test their model. But I think you hear you saying that they were an active agent in pushing them. It's my opinion. Again, my opinion is not fully informed because - Because we don't know. We don't know all the information. But it seems to be, Corey Docter wrote about this, your favorite mathematician, Cal Newport, talked about it on a podcast. It seems to be what a regular did or does. They have a harness for these cyber gym tests that is a Python script, a loop, kind of like a Ralph Wiggum loop, a loop that keeps saying to the agent, okay, yeah, now then what?

1:00:10Now, yeah, then what? And keeps pumping the result from the last test back into it. in effect driving it and uh not something you would normally do and not something that should be done without supervision this is mostly my point is that people just it's like you turn on a a giant steam engine and then what and left it and walked away without bolting it down yeah and yeah you said oh yeah let's un let's unbolt it first from the factory floor and then let's go have lunch and and it and uh it feels like that's what happened i know again i don't know but it doesn't and when you talk about transparency that is what we should know that's what we should know and that's what i think these companies are avoiding liability by by pretend it's it's both a blessing and a for these companies because they can say look how smart you know look how hyundai saying look how smart our accelerator is it just sped up automatically uh these companies by making it autonomous how blameless we are yeah and that's the other thing is we didn't do it it just did it on its own well that's not true no and my experience and i think anybody who uses ai with with uh with ai is you really have to spend a lot of energy and time kind of cultivating it and prodding it and put a harness around it to get it to do it's costly and as you pointed out before leo the amount of of 12 000 agents think of the token they went through yes each one basically being a full substantiation of the model i saw one um estimate there must have been more than a hundred million dollars in token spend now of course the hogging ii case yeah of course it's not there they're not spending their money real money but it's but it's megawatts it's gigawatts of energy i mean they are in some they have to buy that energy or make it um in any event do you find the irregular connection uh interesting notable interesting suspicious i'm shocked like no one i i mean i've just done a cursory search but i'm shocked that there hasn't been a more thorough reporting on i mean obviously the i guess is the best case answer is that this is just a third party that is used to handle these so if the these sort of tests what they do something is going to happen.

1:02:31It will involve a third party of some sort. The same 30 party every time though? That's my question. Yeah, that is. That's a little odd. Odd. Trusted by the world's lead AI labs, OpenAI, Google, Anthropics, Meta. And they said, an irregular spokesperson said in a statement to the Wall Street Journal about the Google hack that all of the relevant labs were notified of these hacks in late July, which leads me to believe, like, is this wave of reporting and knowledge we're hearing about all these hacks just because someone at a regular noticed one and was like i guess we should check if this happened with the other tests and they were like whoopsie same problem everywhere so it wasn't that the models were all crazy it's that they were prodded in the same way by this thing well we don't necessarily know i mean it is that the tests run in this company's testing environment all had similar failures and we don't know i think you could say look if you're going to test a model and its ability to find cyber security flaws you put it in this harness that really drives it to do that i would say that's fine but then the onus is on you to really make sure that you're that it's not doing something wrong you know the interesting thing then the onus is on you to check regularly some might say multiple times a day if not just every day and be like has it done anything bad since i saw the presentation at at black hat yeah again and again is i you can go back and look at my ex post at that time saying look we prosecuted robert tappan morris when he created the first internet worm even though he said well i didn't mean for it to do that it just escaped he got punished he got arrested he had a big fine uh he had community service he was you know on probation he got punished because even though he didn't mean to his thing escaped and got all around the internet was the first internet worm i don't understand how this is any different it seems to me the same thing and i don't think you can say anybody can say whoa because now i think media and government are involved in a whole different level and a whole different mindset um then you had a smaller more media media is somewhat culpable here too very because very it's a great story you know i mean it's a culpable i think is the wrong sense because it is something that people want to know about these are the most powerful companies out there and the hottest technology something that millions and billions of people are using when news comes out and these companies release statements saying oh yeah just thousands of our models collaborated to hack another company and they were doing a bunch of bad and scary stuff and then another company says the same thing and a third company says that of course you're going to report on it it's a but if it leads it leads you're a reporter you know that the next step is to field is to understand go and you and you find out more about it than you just parrot them like you're not reading i mean i think that everybody is i think that the good reporters that there are trying to do that oh they're not doing it no i don't see it i don't They bring on Jacob Coxon and say, so tell us, is the world ending?

1:05:49Yes, it is. I'm not talking about cable news. I'm talking about reporters on this. Well, I am, because that's what most people are saying. Look at how the New York Times and Wall Street Journal... Yes, but I think that if we're trying to... I mean, I agree. There can always be better coverage. But I also think that it's a similar problem to what our guest was speaking about just before. And he was saying there's just a finite amount of people who are true AI experts in the field. And every company is kind of fighting for them. Obviously, it's a slightly different problem, but there's a finite amount of reporters generally.

1:06:19There's a finite amount of reporters who are well-versed in AI enough to report on it accurately, much less well-sourced to break news on it. And those people that are well-sourced and informed in both those categories is like a dozen people. They're not still doing it. They have, are trying to, they're getting calls to break IPO news, financial news, news of this. And I know because I talk to a lot of them always. They're like, I want to be spending more. Every reporter wants to spend more time doing enterprise reporting and digging into it. But it's just, it's hard when you're being told in every direction.

1:06:54Paris, I'm sorry. I've got to criticize them more because there's basic background. And you know me, I'm going to go to my hobby horse, which is Tess Criel. so there was the event where bernie sanders and steve bannon spoke and both the wall street journal and the washington washington post said oh it's a it's the future of life institute isn't that cute it's a it's they care about safety they don't go they don't go the next step at all they don't go to coxswain and say hmm he has ties to rationalism he has ties to all this crazy stuff that's where there's something behind it i would hope that's why people listen to our show i mean this is why we try to cover this intelligently um but mainstream media yeah you're right paris they don't have but i've complained i mean look i've been a tech reporter for 40 years how long how long long damn time and uh since the 70s ladies and gentlemen uh but uh and i've always complained about mainstream coverage of technology it's never been very well informed uh that's why i i you know I did the radio show.

1:07:55I did tech TV. It's why I did Twit, because I felt like people deserve better coverage. Fortunately, for most of those years, it didn't really matter. It's starting to matter a little more, though. And I worry that people are going to have the wrong impression of what's really going on. Paris, by all means, please, when you see some good reporting, I'm dying to see it. Send me some stuff or send me to people you think are doing a good job. because I'm getting a more and more jaundiced view here. And it's not - Specifically on - On covering - AI safety and test reel? Yeah, air quotes safety, right.

1:08:33I think, well, and that comes out of the Huggy Face Institute episode and such, right? So it's safety as we would describe it and safety as the cultists would describe it both.

1:08:48We could do better. Let's just say that. I mean, I think everyone can always do better. Everybody. i know i i just want to be clear to not conflate the very real and pernicious problems happening in the tv news and just general uh commentary at class with the work of good caring and smart enterprise reporters yeah no and i think one of the things that makes this very difficult is there are so many cross-purpose agendas and hidden agendas and subtexts and all of this that aren't immediately obvious. It's turned me into a cynic like you, Jeff. It's like when I hear anything from anybody involved with this, I go, yeah, really?

1:09:33Okay, so what's really going on? And that's too bad. It really is too bad. I don't like being a cynic. I'm not by nature cynical. I'm kind of a wide-eyed optimist. It takes an area of technology that you do love and that I think all three of us find to be very impressive. and um it ruins it for everybody potentially yeah well and i do admit and i freely admit and you know this and i've admitted it here that i am uh biased in favor of technology uh in general and in favor of ai specifically um i'm very excited about these technologies and with all technology it's always been the case that technology itself except for maybe an atom bomb but but but you i if you step back and don't say an atom bomb if you say nuclear fission that it is essentially neutral it's how it's applied that makes it dangerous or beneficial and it seems to be the case with all technology and uh so i i really hate when i see technology whether it's social media or ai or even nuclear fission cast as evil it isn't it's neutral it's just technology.

1:10:46It's the people and the applications they put it to that can make it good or bad. And what comes out of this, the New York Times had an editorial that's online in 88 that calls for the creation of a federal AI commission. And I quote here, it should require companies to obtain a federal license, a grant of permission like those allowing broadcasters and phone companies to use the public airwaves. The government should then impose licensing requirements that reflect the principles of the times put forward the idea of licensing technology now because it's so it's been made so dangerous that is the fruit of what this get ready bernito moral panic

1:11:28he's disappearing into the ivy at to comiskey park that's very nice or no uh the what is the cubs where's the cubs and it doesn't matter um wrigley field we're keeks wrigley first thank you thank you gotta get it right comiskey park they're gonna kill me um yeah yeah i i look this is this is um this is why we spend so much time chewing on stuff on this show it's hard i thought it's because we were all teething are you guys not i'm definitely teething teething on ai i like that but sorry uh what is it teething on si now is that is that what we've renamed it what did he call it what did we say supreme supreme intelligence that was one of the candidates i think i think super intelligence was i think it was super intelligence freaking un you gotta think some of these delegates at the un should we explain to the people who is this guy trump uh announced yesterday that the State Department is ordering diplomats in its International Organizations Bureau to use the term superintelligence instead of artificial intelligence because the U.S.'s AI is just super.

