Intelligent Machines 893 transcript
Please be advised that this transcript is AI-generated and may not be word-for-word. Time codes refer to the approximate times in the ad-free version of the show.
Louis Maresca [00:00:00]:
It's time for Intelligent Machines. Paris Martineau is here. Jeff Jarvis is here. And obviously I'm not Leo Laporte. I am Louis Maruska. And up next, we talk about groundbreaking research that could actually change how we understand AI, the growing threat of AI-powered cyberattacks. Plus, Anthropic has some ambitious plans to share the wealth of their networks. Some fascinating topics up next on Intelligent Machines.
TWiT.tv [00:00:25]:
Podcasts you love.
TWiT.tv [00:00:27]:
From people you trust. This
TWiT.tv [00:00:29]:
is TWiT.
Louis Maresca [00:00:34]:
This is Intelligent Machines with Jeff Jarvis and Paris Martineau, episode 893, recorded October 21st, 2026. It all comes back to cats. It's time for Intelligent Machines, where we cover the latest in super— I gotcha. We cover all the latest in AI agents, robotics, and all the smart things around us. And we do that by having amazing hosts, starting with our very own Paris Martineau. Welcome, Paris.
Paris Martineau [00:00:58]:
Hello. Thanks for— good. Always good to be here.
Louis Maresca [00:01:01]:
Yeah, thank you for being here. Now, you're an investigative reporter for Consumer Reports. And I will tell you that I follow your articles every time you put them out. In fact, we are an avid salad washer in our house. And we also—
Paris Martineau [00:01:14]:
Gotta be salad washed. I will say, that's the thing about working this job. It has changed the way I touch food before it goes in my mouth in a big way.
Louis Maresca [00:01:21]:
Yeah, at one point, you had us throwing out all of our protein powder in our house too. And I think that's a good thing from what I was seeing there in the article. So I appreciate that. Thanks
Paris Martineau [00:01:31]:
for reading.
Louis Maresca [00:01:32]:
Yeah, absolutely. Now, I think one thing that folks can't see, and maybe it's covered a bunch, but you have a bunch of interesting stuff behind you all the time. Are you constantly changing it or is it just—
Paris Martineau [00:01:41]:
I do. I mean, I'm not always constantly changing, though I have reorganized this shelf a little bit over the last 36, 48 hours because I had to, um, my colleagues very graciously nominated me for some award thing, which is wonderful. But as a consequence of it, you have to film a personal video statement, which I've realized is actually hell on earth to have to both write a personal video statement and then record it in selfie mode. You start thinking questions about what is my background? What do I do with my eyes while I'm speaking? How do you hold a phone? So I have reorganized my bookshelf slightly, but it's kind of the same. Great.
Louis Maresca [00:02:21]:
Now, I don't know if folks can't see it on the video— I mean, on the audio— but is that actually a Superhero version of Leo Laporte behind you there?
Paris Martineau [00:02:28]:
No, but it is the— It is a wonderful thing. You're gonna be glad you asked. Well, no, it is the Yes Man, a, uh, '90s or '80s era office gag gift of a I'll turn him on. My dad had this on his desk when I was growing up, where if you speak to him, he'll say things like, hello.
Yes Man [00:02:49]:
I couldn't agree with you more completely.
Paris Martineau [00:02:52]:
Here's my point. Do you have anything to say, Yes Man? No, you're done.
Yes Man [00:02:58]:
Oh yeah, I'm behind you all the way.
Paris Martineau [00:03:01]:
And stuff like that.
Louis Maresca [00:03:02]:
Fabulous. Love it. Awesome. Awesome. Well, welcome, welcome, welcome. And also we have to welcome our very own Jeff Jarvis. He's here as Emeritus Professor of Journalism Innovation at the Craig Newmark Graduate School of Journalism. Craig Newmark.
Louis Maresca [00:03:17]:
Plus he's also the—
Jeff Jarvis [00:03:18]:
You know about that, Lou?
Louis Maresca [00:03:19]:
Yeah, yeah, yeah. I've heard it. I've seen it. I've watched every episode. Absolutely. Plus you're also an author of the great hot book, Hot Type. Legendary— about the legendary machine that started the mass media. And I will say, I have you on the spot.
Louis Maresca [00:03:32]:
Can I get a signed copy? Actually, I have a copy. Can you sign my copy?
Jeff Jarvis [00:03:36]:
Oh, wow. Aren't you great? Sure. Of course. Yes.
Louis Maresca [00:03:38]:
All right. All right. I thought that might have been a hard one.
Jeff Jarvis [00:03:40]:
If folks want a signed copy, since I'm in this plug mode, if you go to jeffjarvis.com, redesigned by my son, there's a link there to the Montclair Book Center where they have signed copies and you can buy them. And please don't leave me embarrassed that they're left with a stack of them. Please. That would be awful.
Louis Maresca [00:03:55]:
I think that won't be a problem. I'm sure it's a great book. Now, are you traveling around for the book now doing—
Jeff Jarvis [00:04:02]:
Yeah, that didn't happen anymore. I went to Haverhill, Mass., to the Museum of Printing. I'm going to the Charles River Museum early next year and the Mark Twain House Museum, that kind of stuff.
Louis Maresca [00:04:12]:
Got it. Cool, cool. Well, first off, I want to thank you both for being here, and I will tell you, I want to set the expectations up front. I am certainly no Leo Laporte, and I'm usually a man of few words, so I would say that this long conversation format, I'm going to adjust it a little bit, and I think fortunately, We have the both of you here, and my plan is pretty simple. Just summarize some things and ask some questions and be part of the conversation. So hopefully that will be how we operate here.