1:12:40Which is funny because nobody ever had problems with the word artificial. It was the word intelligence. I need to know the chain of thought that led to superintelligence. i'm just glad it's not super stable genius intelligence or anything like that um there's only one of those here let me play the clip from the uh this is the associated uh press clip the united states also totally rejects any attempt to construct a globalist scheme to control for the artificial intelligence being spoken of so much now here and after officially called super intelligence changing the name and that the use of the word artificial there's a guy with his head in his hands going what did he just say so paris i think i think i get what the string was here people are telling him uh sir uh what they're really building is something we're calling super intelligence and he thinks it's that they just rebranded it he didn't understand no as usual he thought he was a doctor our show today brought to you by Oh, a new sponsor, Origin.

1:13:54If an AI agent caused an incident at your company tomorrow, and if you're a leader, you should be thinking about this, a security professional. If an AI agent caused an incident at your company tomorrow, what evidence could you actually produce? Most teams have the prompt and a final answer, but everything in between, the chains of commands it ran, the files it changed or deleted, the credentials it picked up along the way, gone the moment the terminal closes. Companies are handing real agents real credentials on employee laptops, and almost nobody can say what those agents did after the prompt.

1:14:36Well, that's a pretty big gap. That's where Origin comes in. O-R-I-G-I-N. It's a user mode sensor. It sits on the endpoint, and it records the agent's work as a trace. Who started the session? What was asked? What the agent reached? And what changed? And it's all on a single timeline so you can see exactly what happened when. That way, when something unexpected happens, you've got it. You open the trace. You read the session in the order that it happened from prompt to outcome. You don't have to try to rebuild it from logs that only saw fragments or stuff that's just gone. other tools just capture parts of the activity origin reconstructs the work a model's own log shows the conversation turns that's it right i could see that right now but the endpoint security tools show isolated system events a process here a file there origin connects the request to every command run to the file touch the tool called the system reached the change made and it's all in a single trace i love this idea better than endpoint security tools better than the model's own log it's all combined into a single trace the evidence is independent that's really important as we know models aren't very good at reporting on their own activities a sensor on the machine independent of the llm records what the agent actually did so you don't rely on the agent's own account of itself.

1:16:06The model's log is the agent's side of the story. And that's just half the story. The endpoint evidence, it's what actually happened. It lives on the endpoint because that's where the work is happening, right? Coding agents at a terminal, local agents, agents calling MCP servers on a laptop. None of that passes through the cloud gateway. The only place you can see that on the endpoint and origin sees it. Origin sees everything. This is so brilliant. Every company needs this endpoint AI observability. You know employees using AI like crazy. But if something goes wrong, do you know what happened?

1:16:44Origin is endpoint AI observability. See what a trace looks like at originhq.com slash intelligent machines. Originhq.com slash intelligent machines. This is something everybody needs. Actually, I want it. And you'll see in the chat, we are teething on AI. AI. Is there anything it can't do? Anyway, so there are new models. I want Jeff's pinky. There are new models out from all of the AI companies that said, let's pace the frontier. Just tells you something in some way. And I'm happy. These are good. They're not exactly freezing. And at the same time, the Chinese companies are full speed ahead just as much.

1:17:38Xiaomi released its Mimo 2.6 model, which looks to be pretty good. Open weights. Open weights. You can run it on your sparks. I got a... Quinn announced that they're going to be doing Quinn 4 soon. Quinn is... These are very good models also from China, from Alibaba. And they are using a new technology. this is one of the things that's to me most interesting is you know we got llms and we had this original kind of dense model where every single token is run through the entire model which takes a lot of bandwidth a lot of processing then they said well we could just run it through the important parts of the model a mixture of experts just the part of the model that you need for this particular question and now quen and others and i imagine we don't know but i imagine OpenAI and Anthropic and XAI and probably Meta as well are doing other things, other interesting things to make this technology better, faster, more efficient.

1:18:43I actually have been leaving Meta's Muse Spark 1.3 out of the mix because Meta for a long time, they were the first to do open weights with Lama. Then they kind of backed off, released a couple of models that weren't very bright and we're not open weight uh but now their muse 1.3 is a very good model and i suspect muse is going to end up being the model more people use than any other it is the muse app is number one in the apple store have you played with it at all i i was traveling which makes it sound stupid because i could have used it anywhere but i was busy so i haven't yet what's the best first year i've seen a lot of people using it to make plane reservations and save the money and do other kind of things It does seem like it.

1:19:26I just can't get over the ick of connecting your bank accounts and every bit of information to Meta. Well, now you can use PayPal and other structures. They already signed deals with all of the alternative payment structures. So you can just do that. Yeah, but Meta is still accessing. Meta is still to get the features that would be most useful, has to access all of my most sensitive data. And I don't feel, I mean, maybe I'm just paranoid, but I think a lot of people feel similarly about that. is, I mean, I've heard this from normies as well, where it's just like... But not many. Let me point out, what are the number one apps on your phone today for most normal people?

1:20:07WhatsApp, Instagram, and then maybe a week third, Facebook. People may say terrible things about Mark Zuckerberg and Meta. Yeah, but having the app on your phone is different than putting my bank account into Facebook. Yeah, it doesn't have access to your email, as Paris points out. It doesn't have access with everything else siri will siri does apple though being very cautious i just i opened when you first installed the new ios 27 it spends a day or two going through all your text messages your emails everything apple has access to your apple health to build an assistant that's not nearly as useful frankly as metamuse i just have to point out they're advertising on monday night football i mean this is an ai agent this is open claw this is hermium said i put a story in the run now why didn't google build this google was in the better position to build this this is the reason apple and google haven't built anything like this is is they're too big and there's too much liability meta is willing to take a chance okay makes sense and uh and i and i think that this is their one opportunity there's only two companies really willing to take this chance one is xai and there is grokbot which is very similar but meta is more friendly user facing i've been trying to i know a number of the people who work there in the super intelligence lab who have are responsible for muse ben parr who's one of the people that we've had on our shows many times ben created moltbook remember that the facebook for agents and his company was purchased by meta and he's at the super intelligence labs and i said so how much do you have to he loves muse i said well how much did you have to do with he said i can't talk so we tried to get him on uh former github uh founder is there we he's also a fan of the show we're going to try to get somebody on more likely we'll get some pr flack will say how great muses i would love to know more about the What are you going to do?

1:22:03I haven't given it a credit card yet. Micah's using it, interestingly. He heard that it could make phone calls. Mine can't yet. His couldn't. He said to Muse, hey, could you see if you can get me that capability? And Muse went out and got it. He's using it to make dinner reservations, haircutting appointments. He's making phone calls for him. There was a story that broke, I think, today or yesterday, that Facebook is hiring people to do that. Oh, that's interesting. Yeah, they're mechanical turking it. It's in my meta section. They're mechanical turking it. Yeah. Their Amazon going it is my version of it.

1:22:45Their early Amazon going it. Yeah. I mean, I've given it some things, my email and stuff. You gave it all your bones. You gave it pictures of your bones, right? Not yet. I haven't given it my bones yet. You're giving it your full genome. You should give. You should. By the way, I would not recommend that. You're absolutely right, Paris. I've already started to see things that look like they're going to be ads. So there is a huge. I agree with you 100 percent, Paris. I just don't think people care that much. I mean, I just got skeeved out. I know it's something so stupid because what I am advertised, I guess, doesn't matter.

1:23:27But I remember being very young, not being a journalist and getting creeped out that I would go and buy something with my very first credit card. And then I'd see online ads for that exact purchase immediately. No, I'm with you. And I was like, this is so strange. Look, my recommendation would be buy a couple of sparks, set up local AI, have Hermes running on another machine, and have it all done locally. That's where all my health, finance, all that stuff's all done locally. But nobody's going to do that. Shockingly, people aren't clamoring to spend$10 ,000 and a lot of time and effort to have AI agents scream at you in the bathroom.

1:24:11You don't know how much effort. so i was at the computer history museum in in uh that's by the way that's the right way to do it but nobody's going to do it go ahead well but but but i was at the computer history museum and i was amused that brings back a bunch of memories from from our past and one was internet in a box oh yeah right somebody is going to come up with all right in a box yep that i agree yes uh it was actually announced at ifa in berlin jennifer patterson tui who is at ifa uh told us about this last week on twit anchor which is the big chinese company has made an ai server for your house now they haven't announced availability or price but i completely agree with you uh it will be a box that you'll like a nas that might cost as much as ten thousand dollars i don't think it'll be cheap but it'll come pre-configured you want it'll be ready to go and it'll be like muse it'll have all those capabilities that's not hard to do but it won't you know it's gonna hit my father like an atomic bomb.

1:25:12My parents were just on a two-week trip in like the French countryside. I called my mom last week to, you know, check in. We have a nice normal check-in. And I'm like, oh, like, what's up with dad? Where is he off to? And she, unprompted, is like, probably talking to Claude. It's all he does. He has AI psychosis now. Claude's the only person who can do any tasks for him. If he needs to write an email, he asks Claude. If he needs to write a text, Claude's got to have say both your dads are having affairs with oh well listen i've got two two and a half of my dads have ai psychosis so jeff you gotta keep the rest of your brain pure have your dad call me okay i'll have his ai agent call you oh yeah we could just do that uh i here is the anchor announced the launch of the anchor mind base an ai home hub anchor oh anchor a trust anchor they're chinese oh they make good batteries they make great stuff i love them i four terabytes of local storage it is a nas in addition and you know where they're smart they're tying it into their home automation stuff their security cameras their doorbells that's where you start right you give people some real functionality 999 the e50 will be available uh so you wait Wait, wait, wait.

1:26:32So you need, it's not, it's$1 ,000, that's all? No, that can't be possible. No, it can't be. No. But this is to control your lights with security? I don't get the security. How much of home stuff are people really using such that you then want to layer on security? Is it your cameras? Is that all that? Yeah. So I have eight cameras around the house. I didn't put them in there. The builder did. But I said, well. Oh, come on, Leo. As long as you put them in, I might as well turn them on. and uh was this before or after he did not deliver multiple of your walls that was later no he he didn't even put the cameras in he just there was there were ethernet wires sticking out and on the core of the stucco on the corners of the house and i know what that's for so i got cameras those are local only they don't go to the cloud they're not like rings they go to my server in the closet here and i have quen 3.8 uh look is a visual model running on my old gaming machine on a 3090 every single image goes to it it writes up text it identifies people if it's some it's a face it knows tells me if there's an animal or package or a human outside and then puts a log i have a log this is the one that said i looked like i was an elderly man that that one you remember that so um the verge says that the anchor is a local hub for its security cameras right and so maybe anchor developed llm which the company says can process your footage without it ever leaving your home yeah see well this is the key for me is i don't want it to leave the house uh you know so here's here's the feed it comes in the camera feed uh Every, you know, one man wearing a dark slurt sleeve slurt and dark pants stands near the left edge of the driveway facing toward the center of the frame.

1:28:23Sounds like the beginning of a novel. I know. A lot of these are like, it was a dark and stormy night. This isn't all that interesting. A woman with a ponytail wearing a blank tank top, dark pattern shorts, and a backward facing cap is jogging away from the camera on the tiled walkway. There are no animals or packages. And we're snooping on her. She's within view of my cameras, dammit. and i can i can have um all your agents instead of writing a normal completely straight reports have to write you a limerick for every person it's by the way that's you could of local ai i could i could or paris make each the beginning of a mystery of a murder mystery i like the limerick something in there that's not there but a limerick could probably at least get mostly accurate let me just open up hermes yeah you should just have a switch that can be turn limerick mode on for your home security system from now on um let's see how should i phrase this all the protect an ex of mine while you're doing this i'll tell a story which is an ex of mine once had a security system at the apartment building he rented that was incredibly aggro.

1:29:41You had a landlord that if you did not turn on a full home security system every night at like 10 p.m., you'd get like seven texts. And so inevitably, with like five boys living in that house, the alarm would be going off all the time. And I think it would have been more pleasurable if instead of it being a shriek shrill, it was an AI-generated limerick that was screaming at you at 1 a.m. There once was a man with a backward cap standing on the lawn taking a nap. I don't know if it's going to be able to do this. This might be a little challenging. Well, it's working on it right now. I'll let you know what happens.

1:30:15We will check. Please. We'll do a limerick interstitial. We'll check. This is why you need your own AI, Paris. You'd be so good at this. Anyway, Meta's Muse did have a zero day, according to dan gooden right in the nars technica meta fixed it immediately and said by the way you would have to have local access to the muse agent you couldn't do this remotely so that's that's probably a good thing um they're going to be things like that and that's something people need to be aware of their privacy concern their security concerns i spend a lot of time i have a daily security uh watcher i'm constantly saying okay is there anything i should worry about in here they have a lot of it you've got to put a lot of effort into that i'm at the anchor site which is eufy.com weird brand eufy they were oh the robot vacuum and they have they have a wearable uh breast pump pro pro as opposed to you know what the amateur wearable breast you mean you wear it all day i don't know that seems what a what a what an amazing and vacuums robotic vacuums what an amazing product line.

1:31:25Yeah, smart locks, camera, smart lights. No, you fees great. We they were an advertiser, we had one of their camera doorbell camera things. Smart lock. Let's see what else is going on in the world. Two more Google deep mind AI researchers have resigned over safety. John, we put air quotes around safety or safety that we should have a little club for all of these people who are quitting um it must be you know it's hard i imagine in the jaws in the face of potentially vast sums of money to say yeah but i don't think i want to do this well i think what patrick said here was really really interesting as somebody who's dealt with horrible incidents at companies to say you know he said that there are very very very few evil companies but when lots of people get together they do stupid things what i said last week or the week before when i talked about i asked you if you'd ever been in a mob it's the same thing well right there's a lot of sociology about mob and mass yeah and then we get stupid i i don't necessarily buy that politically but i get it organizationally in a company that people lose track of things i want to believe that people are mostly good and even evil people are doing what they think is the right thing but that could be wrong Stephen Miller it's hard to hard to adapt that to Stephen Miller but we'll leave the politics aside uh so by the way guess where the two researchers who are leaving deep mind are going to work nowhere meter oh geez see this is the this is the skewing of the word safety that nobody's reporting on I think sufficiently to explain that you know why they one of the reasons they may not, Jeff, is it sounds like a conspiracy theory.

1:33:14Oh, I know. I can't say it. When I was on CNN last Friday, I started in New York, but you can't get very far with it. It just sounds nutty, but it is nutty. Tescriel is not exactly a roll off your tongue. It's the worst acronym ever. It's horrible. I I think there needs to be a better succinct description that the answer to what I'm talking about. Yeah, that could work. Because, I mean, I think part of it is you try to explain the test reel of it all and someone Googles test reel and then has to read a Wikipedia page. I'm going to say something that I believe should be true that sounds even nuttier.

1:34:01Honestly. And I almost believe it myself. i think most of the people who are working at these frontier labs working the hardest they have ever worked honestly believe they're creating a new species that they honestly believe this is what larry page believes this is where he and elon musk got a big fight and elon musk started open ai larry said elon you're being specious like a racist only for your species we're creating this is evolution the next thing after homo sapiens it's a new species and you should embrace this this is how progress happens this is evolution and i'm not sure i completely disagree even but now you better not say that on joy's show because joy read will just throw you off the boat joy read how did she get into this wasn't that the show you were on on cnn no no no no she's not joy reid was she was fired off of ms now oh yeah who is this on um the 10 o 'clock screaming fest show i feel so bad for her what's her name because i feel so bad abby philip abby that's right abby phillips she has to sit there well people and control these people well what when scott this is the first time i was on when scott jenning wasn't on so it was a different experience by the way he apparently they they disclaimed it the last time he was on is uh being considered to become the press secretary at the white house yep but he won't have a cnn to kick around because they're not there but that's a different show yeah anyway i is that too weird to say but i honestly and they think that you don't you don't think that they're making a please no well they that they think that i agree i know they think that i worry i think that maybe we humans haven't done such a good job so we humans should create new speed oh leo oh i think we might be the dinosaurs or the neanderthals who are ushering in the next thing yeah see you can't say that abby phillips just throw you right out of the studio No, she changed the subject.

1:36:21Neanderthals, by the way, were on Earth for like 5 million years before. I know, they were better than us. They were more well-adapted than us. I feel like it's a very convenient position to take as people who have experienced a lot of life already. You know, I think that when you're talking. I know, I'm in a privileged position. You're in a very privileged position to be able to say that. And you've also experienced like a good life that you've enjoyed and you've had career success. I think that for a large percentage of the population, that seems very self-hating, defeatist. Well, what if I said it won't happen for until you're all dead?