Jeff Jarvis [00:04:35]:
Can we plug one more thing first?
Louis Maresca [00:04:37]:
Sure, go for it.
Jeff Jarvis [00:04:38]:
Tomorrow, Paris, what are you and I going to be doing with Micah?
Paris Martineau [00:04:41]:
Tomorrow we are going to be hosting— Thursday, that is. On Thursday, we're going to be doing a podcast and live show about You Can See Everything, the Elizabeth Holmes documentary by Nathan Fielder and Lance Oppenheim that has really captivated the attention, minds, and waking moments of me and Jeff since it, uh, came out. We will maybe be doing a short intro at the top that is no spoilers, but the vast majority of it is gonna be chock-full of spoilers. Spoilers, spoilers, spoilers. Which you might say, how could— as Leo did in our group chat, how can you talk— be talking about spoilers for the documentary? Well, you need to see the documentary to understand more what we're talking about, but it's, uh, well worth the discussion on both tech media, and there's no person I'd rather talk to it about than Jeff and Micah.
Jeff Jarvis [00:05:34]:
And cutting to the chase, we love it.
Paris Martineau [00:05:37]:
We do love it.
Jeff Jarvis [00:05:38]:
So you're gonna— that's our view. Take it or leave it.
Louis Maresca [00:05:42]:
Very good. Very good. When was that again? Again, we're having it
Jeff Jarvis [00:05:45]:
tomorrow at 6:30 Eastern because we don't do UTC. We don't know how to do it.
Paris Martineau [00:05:49]:
We're not sure if any other time zone exists, but if it does—
Jeff Jarvis [00:05:53]:
Because we're in the center of the freaking universe.
Paris Martineau [00:05:55]:
Yeah.
Jeff Jarvis [00:05:56]:
We got 3 East Coasters here now.
Louis Maresca [00:05:58]:
That's what I was gonna say. I feel very comfortable with having everyone on the same coast. It feels great. I'm normally on calls with people all over the world and it feels, it feels hard to figure out. I'm not sure if I'm saying good morning, good evening, good night. I get it all screwed up sometimes. So yeah, great to see everybody.
Jeff Jarvis [00:06:13]:
You just say, hey.
Louis Maresca [00:06:14]:
Hey.
Paris Martineau [00:06:14]:
Hello.
Louis Maresca [00:06:15]:
Hello. Yes. Yes. How are either of you, uh, fall time, like love the fall time? Are you big on fall?
Paris Martineau [00:06:23]:
I'm a big fall head. Are you? I did a corn maze and pumpkin patch the other weekend.
Louis Maresca [00:06:29]:
I love it.
Paris Martineau [00:06:29]:
I'm planning on maybe doing just an aimless drive up north this weekend to just gaze at some trees changing color, but I'm a little under the weather, so TBD on that. But I love fall. My 12-foot-tall skeleton has gone up in my neighborhood.
Louis Maresca [00:06:46]:
I will say that for, for me, I'm, I'm a Halloween buff too. Are you guys into Halloween?
Paris Martineau [00:06:51]:
Oh yes, Jeff's not very spirited.
Louis Maresca [00:06:55]:
Yeah, I was gonna say that. I think, you know, it's not only my birthday— trap you, Jeff— but, um, but it's, uh, also my nephew and cousin's birthday also. And we're pretty big on it here at the Maruska family. We go quite, uh, quite crazy. In fact, I'm kind of sad that I'm not on next week's episode because you would have been able to see, uh, how I did my Doctor Strange, uh, uh, beard cut, if I've screwed that up or not.
Paris Martineau [00:07:19]:
Great point. Should we dress up next week for the Halloween episode? Jeff, because I know you're so spirited, you could come as a spooky skeleton. Jeff just leaves.
Louis Maresca [00:07:32]:
He's like, I'm done.
Paris Martineau [00:07:33]:
This is like us talking about coffee. He's like, I'm not interested, actually.
Louis Maresca [00:07:37]:
The hook comes and he's out. Yeah. Fantastic. Well, cool. I think we should probably get started because there's a lot of awesome news articles out there and topics going on. In fact, I wanted to maybe just start with some of the new models that we're out out there today. In fact, not today, but recently, actually. Nvidia came up with their Neutron 3.5 SuperVL, which, of course, means nothing to folks unless they hear the fact that this thing actually can perform quite heavily even on local machines.
Louis Maresca [00:08:03]:
It's 5 times faster processing in comparison to a lot of the other models out there. It's open-weight, so you can actually retrain it if you want to. They're obviously trying to move beyond selling chips. They're trying to bring things locally so folks can have more powerful stuff. It'd be great to understand what you guys think about this, because obviously NVIDIA is kind of cornering the market here somewhat, at least from the frontier model and from the, from the processing power side of things. How do you think this is going to kind of change stuff?
Jeff Jarvis [00:08:33]:
I, I'm delighted to see— I'm a fan of Jensen Huang. I, I, as everyone who watches the show regularly knows, I'm a consumer and connoisseur of his keynotes. I was just watching his Berlin keynote. We'll talk more about that, I think, in a few minutes. And I'm— with the power that he has, I think he wields it well. on behalf of open and isn't going to let the AI boys get away with closing things down and with doomsters. And is, I think, one of the healthier spokespeople for this industry. So the fact that they put their money where their mouth is and come out with open stuff that's more powerful, I think is good.