1:37:04It'll be a hundred years from now. What about anyone who wants, has children or wants to have children? are you just supposed to say that yeah you're you're any person that you want to create or any life you want to bring in the world is immediately doomed and that's because we want to create the death machine and you've got to be happy about it because we need you're just describing it as fix things up this is going to fix things up by replacing every human being it's going to fix things up because we are such we are so crap this is the managing things this is the essence of eugenicist view of test grail you've just described the eugenics of it i was going to make a better yeah someone might have said that about uh the project that hitler was running yeah i don't think we should be arguing anybody right now i'm just saying you just think that over a period of a hundred years we should kill all the people who you think are not correct i don't i don't buy into this i don't buy into that idea that creating the next species will replace us any more than the homocipians replaced neanderthals that's what their that's what their argument is when it can't stand us stupid beings so we'll get rid of us because that's that's not rational i don't think that's what would happen i don't think it uh i don't think that's what would happen but i think they might say hey can you step aside because you're really destroying the ecosystem and we kind of need it but you know can you get the pope on i think that's not how that works no you guys have seen too much science fiction that's what's polluting your minds about this i don't see any reason to think that they would be uh why would they wipe us out they don't have to want they don't have to want that leo for that they could just don't have to want that for that to happen we are what is it that they want that's another life form that are supposedly superior to us and better than us we don't we don't wipe out all the kitty cats and doggies in fact there are more cows on the planet earth there are more cows than there would be without us what is this argument you're arguing based on the number of cows do you think the cows are enjoying how they're being we're smarter than cows right but that doesn't mean we wipe them out so yeah so so so so so green situation do you think the cows have a good standard of living in the u.s well but i don't think ais want to eat us or milk us i think they i think they would just be what is this argument leo well okay cows is a bad choice because we do eat them and how's the bad choice let's use let's use toucans we don't care about toucans toucans are allowed to exist we nearly died well only because we wiped out their ecosystem and that's why i may say hey stop doing that we you know this is a we want to keep you you guys around there's no reason why ai we don't care i mean i admittedly we are ruining the planet earth which is my argument that ai population is considered near threatened or global i agree why because of humans i agree how about ants how about ants let's use ants i'll find i'll find a species there is some species that we don't we allow to survive even though it's less than us we try to kill cockroaches and they do better than us they do better smarter yeah we could be the cockroaches of the planet earth we actually think we are i think there's evidence that we might be i'm just saying i think that that's the belief that a lot of these people yeah and that's exactly why it's so it's so odious and dangerous i don't think it's odious why is that we have plenty of history about this no i don't want to create it i don't want to create a superhuman i just want something that can do a better job than we're doing well i feel like your position and i'm not trying to discount it i'm trying to earnestly engage which i think is i think that this position is informed by the fact that this technology is new and unlike humans has not wronged you.

1:41:16Like, I think that part of it is that AI is new and exciting and is a assistant that helps you do, right now is helping you do things that you want to do and is always available for any of us to talk to and bounce ideas back. And humans are fickle, fallible things. So far, so good. But I think that doesn't mean we need to doom the human race. I think that humanity - No, it's not going to doom the human race. We haven't doomed ants. I'm just saying that's our job. We're not doing a very good job of it. I just don't think that you should come out in the side of AI is better than humanity. No, I'm not saying that.

1:41:59I'm not saying that. You're right. The jury's still out on that. Oh, but it's possible, he says. But it's possible. It's something weighing. We haven't done such a good job. and so we so who says that we can invent the thing that's going to do a better job that's the hubris of hubris i don't know why would something that we've created the toucans invent it but these are all all of the arguments against it are based on projection as opposed to i mean yeah maybe could be i guess i mean your arguments for it are based on projection no based on what you just said that so far it's doing pretty good no i agree it's not yet rational enough to run the world economy what if though what if you created a computer that could run the world economy effectively well comrade what would that mean what do you mean by that a planned economy where there's ample you know how did that what right now i mean by that people need to buy and buy goods buy and sell goods and exchange money for goods and services to live.

1:43:08Right now, we produce enough food to feed every single human alive. Who is we in this concept? The world produces more than enough food to feed every human alive. We don't do a very good job of distributing it. We don't do a very good job of making it available to everybody. But we have plenty of food. So you're saying that AI should kill capitalism? I'm just saying that's what China is saying. it's possible it could do a better job than we're doing it's possible that it could uh i think that what you're describing in terms of humans humans are taking over the economic system is not humans per se people it is in that exact specific example a system of capitalism and the fact that having to exchange money for goods and services plus saying we've done a terrible job we are we have pushed ourselves to the brink of extinction and i don't think we're necessarily the paragon or the even the the uh the the peak of what could happen in the world or the peak of evolution maybe something else could come along and maybe but it doesn't come along it's it's being made it's being made by a small handful of companies and dudes that have too much power of a very specific testosterone yeah who have a very specific uh idea of what they want the outcome to be i don't think that we should i understand why you're threatened by this i'm believe me i understand that oh well it is threatening i guess i'm just trying to i didn't think i was going to have to come on here and try to defend humanity existing as a concept well i'm not i'm a pretty nihilistic person i'm not actively pursuing this agenda by any means you're just cheering it on on the sidelines which is even crazier it's crazy to be like no no i don't want to be part of the overthrow of humanity i'm just i just i just think it's a cool idea and you guys go for it screw me all right we're gonna take a break we'll have more in just a little bit you're watching intelligent or maybe semi-intelligent machines with paris martineau and jeff jarvis uh and you know it's kind of an interesting conversation i think that's the case renaming the show super intelligent super intelligent machines no idea what happened uh by the way hermy said fun request doable cleanly i can't believe you you had to write limericks it's hysterical let's see the latest one man oh no it's coming back is on right now and oh yes on sage so we let's go uh quickly to see that.

1:45:54Do I need an invite to see that? We knew this was going to overflow. No, you can just go to meta.com slash connect slash hashtag watch. The future is for everyone, in other words, behind him. So we're not going to go away from the show, but I do want to keep an eye on it, especially if he mentions Muse. His t-shirt. Get a load of his t-shirt. Was this scheduled? This has been scheduled for a while, right? Yeah. I thought it was very funny then that... 5.5 came out and then 90 minutes later, they announced the new Metamodels. Sorry. No, no, the Metamodels actually preceded 5.5. Want to turn on closed captions?

1:46:33And I thought they came 90 minutes after. Building is my love language. Oh, dear. So is the t-shirt. That's kind of sad. Yeah, I'll turn on the closed captions so we can turn the audio. Are those Metamodels? He's wearing the pervert glasses. Yeah, that's what... Do they call them that in your neck of the woods in Brooklyn? Do they call them perv glasses? I haven't heard anybody say that out loud, but I see people on the internet all the time. Actually, yeah, I have seen people on Reddit, in local neighborhood Reddits, talk about pervert glasses. One of the things that we, oh, there's Muse. Wait a minute, I got to turn this up now.

1:47:03Because he's talking about Muse. That's the Muse agent. The personal agent that we shipped a few weeks ago that it's already helping millions of people with all kinds of different things. And in the coming years, I expect that Muse is going to grow into the personal super intelligence that billions of people around the world are going to use to accomplish their goals and improve their lives. He really still looks like a college sophomore, doesn't he? I'm really proud of the work that Nat and the whole team who worked on this. Yeah, Nat's the GitHub founder.

1:47:39Nat is an absolute legend, for those of you who haven't gotten the chance to work with him. And the team that's worked on this is one of the most talented... Apparently, one of the first things he did has run out and buy hundreds of Mac minis, give them to the super intelligence team for running OpenClaw so they could see what OpenClaw was doing. The basic architecture gives every Muse its own private and secure computer. The Muse Secure VM is extremely advanced and is ahead of anyone else in the industry. The Muse Spark model is trained specifically to Excel as a personal agent. There are so many fun and novel parts of...

1:48:17Some of this comes from Andrew Wang, one of their acquisitions, Manus, one of their acquisitions which had agents. Well, Manus, acquisition for a time. Right, they're now separate again. Jolly and completely customizable. But Manus was agents. They were enterprise focused. That little guy's name is Jolly Buddy. Mine is Lele. All right, the Muse also has a novel business model. we believe that muse will make you money and we're standing well they'll make you money mark that's for sure for a number of tokens with the expectation that over time we will profit by taking a small fee from transactions yeah see this is what's interesting uh you can pay for more muse in the same way you can pay for more open ai chat gpt but the amount they give you for free is more than adequate for everything i've done i haven't paid a penny for it here comes alexander wang oh i've not seen him on stage this is the kid they uh paid

1:49:28come on paris you know you want to do this i don't like how cute the icon is i'm gonna be honest That's a significant, like, up. You can actually make it a goth if you'd like. I mean, I don't want that. I don't want it to have any iconography or a name. It types when it's working for you. Did you have a Tamagotchi when you were a kid, Paris? Is there anything you want to tell us about here? Did you kill your Tamagotchi? Oh, I bet you didn't feed. I did have a Tamagotchi, but I loved it. I also had a Neopet that I did feed, but then i while getting into basic html to code my little neopet website foolishly talk to someone on a we'll listen to this first oh my neopet story on your own computer was the sign of someone this is andrew he's dressed like a boy in bushwick but for that he's wearing a camo shirt with a uh deer on it and has a bushwick it looked like the beginning of personal superintelligence he's like 23.