Jeff Jarvis [00:09:12]:
He's obviously in self-interest here. He's going to try to build it around CUDA and his chips. And he showed a lot of Vera Rubin in Berlin. So that's the next crush that's coming on, is he wants more powerful things to run more efficiently on these chips.
Louis Maresca [00:09:30]:
Right.
Paris Martineau [00:09:31]:
Interesting. I mean, so can you put this into context for me? Is this the first chip of this size and power that has been kind of exclusively designed for work with OpenWeight models? And also how does that work? Is there something about this that has to be targeted towards OpenWeight that previous technology had not been?
Louis Maresca [00:09:54]:
Yeah. So I think the biggest thing is that they are focused on— they trained this thing and fine-tuned it to actually focus on agentic capabilities. So letting people run agents locally on their machine and using this agent, using this model as the main model for the agentic loop. And so they focus on fine-tuning it so it can run on these smaller machines, right? The DGX machines, or even from my understanding, it even runs on some of the RTX-based machines as well. So now you're getting down to, okay, let's try to run our agents without having to worry about cloud models, run it locally. And I feel a little bit, at least I would feel more, a little bit more comfortable with it running locally, but I'm sure that others will as well. I think it's nice to be able to see that type of technology coming out.
Paris Martineau [00:10:38]:
Luuk, also talk us through how you use AI and what your experience is with both open-weight models versus— I'm sure you work closely with Copilot and other stuff.
Louis Maresca [00:10:54]:
Sure. Yeah. For me, it's interesting because I like using Copilot and OpenCLAW because they allow you to use any model. I think that's the biggest thing. So I have a Copilot harness that runs in my house that basically just a wrapper around the GitHub Copilot SDK and essentially just runs as the main kind of agent for the house. So, and the nice thing is I can swap models in and out of there. I can do ones like, I can do the Quin model that I run locally. I can do, you know, any of the frontier models when something gets really complicated.
Louis Maresca [00:11:32]:
We have to go off and do some complicated computation that probably that model wouldn't be able to handle locally. So I would say That's the coolest thing about some of these harnesses and the capabilities of running models locally is you get a little bit more fine-grained control and it can run on consumer tech. So that's how I run it. And in fact, I have a much simpler environment than Leo has. I know he has like a bunch of— He has a bunch of agents and voices and things running throughout his house. I have more of a centralized way of doing it where all the screens in the house and the devices kind of are remoted into the single agent box. Wait, wait, wait.
Jeff Jarvis [00:12:09]:
So explain. So you're doing the whole house on AI?
Louis Maresca [00:12:13]:
I do the whole house on AI, but it's only local AI. Like, there's no remote.
Jeff Jarvis [00:12:17]:
So
Paris Martineau [00:12:17]:
what does that mean, your whole house is on AI?
Louis Maresca [00:12:20]:
Oh my goodness. Okay, so you're gonna get me in this. So I would say the first thing is all of our data is in a local vault. So that means like whether it's our calendar or health information or whatever. And then anytime the kids come home or they drop in their grades or whatever, all that stuff gets entered into the vault. And then there's a graph on top of that vault. And so anytime there's an Alexa in the house that needs to give us like an alert around, hey, you need to be at, uh, you know, baseball practice or whatever, it— the agent goes and pulls all that information and lets us know on all the devices, alerts us, sets the destination on our Teslas or stuff like that. So it's— it does quite a bit of work to keep us up to speed.
Jeff Jarvis [00:13:01]:
What does your family think of this?
Louis Maresca [00:13:03]:
Um, you know, I think at first they, they were worried about, you know, obviously onlooking, being watched, that kind of thing. Uh, but the fact that I'm a father of 5 boys, like, I kind of don't care.
Paris Martineau [00:13:15]:
Yeah, at a certain point it becomes very necessary.
Louis Maresca [00:13:18]:
That's right. So I think that's the biggest thing. But I, you know, I think now that it's running locally, it's a little bit more, you know, we're a little bit more comfortable about it.
Paris Martineau [00:13:28]:
So, but it's curious, what do you think then about, uh, products like CC, which is something that had been rolled out Google Labs as an agent built for families. It seems like a more centralized version of what you're talking about, which is obviously less secure.
Louis Maresca [00:13:42]:
Right. Yeah, I think they're all cool. I mean, Muse, CC, like, they're all very cool. But I think that's the biggest thing is people have to be comfortable with saying, okay, now everything that I— everything Google that I have, or everything Meta, or more financial, I'm gonna have to give it to them and let them process it. And of course, they all say that they are compliant or they keep it safe or whatnot. But I don't know what, what you guys think about that. But I, I've always been— I think I was on a tweet a couple weeks ago and I, and I was worried. I told Leo that I was like, I don't wanna, I don't wanna give my, my stuff to these, these companies because, uh, and then he's like, oh, they already have it.
Louis Maresca [00:14:18]:
But I know what you guys think.
Paris Martineau [00:14:20]:
That's Leo's idea of all this though. Leo is the same person that uploaded all of his medical data to MetaMuse and was wearing a pin around that recorded everything he and everyone around him said and didn't think of the fact that that meant that he was giving everyone else's data to Muse or Bee until someone pointed it out.
Jeff Jarvis [00:14:41]:
So what do you run all this on?
Louis Maresca [00:14:44]:
I run it on 2 1590s. So I have a very big beefy machine. I do 2 1590s that run on a local box here. So Windows machine.