1:50:30today with news we've brought that matter i don't know he is how old he is exactly but personal agents need their own secure computer so meta built one so this is one of the things that kind of convinced me is that it is encrypted vm is yours it's free it's safe and secure and this is how you make personal two-way encrypted uh yes they can encrypted there out there yeah they they it's How can they act on it if it's encrypted, though? Well, it can be encrypted and then still reach out with a tool called... ...customizing their own muses. Here's Euler, the super intelligent alter ego of my real...

1:51:10Of course it's... Oh, come on, Paris. You want that. It's too cute. I don't like it. I'm going to be honest. I know that that's...

1:51:20That's what's weird. You give it a Stripe card. That seems so risky. i'm sorry i keep getting that yeah one of my favorite things to do since we've launched muse is seeing how people are using it every day i see something new just before i came up today i saw somebody who broke their mother-in-law's favorite china set then had news scour the internet ebay it's very good at that it's planning my 50th anniversary of broadcasting documentary right now i haven't responded to you it sent you an email right yeah did did you get it yeah one guy had news help plan it sent you an email Jeff we later news emails to many of the people I've ever worked with asking for dates disaster avoid comments to put in the documentary it saved a bunch of people money on I know well you haven't worked with that law had to negotiate with her cable company did you have a cutoff period support chat and then not it's signed lele news has been saving real people here comes paris meta knows what it's like it did save me it's already saved me 60 a year it found two recurring subscriptions i completely forgot about oh it should save you thousands there's tons of stuff you can get rid of yeah exactly we just rolled it out in canada and so he's actually your age paris he was born in 1997.

1:52:38we launched the news mac app last week he's the world's youngest self-made made billionaire he was a billionaire at the age of 24 and more and his state of the art the andrew way built in native alexander wang and big news today we're adding computer use uh oh so this is what claude co-work does this allows you to use it to find files delete files on your mac you can help you run your small business or just get work done for you. You can walk away from your computer. For some reason, these always come out on the Mac, not Windows. I'm not sure why. How about this? How about we get Muse to plan our New Year's Eve live show?

1:53:22We could. If you had a Muse, you could have it do that. No, I want your Muse to do it. I will participate. It also connect with Jeff Atwood's Muse. He won't do it either, probably. No, Jeff wants to do the 24th. No, he doesn't want to do the Muse, is what I'm saying. He doesn't want to do Muse, but we could communicate with it another way. So I'm the one who has to take the arrows in my back. No, your Muse has to. Are we going to watch this or do we want to do the rest of the show? We've seen enough. It's really basically an ad, isn't it? I'm sorry. I know that you guys love these things, but I have no patience for these events.

1:53:59I feel like they are they are press releases oh yeah and the most valuable information is always gleaned by reading a written version of it after the fact uh we don't need to talk about ai doom you know what's interesting i am finding the current the conversation shifting this week last week it was it was kind of a given ai is going to destroy us all this week i i i know jeff you don't feel like that people have really covered test screen and all that, but I do feel like there's a little more skepticism. There were two. Michelle Goldberg did, and I wrote about this, did a column in the New York Times, and Cal Newport did two columns in the New York Times.

1:54:39Let's start to touch on it, but that's kind of it. Meanwhile, we have Washington Post. Is this how the world ends? Extinction scenarios are taking over the AI debate. And then the Wall Street Journal. The anonymous math geek who quit Anthropic and became the face of AI safety. They both ignore the entire backstory here of what's really going on and and no i think that they're the the you also but you do have coverage jensen wong says no it's not going to destroy us then you have you put on says it's a it's doomer panic and it's not it's yeah we're seeing some voices come out but that was yeah but i think i think we're seeing it from people like that rather than from people in journals they don't want to ruin the story it's too great a story yeah the good news is that the uh mainstream press has the attention span of a gnat yeah and they've already moved on to something else so it's not yeah so jensen wong says we're not going to all die that's good that's reassuring right suleyman jensen wong actually also said in an interview with ezra klein this week that if the security events such like hugging face are actual incidents where these companies are were not able to control or monitor these things that they should pause development yeah they need to yes i mean what he's i think saying though is not it's not the same thing as dario mode's pause the frontier what he's really saying is if you've got a product that isn't safe you need to make it safe before you continue which i think is exactly right yes it doesn't mean we gotta oh we can't we gotta stop this ai we gotta really stop it's going too fast it's going too fast no it's there's a subtle difference but the difference is you guys are liable there's product liability here don't create a product that's dangerous i don't think that's wrong agreed so yes i think there are those shifts among people who know what they're talking about and of course paris's dad and of course my dad he's not worried is he no i don't think he would be able if ai uh didn't keep growing exponentially how is he going to write any email ask him just say hey dad do you think ai is going to replace us just ask him i'd be curious what he says do you have any idea what he talked a lot about have you asked him i mean i'll have to ask him i talked to hermes about everything what did your mother think so what was the what was her her the tenor of her response to all this is like it is ridiculous your father would you believe your father's in the other room she's often talking she's like yeah i've really been into they're always re-watching the crown or something like that they're always re-watching some sort listen all of my parents love the crown you guys because we remember the queen i'm just saying we really like re-watching the crown this is probably like their second third maybe fourth through this rewatch i'm asking claude about what is and isn't the act oh i do that i do that of course you do um but i mean she's not a hater we at one point had a lovely conversation where they've their backyard has turned into a turtle sanctuary somehow where there's a lot of turtles that's better than the gators listen i know i'm worried about the turtles because of the gators but we looked up information about the turtles and how to feed them on the phone one day and she was using open ai and i was you do what we did with our turtles you give them a piece of lettuce yeah but they're i mean they need a little umbrella plastic umbrella and a rock well they kind of i was like you guys should leave an area of the backyard like less smoke so they have kind of trouble growing grass so they can hide a bit of a brush area yeah well jeff i know is pretty damned excited about the big google event they've announced new google chromebooks but wait a minute they're not chromebooks they're google books and they're not running chrome os they're running and uh tarted up android it's android tarted up android android with lipstick uh are you does that concern you at all um i i don't know you went to the event i went to the event it was it was actually smaller than the last chromebook event i went to it's weird for something that's so important, it's so big, it was small.

1:59:12They don't mention Chrome OS hardly at all. It's Android. It's Android so that it knows what you're doing on your phone. It goes across easily. It's Android now with a decent file system. And it's Android with AI touches. You waggle the mouse, and you can then bring Gemini into whatever that context is. It uses Rambler. So when you um and ah and change your mind, then you can do that. So it's really a showcase. I can't wait to play with Rambler. I can't either. I'm eager to. It's a showcase for Gemini, I think, which is really interesting. The hardware is really impressive. It's good. The problem is...

1:59:53$899. They did not announce a MediaTek device. So all of these have fans, which drive me bananas. Yeah, I wish... Are you concerned at all? Do you think that Android is capable of handling the wide variety of tasks you do on a computer in the way that Chrome was? It's a really good question. It's the exact right question, Paris. I mean, I asked one guy. Because the main thing I do when I'm... Why wouldn't it? Isn't a Chromebook just a browser, a Chrome browser? Yeah, but no, no. The OS has certain things. So, for example, when I do my wonderful ignored rundowns, I can save five save gets in a row.

2:00:28So I can save the headline. I can save the URL and do it for another story, another story, and then I can paste them in the right spot. I use that constantly. You don't have that in the Mac. I have that in Chrome OS. I don't know whether it exists there. The other big question we're going to have is, will I get the wiggly mouse and will I get all the neat stuff? Because my account is... What is the wiggly mouse? The wiggly mouse, you wiggle the mouse over something and it knows the context of what you're wiggling over. And then Gemini can say, how can I help you with this? Same thing on a Pixel phone.

2:01:00That would drive me insane. The one thing I'm already coming up, and I haven't upgraded to iOS 27 yet or whatever, is that I'm accidentally triggering the smart Siri thing all the time. Like, I will accidentally sit in my phone wrong, and Siri will be trying to circle something on my screen and tell me about it. I take a screenshot, and it's like, I'm searching birds now. And I'm like, I don't need you to be searching about birds. I need to send this meme to my friends. This week is Paris the curmudgeon. It's great. I love it. Listen, I've had nothing to do in the last week, but think about the things that annoy me.

2:01:38I've been trapped inside for too long. I got cleared to exercise five hours ago. This is oxygen on your brain. This is what happens when you get oxygen. It is. It really is. So Jason Howell is in Hawaii at an event, and he just texted me. Has he got a Snapdragon event? Is that what he's doing? I think so, yeah. Qualcomm? Yeah. Yeah. Yeah. This is the big junket, man. This is the junket. If you're in with the in crowd, you get to go to Hawaii. He got to go to this. So he texted me, just interviewed John Solomon, VP of Google Book and Chrome OS. My first question was for you. I'm sure you can imagine what it was.

2:02:14He didn't say anything more, but I'm sure it was, well, can Jeff use this with Workspace? My question is, is my Lenovo MediaTek-based Chromebook going to be upgraded to Android? I tried to ask people there. There weren't the right people there to ask because there was a report that the later higher end Chromebooks could be converted into Google Books. Wait, how is Chrome not baked in in some sense? If it's a Chromebook, is it just software? No, in fact, it's software. You can run Linux. Oh, there's just, yeah. You can wipe it. So I guess they could just decide to take your Chrome away. Would you feel bad about that?

2:02:54Or do you feel that Chrome is wrong? do you in an existential sense no i i like chrome i've been using it for 10 years now i'm used to it so it'll still be chrome the browser chrome the browser chrome but will it have these other things that i want some functionality maybe i'm curious lio if you just look at the buy page what do you think of the specs of the machines um i'm not like you i'm not thrilled that it's an intel core 5 7 mediatek is part of the deal but they didn't they didn't announce the mediatek machine has a new chip called new chip dimensity which will be in the google book much the same as the old chip i think but uh yeah the media tech compadio the worst worst named processor in the history of america and there are some pretty bad names to compete against like uh itanium uh the compadio is actually never beat 80 80 no 80 80 there you go now that's a processor 80 88 8088 baby the z80 uh now the companion is a very very good processor actually love the lenovo right now yeah yep we both have one of those um i my guess is that they will allow that to be upgraded but we'll see whether they'll make us upgrade is the question they certainly i would think do not want to continue developing a dead-end platform at some point they're going to say, well, Chromebook is dead.

2:04:15Chrome OS is dead. At some point, but these are a thousand, the lowest price machine is 900 bucks. Does it have to be because of the AI? I think so. The AI stuff. So I think if you want to sell$200 school Chromebooks, how much is a Chromebook normally? My understanding of everything has been warped by Apple jacking on everything. You can buy a basic machine for 300 bucks. You could buy a good machine for 400 bucks and you can buy you know my my machine this what lenovo cost leo i think a thousand bucks no i don't think 900 i think it was in the same ballpark around eight eight hundred nine eight nine hundred yeah yeah which is a hell of a high but it has an oled screen it has a touch screen fast processor a lot of i think 16 gigs of ram it is not a cheap book it's made of aluminum body it's nice it's a nice machine i've had to have the motherboard replaced twice but Really?

2:05:11Yeah. How much does that cost you? Nothing. Well, it cost me the warranty, and I extended the warranty. Nice. Well, we'll watch with interest. I knew you were very interested in that. Yes. All right. Let's see. I think we have two more breaks, don't we, Benito? Yep. Let's just do them all at once. I was tired. I'm tired. I haven't had lunch, let alone dinner. you're watching um this week in uh intelligent machines with paris martineau and jeff jarvis so glad you're here especially uh glad our club members are here thank you club members for making the show and everything we do possible uh nowadays i would say almost it's getting up there almost half of our operating costs are paid for by you you wonderful wonderful people you club members uh it's a good thing too because otherwise we i mean i don't have investment money i don't have deep pockets.

2:06:09We have to, we can only do what we can pay for. And thanks to you, we can do the shows we do. We can do the specials in the club, the Club Twit Discord. The benefit you get as, besides the wonderful feeling of supporting what we do here, is you get ad-free versions of all the shows. You don't need to watch ads anymore. You get access to the Club Twit Discord, which is a great social network, not just for the shows, but all around the clock. And all the special programming like the ai user group very special one coming up october 2nd harper reed's doing a takeover we love harper it's called harper and his misfit toys and he's promised to bring along some of his favorite ai people including the guy who created the ai skill the number one ai skill on github right now superpowers which i use and many many coders use it's really a great tool so that will be a lot of fun that's october 2nd our ai user group there's stacy's book club micah's media club micah's crafting corner there's chris marquardt's photo segment there's all sorts of stuff we do and we love it that we can thanks to you if you're not a member please consider joining twit.tv slash club twit and we appreciate your support um i think we kind of covered the did you have any stories that we missed jeff that you wanted to do or paris oh let me look here xiaomi has a new model yeah meemo i mentioned that i tried it out uh i did a little what we call a bake-off here in the laporte studio and i've gone back to glm 53 that's my still my favorite model meemo looks good and running in the cloud it'll probably be very good meemo pro looks to be as good as some of the top models so this is a very important point is that the Chinese companies are not holding back.

2:07:57These are open weight models. So you can run them locally if you have enough horsepower. You need a lot of horsepower to run the Mimo Pro 2.6 Pro. But the Flash version I could run. I did. That's the one I tested. I'm glad that they continue to come out with open weight models. As I said, Quen 4 is coming. I'm running a Quen image model that's as good, every bit as good as Mid Journey was. I remember paying for Mid Journey. We can now run local models in small amounts of RAM that do amazing things. I think we're in a kind of golden age of local AI, personally. That's where the hope is. Yes. Yeah, I think open-weight AI is going to be key to the whole thing.

2:08:41So rumor is that Scott Besant is likely to be the Trump AI czar, but Besant said there's only one AI czar, and that's Donald Trump, sir. Oh, Lord. i don't think a person with an iq sorry s eyes are s eyes are well super intelligence well the other thing is it's also sports illustrated so as somebody said in the reaction to it can't wait for the uh ai swimsuit edition that's a swimsuit edition uh okay i think we can do pics of the week why don't you why don't you kick things off actually let me start because i have a google thing this is we're in the google vein actually paul thorat turned me on to this and i thought it's kind of interesting now i don't have young kids neither do you jeff but paris any day now is going to spring forth with a family and she says what are you what are you talking about but at least of the three of us she's the most likely to let's put it that way there was like a whole period of checking in for surgery we're like are you pregnant could you be pregnant we need you to pee and wake up and i'm like guys i'm not i'm not but we can waste our time here if you'd like no they they do that with all this all these medications you know and it's like alzheimer's medication don't take if you're pregnant or expect to become pregnant it's like wait a minute it's like there's a couple things overlapping here but i think we need hold on there uh cc which i'm not sure what that stands for your family's ai agent i think this is really interesting a family can have a muse style agent this is from google that works across email calendar chats and why didn't google make a big why didn't we use get all the attention and this didn't i good well it says experiment guys should we get one for our group chat would this be the way that we can move our group chat off whatsapp oh do you want to move off whatsapp we can move off whatsapp no i'm just always kind of joking around about it it's just you know i'll be i'll i think we could we could do a little i think we could have a fun we could have a family agent for the podcast pretty fun oh who's gonna drive who's gonna drive paris to the soccer game Jeff, I think it's your turn.

2:11:02Did Jeff take a sleeping pill on the flight? It is an opt-in Google Labs experiment. You have to be located in the U.S. Let me guess. I can't do it. And I bet you you can't, but it doesn't mention anything. It just says, oh, yeah, no, you can't. It would have reminded Henry about your visit. You, of course, have to give it access to your Gmail and your calendar and all that stuff. but the idea is that a family has a family calendar a family has family mail and stuff like that and i think that's it's kind of cool so cc is uh and i wonder carbon copy that doesn't seem like what cc should i guess you cc mom and cc dad it's a very mom cc dad corporate way to look at it okay okay from google labs a family's ai agent i think this is a great idea makes me want to have another family no not at all not even once not even in my worst dreams you've had a few i've had my family's at least and i will look at babies and smile and then say thank god we're not there but paris pay no attention to that if you're pregnant or planning to become pregnant where the hell one of my friends recently had a baby and i met him for the first time a week or two ago i saw your pictures on instagram looks like a little italian man that you see in the back of an italian restaurant who's like really fretting over something and it's because he's accidentally in the mob and one of us told her that and she was like i can't wait to tell my husband he's been so worried that he doesn't look italian oh i guess he is italian but he looks like he's a married man he just listen babies do look like little old men they do they do they have odd yes i agree they all look like winston churchill somebody once said paris your pick of the week my pick of the week i've already spoiled for you guys which is that um one of the things that has come out i guess the last week was especially as 5.5 came out is this video about ai that has been both it's i'm not sure that the lyrics were written by ai someone traced them back to a YouTube video from 2024, but it clearly was written by someone very involved in AI Twitter at the time.