Jeff Jarvis [00:14:52]:
So do your kids have conversations? Has the AI become a character in your family?
Louis Maresca [00:14:57]:
Yeah, unfortunately there are many characters. In fact, I was, I was going to say, even for Halloween, we built this system that when people walk by our door, we have Stephen Fry, John Malkovich, Alan Rickman, Peter Cullen, Benedict Cumberbatch. They're all speaking to the people coming up to the house.
Paris Martineau [00:15:11]:
Oh, wow.
Jeff Jarvis [00:15:12]:
Oh, Paris. Oh, Paris, I'm here for that.
Paris Martineau [00:15:14]:
How do they trigger? I'm shocked that my neighbors haven't done this yet because I live in an area where within like 2 blocks of me on either side, there is more wacky Halloween decorations that get increasingly complicated every year than you could ever possibly comprehend, but they haven't gone that far.
Jeff Jarvis [00:15:31]:
AI Halloween is going the next step.
Louis Maresca [00:15:32]:
Yeah, it is. It is. Well, the hard part about this one is I'm a little worried that they are going to misbehave because they are running on a harness that's homegrown. So there's some responsible AI built into it, but who knows, some kid comes along and tries to figure out how to hack through it. But they all are autonomous, basically, and talk to each other when you walk up. And there's no camera, but it listens and hears the kids talking. Uh, but yeah, that's basically how it's built. It's pretty cool.
Jeff Jarvis [00:16:00]:
Scares the hell out of them, which you're supposed to do on Halloween.
Louis Maresca [00:16:03]:
Yes. Well, when you hear Benedict Cumberbatch yell, it's pretty funny.
Jeff Jarvis [00:16:08]:
We need video of this. We require video
Paris Martineau [00:16:11]:
of this. Yeah, we are going to need a full video.
Jeff Jarvis [00:16:13]:
Yeah, there's an animatron. This is Bonita, by the way.
Benito Gonzalez [00:16:15]:
Sorry, sorry, real quick before we move on. Like, so you better enjoy this Halloween, Lulu, because next Halloween I'm pretty sure kids are going to start figuring out how to hack that stuff.
Louis Maresca [00:16:25]:
Yes, I know. I know. Well, they're probably already good at it. They'll probably just ask a bunch of questions to it and let them yell at each other. The AI yells at each other and then, then it's going to run off the cuff. So we'll have to see how it happens, but we tried to put some safeties into it.
Jeff Jarvis [00:16:39]:
So the Neutron runs on, as you said, uh, RTX Spark, obviously on a DGX Spark and so on.
Louis Maresca [00:16:45]:
Right.
Jeff Jarvis [00:16:46]:
Um, I don't want to jump ahead because there's more NVIDIA news, but, uh, Mu's open version It is designed to run on Apple, which I found interesting.
Louis Maresca [00:16:58]:
Right.
Paris Martineau [00:16:58]:
Have either of you guys played around with Muse at all? What has your review been, Jeff?
Jeff Jarvis [00:17:05]:
Mine's limited. What I really liked about it, uh, and I think I mentioned this briefly last week, is I'm working on another book idea, which is basically procrastination— just not write the book I'm supposed to write now.
Paris Martineau [00:17:15]:
And so of course, instead of being like a normal person procrastinating by doing something like pleasurable and low stakes, you're like, I'm going to procrastinate by working on Another book. Another book, right. An illicit book.
Jeff Jarvis [00:17:26]:
Right. Um, so I was driving home from my son's, um, and last week, and I wanted to— I was bored with the book. And so I started, uh, wanting to think about this idea. So I thought, oh, I got— I tried it with Google, but it would forget every, every response. It would continue. So, okay, let me— this is time to try Muse. And it was great. It was really amazing because it, it held on to the ideas.
Jeff Jarvis [00:17:49]:
It was additive to it. I could have a conversation. I could come back tomorrow, a week later, and have a conversation. So I really like that notion of having a, having dialogue with it.
Louis Maresca [00:17:59]:
Right.
Paris Martineau [00:18:00]:
And, uh, I'm curious as to when, so when did you do this?
Jeff Jarvis [00:18:04]:
This was a week ago.
Paris Martineau [00:18:05]:
How many messages unprompted have you gotten from Muse since then?
Jeff Jarvis [00:18:09]:
It's interesting. It tries for a couple days to draw me back in. Right. Uh, I think I mentioned last week that it, oh, I, I, I found an error when I told you yesterday, or that thing I was going to, I was going to read these poems to you yesterday and I have them now. I can read them to you when you want. It tries a couple times and gives up.
Paris Martineau [00:18:26]:
I wish that mine had that understanding of knowledge when it had been beaten, because I've received more than 20 messages without me even opening the app. And when I don't open the app for a couple days, it gets more and more desperate.
Jeff Jarvis [00:18:41]:
What did you, what did you do when you first opened it?
Paris Martineau [00:18:45]:
So I discussed this briefly last week, but I was, I did this test that I'd been like asking to do of all the AI agents, 'cause I know it's something that is possible of like, right, right. Find clothing in like 2 very specific shades of like blue or green, like with these specific parameters, like try and use it for like large-scale search. Because if you try and do this over just Google search, you just get absolute slop from whoever has paid for the most in Google Ads. Sure. I was like, I don't have enough time in my life to actually find all of the different interesting online stores I want to and then go through it specifically or hire someone to do that. So I was like, maybe I could do it. Muse probably did the best of any of them. not 100% hit rate, but it found a couple things that I thought was interesting.