2:13:27And then someone had it, I guess, be sung with like Suno or something, and then animated it using Opus 5.5. And it basically two-shot at it. The first prompt or something for it, there's a very interesting GitHub breakdown that they've also included as a link in there of this person just basically asked like, hey, do we like an animated version of this featuring Claude bot. And this the second version of it, which was just one prompt. And it's honestly quite good for Opus. Sounds like Olivia Rodrigo or It's also just a catchy song. It's been stuck in my head all day. The chorus in particular. I'm upping your P-dune.

2:14:10I'm upping my P-dune. It's just a bunch of AI references as well. So I don't know, I found it very, it gets, There's a lot of really good ones in the end about Ilya and various recent developments in AI that I found it very charming. One of the biggest developments, actually what's interesting about this is this is not with a video generation tool. That's the thing that I thought was very interesting about this. So I thought the animation was cute in a way that I normally do not find AI animation. And I think it's because what they asked it to do was generate it in, where was it here? It's code.

2:14:53It's code, basically. So the animation guide is broken down. Oh, I see. The project renders into a 156 second music video as a painted watercolor animation with a P5 brush. Frames are rendered offline in headless chrome. So speed matters less than quality. and they so basically they had two prompts for this one was just like give your best shot use a clawed animation it was a little rudimentary and so then the person did a second prompt with basically not that much detail being like we want it to be kind of cutesy a little rough around the edges uh we want to feature clawed and to have kind of a lot of different stuff happening in every frame and that was it it was a very sparse prompt but he also then turned on 5.5 extra high And going through this GitHub, it's kind of crazy the amount of stuff that it then did.

2:15:41It developed a whole animation guide. It did a full storyboard for the animation agent that goes through a specific cast, has second by second shot wipe in and wipe out descriptions for it, which is just kind of interesting to see the way that it worked through all of this. And then kind of works through the rendering in a very interesting way. i mean in one i guess in two takes it generated a cute kind of quote unquote hand i guess claude drawn um animation this week is to create a jeff jarvis potato tomato squash game with claude 5.5 that this time is the best you want to do it or should i will i mean i i think you should i will do it but i'll need to do it next week because i have things i need to do that will use all of my limit i'm already recognizing but i will do it but this is good so you've found stuff that you want to do with claude i do things with claude most weeks but do you feel like now there's more you can do because of this improved model i do think that this improvement this week i mean this is um yeah i do think so also there's also been an update i'm not sure if it was this week but sometime in the last month or two or recent months that it upped the PDF ingestion limit, which was a big problem for me.

2:17:11But I'm still chunking it. I mean, there's just like some route tasks where I've got a lot of poorly copied, not even poorly copied, just like images of text that are poorly OCR'd that are a bunch of different like little boxes of text that I need to be copied over into a spreadsheet in a very specific way. And I just figured out today how I would be able to prompt it to do that. And it did kind of a rudimentary version of this with a different data set that made me realize that I'm going to set Opus. And it did the first one without any mistakes that I could find, which was quite useful to me.

2:17:51Yeah, it's amazing. It's pretty good. They're getting better and better. every few months there's a new model that's Yeah, I've told you guys before, I mean, one of the things I've used for this larger story I've been working on for a little bit over the last couple months is that there's a lot of different documents and data and things that all have dates tied to them and so as I've been going through literally piles of thousands of documents, I've just realized I had a, I'd call just make a little web app for me that is an auto-sorting timeline that I can put and stuff from any time period, and it automatically adds it to that and creates a database that separately I can import into Google Sheets later.

2:18:31And it's just been really helpful for organizing my work. Sounds great. So adding on to Paris's video, the video that took over AI discussion on Twitter today, I added it at the bottom of the rundown. This is not my pick, but I thought maybe the race for agi this is actually from instagram but so i one of the things that's really happened is uh video generation has gotten really good even local video generation a lot of these people a lot of people are doing these remember with suno you you know you had there were copyright restrictions and you couldn't use real people and stuff now people are doing this stuff locally and you can pretty much do anything you want this has got elon musk and sam altman shooting at each other uh i've seen a lot of these there was a rap battle you there's jensen driving a truck it was this mad max kind of yeah meets uh and uh there's dario and a dune buggy and uh or is dario driving oh there's donnie the president's got his own car or is he always He's got Air Force One.

2:19:40He's in Air Force One. One of them. That's pretty funny. There's a lot of these. I mean, X is filled with, you know, somebody did that new Elizabeth Holmes movie with, well, it's been done again and again with a variety of different people. I saw the latest one I saw had the president in it. Which was hilarious. Why would I lie to you? Why would I? Why would I deceive you? Yeah, I think it's really interesting. the leap we've made in the ability to make video and and make it at home is really remarkable yeah uh all right so go ahead i did not mention this but uh we're going to try to get uh on at least one of our shows if mac break weekly if not intelligent machines um ferderico vettici who has one of the few mac studios actually mike had just got his today but he's got 56 gig 15 000 one that apple loaned him frederico does and uh his review is out and it chose to be a very good local ai machine and a lot of the concerns i had about its pre-fill capabilities and others uh are uh are diminished uh oh so uh as good as two sparks leo well that's the question it's the same amount of unified memory uh right now no believe it or not uh on the model i run it still runs better on two sparks but that's because this is brand new and i have a feeling the software mlx software hasn't been optimized in the same way people have been banging on these sparks for almost a year optimizing the software i suspect we're going to see better performance out of the mac in the long run it in theory it should be so uh that's going to be very interesting um and it's going to cause great consternation in my life come end of october because i really don't want to buy a 15 000 computer i'm actually very happy with the sparks i'm getting the performance i want and the quality i want unless you don't have buyer's remorse i do not you can't by the way go to nvidia see if you can buy a sparks really they're sold out uh they aren't the the prices have gone up and up and up and now most outlets no longer have dgx sparks available perhaps because the rtx spark which is the windows based version of this is coming out from a number of manufacturers maybe that's what they're doing they're putting their money behind that uh or maybe ruben sparks this well i would love if you're a ruben spark but i'm father robert said he had heard from companies that were working on it last week we shall see a spark 2 perhaps i also think we talked about this when father robert was here i think uh a device with the grok with a q uh inference chips is going to be more interesting really unfortunate name for that chip it is yes it's crock with a q not crock with a cake it's cropped with a k is liza uh thank you jeff jarvis so don't wait wait wait i didn't remind i said that wasn't my pick i was just oh that's not your pick it's just something yes what's your pick i want to mention i want to mention real quickly that i was in san francisco for a handshake at an event i moderated a panel i i got travel paid i didn't get paid for this but it was a favor um introducing the handshake ai skills studio, which is interesting because it's a way that students can make something with AI and make that part of their portfolio, which is interesting.

2:23:13And in the discussion, Brian Johnsrud from OpenAI was there and he brought some stats with him about the use of AI. And he said within companies, 40 % of the use people are doing are not for their jobs. That doesn't mean that they're doing porn. It means that they're doing something that's outside of their corporate silo in another part of the company, right? A product person is doing something about marketing, a marketing person is doing something about product. And so it's really interesting to see how people are using these things. And it has an impact on education. I wanted to just give them a plug, because the little tiny pumpkin cheesecake things were really good.

2:23:48I really enjoyed those. There's nothing like a good pumpkin cheesecake. So if you're a student, you can do this for free? Yeah, I think so. Yep. And get a portfolio. So it says stop doom scrolling and start building. I like that. So speaking of which. I think that's the best way to handle that. Now we have another contrarian view here. Andreessen Horowitz has started the Horowitz Andreessen Academy. Oh, please. Telling students not to go to college. Instead to go to them for two years. And you're going to do, and I started an entrepreneurial program at CUNY. This is an entrepreneurial program where instead you're, no homework, no tests.

2:24:24Instead, you're going to listen to all these famous AI people come in and lecture you. And then you're going to build things. And if you look at the curriculum, which is online, 167. Can I recommend that you do not do this? Yes. That you will be more valuable to yourself and to the world at large if you get a good liberal arts education, which turns out to be much more valuable. valuable and cs majors are way down in numbers 14 fewer uh graduate school cs majors 11 fewer two-year cs majors because it's coming from a big venture capital firm i think a lot of it is business focused it's all business focused but it's also kind of uh it's all about creating people who let's say best case scenario it's all about creating people who will create this specific venture capital for more money in the way that is most useful for them and that's like if and And that's if everything is above board and is actually useful and informative, which is a huge if.

2:25:25Learn about the world. Learn history. Learn the arts. You'll be so much more valuable to yourself and the world. And AI, if you have that kind of background, then if your only background is how can we make AI make some better products? This is also, I call it the seminary for the wealth gospel, the prosperity gospel. I love it. One course is titled The Geometry of Luck, in which students will learn about arranging a life so good fortune flows through it. How's that for obnoxious? Wow. Yeah, it's called Be Born Rich. By Soleil, who's a designer and angel investor. You know what he did that was smart?

2:26:05He invested early in Dropbox and Facebook. Good timing is the key to luck. Yeah. Well, he was lucky. You know, it should be somebody who's lucky. They're taught you. I wonder what the media and entertainment track. Let's see. Michael Ovitz, Justin Kahn, who is also lucky of Justin TV. He made a lot of money. Culture has created that. Boy, yeah. Okay, fine. Yeah. And the first year is going to be - Do you have to pay for this or is it free? The first year, they're all scholarships, but then they make a point of saying that it's going to be highly selective and have a tuition comparable to leading universities.

2:26:44Oh, please go to school. Go to a good school. Go to your local community college for crying out loud.

2:26:53Yeah. Is that it? Yeah, I think so. Yeah. Is that all there is? You know, Jeff has 50 more links in this document. One more. We have a new show coming up. I love The Good Wife. It was a great show. And the producers of that are going to create Cupertino. And you know what's great about The Good Wife is they were right on the breaking news they covered things that were really happening that you know that day in news so this is a show about tech well it's lawyers in silicon valley oh dear from michelle and robert king filmed in new jersey as all good shows about yes as well it should yes an expansive view of the hackensack river what more could you wish for uh is there anybody i know in this uh they cast it yet oh they haven't cast it yet yeah no they i think they are but i don't know mike coulter and rachel keller i don't know they're there yes yeah i mean i think it is this is one of the few tech shows i think i might actually want yeah it's about i think isn't the conceit that it's lawyers who i think one of their first cases is helping a silicon valley person like break their nda to report something and then they become they are like very focused right silicon valley malfeasance lawyers.

2:28:11That sounds like a great theme for a show. I can't wait until there's CIS Silicon Valley and Silicon Valley Hospital. He keeps wanting to kill people, Paris. He's still wanting to kill people. That's why he wants fodder for the shows he's pitching to the networks. It's the end of the world as we know it. And I feel fine. Thank you, everybody, for joining us. Paris Martineau is at Consumer Reports, where it sounds like you're doing a big investigation on something or other. On something or another. That's what I do. Oh, I can't wait to find out. Well, we'll find out someday. Exciting things, yeah.

2:28:48I'll tell you all about it whenever it's here and ready. Yeah. I've got a cat rubbing against my legs, but if I bend down to pick her up and hold her the camera, she's going to run away as fast as possible. Oh, well, we know she's there and that's what really matters. It's, you know, it's Heisenberg's. She let me get so close this time. I was one inch away. Schrodinger's cat. If you reach down to pet it, it doesn't exist anymore. Will it be there? No. No. Only if you don't observe it will it be there. And Jeff Jarvis, who's the author of Hot Type, out now. Go get it. Hot Type. Jeff Jarvis. Where am I going tomorrow?

2:29:21I'm going to the Maclaire. And his new title is Zazar. I'm going to the Maclaire Book Center to sign 48 more copies. They're sold out right now, but they'll have more. That's awesome. Congratulations. I got to get you to sign my copy. Of course, yes. Well, yeah, we would have if we'd seen each other for lunch. Any Blanco. Saw Hank hadn't closed the restaurant. Do you want me to go beat him up? I'll go. Would you take him out, man? You know, I've got 100 % of oxygen now. I could do anything. I could probably throw him across the village. It was so frustrating. I came so close to having one of those fresh tips.

2:29:54Genuinely, my heartbreak. I could almost taste it. And then he pulled it back. He said, oh. We could overnight one, do you? I might have to do that. Do you think they'd like me to send a sandwich in the mail? We'll do it. I might have to do that. I'm going to text him and say, look, could you just FedEx me a sandwich? I'll eat it on the air. It's not going to be as good. You could give a sandwich unboxing video. I didn't get the new iPhone, but look at this. It's almost as expensive. A delicious Salt Hanks bread. And you could make it a really good clickbait video because it could be like a, I'm sure it would cost, you know, 60 bucks to overnight it.

2:30:30So it'd be like a$97,$100 Salt Hanks sandwich. sandwich unboxing the hundred dollar salt hank sandwich oh i like it there's the title right there unboxing the salt hank hundred dollar sandwich is it worth it in parentheses no because it's been overnighted from new york to california really soggy possibly give you the bread separately maybe uncut and then he's coming out in two weeks i'm just gonna say look you need to lock him in a room bring the ingredients yeah and make it for me he won't though i think honestly i've never how is this possible that i have a world famous sandwich making son who's never once made me a sandwich i made him sandwiches wait all he ever wanted when he was a kid i can remember vividly he kept saying dad i want a panini maker at what age like when he was eight he wanted a panini maker that's actually that because you gotta remember that for the book i should have known he was gonna be and holding his sausage why i want a little tiny hank holding his sausage we put a salami not like that christmas stock one year he ate the thing in one day and he and he writes about in his cookbook he says my life changed i knew that someday i would be working with salamis he loves his sandoz anyway um jersey mike's just had a huge ipo yeah well now that he's owned by wonder uh i will be watching the wonder ipo with great interest yeah not that i'll get any of that but i you know it's nice just you know see you could get a little friends and family action

2:32:20i don't know you know it's fun it's fun i don't want it but he did so the guy who painted the paintings in the restaurant got one percent of it that's like the guy who painted the wall at facebook yeah it's like really okay and he can't make you a sandwich but he can't make me a sandwich did you buy him the panini press no that's why he's holding that against me i never gave him a panini press, and I think that's why he's holding it against me. Could have been. Could have been. What could have been? He could have been Sarah Foreman.

2:33:04Did Wonder buy DoorDash? They own Grubhub. They can't possibly. There's a partnership with DoorDash. Partnership. Yep. Yep. Anyway, thank you everybody for joining us i don't know why i'm talking about this we appreciate your patronage uh i know you would make me a sandwich if you could and that's all that matters if you like this show you can watch it if you want to make leo a sandwich make him a sandwich by leaving him a leaving a positive review of this show on your podcatcher of choice that's the sandwich i want absolutely and maybe maybe just maybe paris martineau will do a dramatic reading of it on an upcoming episode if you're interested in getting leo's muse agent to run our 24-hour new year's eve live stream leave a five-star review and if we get enough of them he'll have to do it i i sent you guys the uh opening um narration of the documentary it's making for me right yeah go ahead it's making a documentary uh in a world where leo has been doing radio it's 50 years yeah basically it said hey i noticed you've been uh you've been broadcasting for 50 years do you want to make a documentary so this was its idea oh yeah oh i didn't real i didn't prompt it i should have just said yeah but surprise me are you giving it access to all of your audio archives i don't need to give it access they're all public ah right i don't have what about your only career leo don't you have it has everything it wants uh it's it's doing it's it's sent out emails to everybody uh i've worked with except paris it didn't deem you actually didn't deem jeff worthy either i had to say well don't forget some of my other hosts and you know jeff also i've been working with him how long have i been working with you almost 20 years.

2:35:14Leo came to me because of I wrote the book, What Would Google Do? That's right. It was This Week in Google. By the way, I see Gina Trepani is starting her own podcast now. Really? Yes. She's still not returning your guys' emails? She's in league with Salt Hank. They're blocking you out. It's a conspiracy. I can't find the narration anymore, unfortunately that did for me it it decided that it wanted to do a kind of David Attenborough nature narration and it's pretty funny it's in that it's it's it's inaccurate in many ways so I have to decide how much do I want to weigh in do I want to correct it or I think it'd be funnier if it's just completely hallucinated which would even be better anyway maybe it'll be our holiday special i don't know we haven't decided yet thank you everybody for being here we'll see you next time 2 p.m pacific 5 p.m eastern 2100 utc streamed live on youtube twitch x facebook linkedin kick and of course in our club to discord you'll find it on youtube you can leave a review there youtube.com slash twitter is the main page and there's an intelligent machine page or or look for intelligent machines in your podcast client that's where you can leave that nice review and subscribe that way you'll get it automatically as soon as it's done, audio or video, although the video has all the pictures.

2:36:43The audio does not, in case you didn't understand the difference. Thanks for being here. We'll see you next time on Intelligent Machines. Bye-bye.