Paris Martineau [00:19:28]:
And I was like, great, useful. I have since received like maybe 40 or more unsolicited responses from Muse. And it's just like, I will not respond for a couple days and then I'll be like, well, here's some more. Here's another thing. Well, you didn't even ask me about this, but like, here's the thing. I, you, I never— you haven't responded for a bit, but here's something else. And I'm just like, it is like a clingy girl.
Jeff Jarvis [00:19:49]:
Have you tried to tell it to shut up?
Paris Martineau [00:19:50]:
No, because I don't want to reward— like, clearly what it is indicating to me is this is the end goal. This is what happens when you have a company that's whole shtick has been just built on like engagement maxing. I mean, that is for a long time what all the features on Instagram and Facebook, and I assume maybe even WhatsApp, have been targeted towards, which is just like build in dark patterns or move stuff around to where people engage with your product more and more, even if it makes makes for a less pleasant user experience. I can't tell you how many times, like, a button I use often on Instagram has just moved to a different menu without any real fanfare or knowledge, just because then the button I go to click instinctively is AI or something, or something that they— a project manager clearly wants to— their job is contingent on them getting more clicks on. And it feels like this is the same thing. There was a, um, study that came out this week that I haven't found— have to find. I didn't put it in the rundown. Basically showing that the amount of people that still had the Muse app downloaded on their smartphones after it was either a week or 2 weeks after they downloaded it, it was like 23%.
Paris Martineau [00:21:03]:
Like, most people are downloading this, using it, and getting rid of it.
Jeff Jarvis [00:21:06]:
Really?
Paris Martineau [00:21:07]:
Yeah.
Jeff Jarvis [00:21:07]:
There was another figure here. Where do they have it? Uh, Muse. Um,
Benito Gonzalez [00:21:13]:
I think, Paris, your mistake was asking it for shopping advice, and that's why.
Paris Martineau [00:21:18]:
Well, that was the one thing that I thought. I was like, maybe Meta would be uniquely qualified to do this because like they do all of the brands that probably have things that I want to purchase, have uploaded their data to Meta either through Instagram or Instagram or Meta ads. They all are probably trying to hack and game this. They ostensibly have the information that would be useful. Like, I was stunned that I spent so much time on Gemini. I'm like, you have the entire internet in you. You should be able to find where I can find a royal blue pair of straight leg pants. Like, that shouldn't be that hard.
Paris Martineau [00:21:57]:
But it was. It was impossible.
Jeff Jarvis [00:21:59]:
So Meta's Muse, by the way, earlier today Jason Howell was talking about that. Every time he said Meta's Muse, he thought it's Metamucil. Um, which is not a good brand name.
Paris Martineau [00:22:08]:
Oh no,
Jeff Jarvis [00:22:09]:
no. Yeah, it clears out the pipes of the internet, you know. Um, anyway, uh, Meta's Muse has amassed 10.9 million downloads since its launch on September 8th. And, um, by contrast, US downloads for ChatGPT in the last 30 days have declined at 25%.
Louis Maresca [00:22:26]:
Right. Yeah, they said the daily active users though was actually still 2.5 million, which is actually pretty high, like engagement, I guess you could say. So I think the— but they don't, they don't talk about people install it. They don't have the information around people who uninstall it either. So I think that's probably the interesting part of that.
Jeff Jarvis [00:22:44]:
Or just, just leave.
Paris Martineau [00:22:45]:
Yeah, I'm trying to find where I found this. It was, um—
Jeff Jarvis [00:22:48]:
I was going nuts because once I started the conversation on my phone, I wanted to continue it on my computer and I couldn't figure out what account it seemed to think I was. I couldn't merge the two. I couldn't continue the conversation. And I would ask it, how do I continue the conversation? It would give me a bunch of bullshit. I'm sorry. Hey, a bunch of BS. And, um, I still haven't accomplished that.
Louis Maresca [00:23:09]:
Right, right. There's, there's some interesting things. So one of the things I'm— I, so obviously Muse is one of them. There's the Dots. Does folks use Dots?
Jeff Jarvis [00:23:19]:
I haven't used Dots yet. Have you used it? Either of you?
Louis Maresca [00:23:23]:
No, I haven't used Dots yet. There's all these obviously personals. They're all going after obviously personal assistant, consumer-level personal assistant stuff. Um,
Jeff Jarvis [00:23:32]:
you know, Muse has just got— it's the advantage it has is it has a large consumer, uh, you know, consumer base to convert.
Louis Maresca [00:23:39]:
Right. Well, they were first to turn it over too. I think they're first to get to market. So I think that's also helpful. Yeah.
Jeff Jarvis [00:23:45]:
I'm surprised. I'm shocked that Google hasn't done a better job of this, of the persistent agent. Uh, they already have my email. They already have my calendar. They already have everything. And, um, they could have You know, come up with their own cute little beast and I could have had a relationship with it.
Benito Gonzalez [00:24:03]:
And it could have been really good and then they'll cancel it in 2 years.
Jeff Jarvis [00:24:07]:
Yes, that's what my son says too.
Louis Maresca [00:24:10]:
Do they have a special name for theirs? Because I know they have— they came out with this kind of personal Gemini assistant in this past week actually.
Jeff Jarvis [00:24:18]:
They did?
Louis Maresca [00:24:19]:
Yes. Yeah, it's one that you got to give permission to everything, unfortunately, like Gmail, Calendar, Maps. services, YouTube. Uh, but it—
Paris Martineau [00:24:28]:
there's no middle option.
Louis Maresca [00:24:30]:
There's no middle option. And unfortunately too is they, they went Android first too. So I guess that
Paris Martineau [00:24:36]:
maybe this could be good for you, Jeff. Yeah, I'm gonna turn it all over. I mean, the reason it took me a while to even get it set up with Muse, because I didn't want to use it on any of my Meta-associated accounts. I like used a completely fresh email address. Smart.
Louis Maresca [00:24:50]:
Yeah.
Paris Martineau [00:24:51]:
Which I'm sure Meta can— I, I'm certain that Meta has found out that they're connected because I'm now getting Instagram ads for a lot of the products that had surfaced, which I guess is fine. I just didn't want to act like— I'm, I know that Meta is probably using every scrap of data of mine it can from every account in ways that I do have not actively consented for because this has been kind of their trend. They've been got for some version of that multiple times over the 10 years. But I was like, I don't wanna actively sign up for another thing and then also blast out a notification to everyone who follows me on Instagram that I'm using Meta Muse. So I signed up for—
Jeff Jarvis [00:25:30]:
It's also just your IP.
Paris Martineau [00:25:32]:
Yeah.
Jeff Jarvis [00:25:33]:
So, right,
Louis Maresca [00:25:35]:
right.
Jeff Jarvis [00:25:36]:
Um, go ahead.
Louis Maresca [00:25:37]:
No, no, go ahead. No,
Jeff Jarvis [00:25:38]:
no, no, I was gonna change the subject.
Louis Maresca [00:25:40]:
Oh, cool. Yeah, that— we, we just definitely— I think there— the other thing is I wanted to point out a couple other new models that came out this week. Okay. So we have some new image models that came out. which are actually pretty interesting. And it kind of gets— yeah, so we have the Flux-3, uh, Image-Klein model, which is also another open-weight model. And I think the coolest thing about all these kind of new image models that I, I kind of feel is obviously they can run on smaller, again, consumer devices. In fact, 4070 device, this particular one.
Louis Maresca [00:26:04]:
I think this is— the interesting thing about this one is it's also an editable image model. So it can actually edit model, edit images. I think this one handles up to 8 different images at a time. It can do up to 4K images. So I think the biggest thing here is just being able to run these things locally versus like having to have to go to Google Images or Google Gemini or Nana Banana, whatever they call it. And of course, all the other models that are out there that again still add the watermarks to images. But I think that kind of wanted to segue into this other part of the conversation, which is How do you guys feel about the fact that these image models are getting easier to run on local boxes and there's no watermarks?
Jeff Jarvis [00:26:47]:
Yeah, all these efforts. Yeah, yeah. But all these efforts to think that we can use a technology solution to this problem are going to go nowhere. So let's just— we're there.
Louis Maresca [00:26:56]:
Yeah, we're totally there.
Paris Martineau [00:26:58]:
I mean, Jeff, your response reminds me of a response in the movie You Can See Everything, which I guess we'll discuss tomorrow.
Jeff Jarvis [00:27:07]:
You know, I think that there is
Paris Martineau [00:27:09]:
a— And there is an element of like nihilism in how people think about these technologies where they're like, oh, well, it's all inevitable that we're all just going to be flooded by fake media or that everyone's going to take any person's image and turn it into deepfake pornography. So, you know, we might as well just not even try to stop it because it's just inevitable. It's like, no, I think that like we all should just be Trying our best to make the world slightly better, even if it does sound hard or impossible, you know. But I don't think we can— We can think about the consequences of our actions.
Jeff Jarvis [00:27:45]:
Right. But I think that comes from what you've just said. That comes from norms and shaming people and stuff like that. It doesn't come from saying, I've invented the technology that's going to solve all of this.
Paris Martineau [00:27:53]:
Yeah. I mean, but I think the people that we should be shaming should be the people who've invented this for rolling something out to the world without thinking of the consequences of it and trying to at least come up with a good faith way to try to mitigate some of those things. I mean, I don't think we should ever have the expectation that everyone's going to be perfect and come up with a way to completely improve the actions of all living human beings when using your product. But like, just to have something in there that where you're trying would be good.
Louis Maresca [00:28:24]:
Yeah, like Google. I mean, they obviously, they have the SynthID, which is, you know, pretty advanced way of adjusting pixels into a pattern of some kind. You would think that even these open weight models would at least attempt to do something similar to help do some—
Jeff Jarvis [00:28:38]:
It's the plus and minus of open weight. I mean, I'm a big proponent of open weight. Leo's now become religious on the point too. Um, but the opponents of open weight say, oh my God, you can take off all the guardrails. Yes. And I go back to my prior point. You can't presume that the guardrails are going to protect you. It's a fool's errand.
Jeff Jarvis [00:28:56]:
You should try. That's fine. But we've just got to acknowledge that these things are going to be out there. And I think it's worth the price. Personally,
Louis Maresca [00:29:04]:
I think that— I mean, you need competition. You do push these frontier companies forward. And I think the fact is you're also bringing it down downstream. You're bringing it to lowering the barrier for people to use these things, which is even more important in companies to be able to compete. I think that's one of the biggest things is you can't just have big tech competing at the AI level. Having folks to be able to fine-tune these and distill them and do what they need to do for their businesses is awesome. So that's why I love OpenWeight as well.
Jeff Jarvis [00:29:29]:
So I thought I'd put it in here, but the, um, the new— since you're on new models still, the new Metamuse is open, open weight. They put out an open weight, and as I mentioned, it's, it's built to run on 2— not count them, not 1, but 2 of the expensive Apples. And you need at least, I think, 512 in each, which you can't even buy.
Paris Martineau [00:29:52]:
How much would that even cost if you could buy them?
Jeff Jarvis [00:29:54]:
$250. I went through and I priced one, $256. What's the What's the model number for the— model name for that Apple? The Studio?
Paris Martineau [00:30:01]:
Mac Mini?
Jeff Jarvis [00:30:02]:
No, no, no, no, no.
Louis Maresca [00:30:03]:
Yeah, it's the Studio. Ultra Mac Studio?
Jeff Jarvis [00:30:05]:
It's the Studio. Studio. So I priced one, the better chip, $256, which is not enough. It needs to be $512. And again, this is 2 machines. And I think I priced it at 2 terabytes. And that came in at between $11,000 and $12,000. And you can pick it up at the store in February.
Paris Martineau [00:30:30]:
Hmm. Oh my God.
Jeff Jarvis [00:30:32]:
Yeah. So I don't think that Meta's open— I mean, good for them. They're doing an open weight model. Hallelujah. I'm glad for it. But it ain't exactly going to blow up the world.
Louis Maresca [00:30:44]:
I think from a consumer perspective, you're right. But I think some businesses might be like, wow, this is— there's an 816 billion parameter model I can go fine-tune and distill for my, for my for my work.
Jeff Jarvis [00:30:52]:
For an ad agency or stuff like that. Yeah. Right. Yeah. I think so.
Louis Maresca [00:30:56]:
Right. I mean, you have companies like, what is it? Thomson Reuters is out there doing fine-tuning QWEM models. Now that when they see this, they might actually rethink that whole process. This is actually a pretty powerful model too. It's multimodal. Yeah.
Jeff Jarvis [00:31:12]:
Of course they always, so there's, they say that it's the most powerful US open weight model. So is that them admitting that it's not as powerful as the Chinese open weight models?
Louis Maresca [00:31:21]:
It could be. I mean, this is the— was it the GLM-53 has 83 billion parameters, the 3-8 QWEN. So those are all pretty powerful. I'm actually trying to find, vigorously trying to find the benchmark list.
Jeff Jarvis [00:31:33]:
They had it in their announcement.
Louis Maresca [00:31:38]:
But yeah, I'm gonna guess it's probably way up there.
Jeff Jarvis [00:31:43]:
Benchmarks. Okay, hold on. I'll Ship this into the under meta down in my section, line 111. Leo hates it when I say lines. It breaks the fourth wall. It ruins the magic of it. Right, Paris?
Paris Martineau [00:32:06]:
If you've heard that, please wipe it from your brain.
Jeff Jarvis [00:32:11]:
Yes. So if you scroll down, there's the benchmarks. They're comparing themselves. So 61.8, that's— I don't understand any of this stuff. JobBench, MCP Atlas. I don't understand a bit of it.
Louis Maresca [00:32:23]:
Yeah, you just— some of these are, they're hard to know just by their names. Like the AI Intelligence Index is a pretty good one for agentic work. But yeah, Toolathons, there's a bunch of them in here, but I think they they, they seem in comparison, I guess you could say, like up, up to, up to par with some of these models— GLM, the Kimi K3. Uh, but those are— of course, Kimi K is a much bigger model too, so 2.8 trillion. Um, so it can't run on consumer, but the GLM-53 is right up there with it, and that one is— can be running on consumer-based hardware. So I guess that's why they say US-based. That's cool.
Jeff Jarvis [00:33:06]:
In terms of consumers, I don't think we're going to see local stuff for a while yet until it gets a lot smaller. I think it's still going to be hosted. It's like Muse. It's going to be hosted.
Louis Maresca [00:33:14]:
Right. Well, I think the biggest thing, I don't know if you guys saw, but I'm very excited about this. And I bought myself a, um, kind of my own birthday present. My wife hates when I do this, but it's the, um, the DGX, the Ultra PCs that are now coming out, the, the NVIDIA PCs. Right. Um, they're, they're, they're not, not cheap, but for, for $2,500, you get a laptop that can, can run, you know, models locally, just know, with, with the full power of, uh, NVIDIA. So
Jeff Jarvis [00:33:41]:
how much memory?
Louis Maresca [00:33:42]:
It's 60— there's 48 gigs on this one. It's unified memory, so it's not only local for the operating system but also for the GPU as well. And you can get a 64, I think, for like $3,900. So I mean, think about it, you just said $12,000 for those MEMP Studios, but this will probably run Just as powerful models.
Jeff Jarvis [00:34:04]:
Oh, it'll be— it'll— that fan will be going berserk.
Louis Maresca [00:34:07]:
Yeah, I agree. They, they try— of course, they, they gloat that they have new architecture and the cooling of it, but I think that's the thing is we'll have to see what— when people push them really hard.
Jeff Jarvis [00:34:18]:
I, I went— when I went to the event for the— this is a very— this is like the kiddie version of what you're talking about.
Jeff Jarvis [00:34:22]:
When I went to the Google Book event, And I saw the primary colored computers.
Jeff Jarvis [00:34:29]:
And I went around the room to everyone and said, does it have a fan? Yes. Does it have a fan? Yes. I hate fans.
Louis Maresca [00:34:35]:
Well, we've been on the topic of NVIDIA. I think we should definitely talk about this, this new container. Yeah, this is cool stuff. I mean, this is the drag and drop data centers essentially. Is that what it— this is the main announcement that happened today, right?
Jeff Jarvis [00:34:49]:
Leo's going to have one in his garage before I'm
Paris Martineau [00:34:52]:
sure the agents are working on it right now for him.
Jeff Jarvis [00:34:55]:
Yes. How do we, how do we grow? Yeah, it's fascinating. It's fascinating. I put a video up, but it's, as you say, it's drag and drop. It's a completely done unit. All of the compute is in it. Everything is in it. You just, you put a cooler on top.
Jeff Jarvis [00:35:07]:
And by the way, it's not water-cooled. It just releases the heat and you plug into the DC power and that's it. And so then you, uh, it reduces the install from months to 14 days. And if you have the prepared space, you just plop, plop, plop, plop, plop, plop, plop, and you got a data center. And what occurs to me is that I don't know if you need the big ugly data center that we have, that you could have, um, you know, put it in other places, other kinds of places. It's, it's, it's, it's, it's a container. You just move the container and boom, you've got a, you've got a data center and you add more of them and you got a bigger data center. If you go to the, you know, can we go to the video or no?
Benito Gonzalez [00:35:48]:
Let me find it.
Jeff Jarvis [00:35:49]:
I put it in the rundown under Nvidia. Surprise, surprise, surprise. Um, sorry, line 119.
Benito Gonzalez [00:36:01]:
I got you, I got you.
Jeff Jarvis [00:36:02]:
At the time code 30:40.
Benito Gonzalez [00:36:05]:
Oh, it doesn't let me. Oh, how? Okay.
Jeff Jarvis [00:36:07]:
It won't? It's a YouTube video. It should let you go straight. It's their page. And then you click on it, then you go to the YouTube.
Benito Gonzalez [00:36:13]:
There you go. There you go.
Louis Maresca [00:36:15]:
Oh, cool.
Jeff Jarvis [00:36:15]:
So there, there it is. Yeah. So—
Paris Martineau [00:36:17]:
Got a container full of data centers.
Jeff Jarvis [00:36:19]:
Container, right?
Louis Maresca [00:36:20]:
That's incredible.
Jeff Jarvis [00:36:21]:
Isn't it amazing? And then, so this is what they always love to do. That's the, that's the cooler on top, which lets just hot air out into the environment. Then there's the power. Um, and then boom, boom, boom, boom, boom, boom, boom. Add a whole bunch of them. You take over the world.
Louis Maresca [00:36:38]:
Unbelievable.
Jeff Jarvis [00:36:40]:
The big question here is how you get the power. Right.
Louis Maresca [00:36:44]:
Power and security. Like, I don't know how secure a container is, but if you drop them in the middle of, in the middle of the cold tundra, I guess, you know, and put a fence around it, as long as you have power,
Jeff Jarvis [00:36:56]:
you got—
Paris Martineau [00:36:56]:
Imagine how much more exciting the Ever Given container ship being stuck in the Suez Canal would've been if it was full of data centers.
Jeff Jarvis [00:37:03]:
Yes.
Louis Maresca [00:37:04]:
At least it's autonomous, right? I mean, Runs itself. That's incredible. Did they talk about at all the cost of what one of these would be?
Jeff Jarvis [00:37:13]:
No, not really.
Paris Martineau [00:37:14]:
How do you cool this?
Jeff Jarvis [00:37:17]:
The thing on top of it just draws out the heat without— so the unit itself—
Paris Martineau [00:37:24]:
So that means they can't be stacked?
Jeff Jarvis [00:37:27]:
Stacked, no.
Louis Maresca [00:37:29]:
Right.
Paris Martineau [00:37:29]:
Which does cause problems when you're thinking of containers. That's kind of their whole deal.
Jeff Jarvis [00:37:35]:
Yeah. The Vera Rubin chip itself is water-cooled, recycled water. And Jensen Huang showed off that. And then he also showed off the— there's no back filled with wires and cables. It's cableless, basically. It's pretty amazing. That's amazing. And then the connectors is fiber straight to the chip.
Jeff Jarvis [00:38:06]:
And it's phenomenal. It's just phenomenal. And I think that that provides an opportunity to rethink data centers. You raise a really good point, Paris, about the stacking. You know, my idea is that you take a mall and you turn it into a data center, which is going to happen likely in San Francisco.
Paris Martineau [00:38:27]:
You know, back in my day, they used to be taking the malls and turning them into Amazon distribution centers. But now you got data centers in there.
Jeff Jarvis [00:38:36]:
Yep.
Benito Gonzalez [00:38:37]:
Is there a physical problem with trying to stack a data center high though, because heat rises?
Paris Martineau [00:38:43]:
I mean, that's what I think, but that was the issue with these containers is because it's a contained thing inside a little metal box that is going to pose some problems if the heat, uh, if the area where you're supposed to be dispensing of the heat is the top, you can't really have that then underneath something else that is Doing the same thing. I think that's probably part of the issue. I was talking to somebody last week who, um, her brother designs basically, uh, data centers or like works with the electrical company that helps set them up. And he's like, yeah, we now— our waitlist is so many years long, we've stopped accepting clients because we don't think we have the energy capacity for it at all. Yeah. It's just not feasible.
Jeff Jarvis [00:39:30]:
I saw a story from somebody today. I didn't put it in the rundown that they have the chips, but they can't get the power.
Paris Martineau [00:39:39]:
All we need to do is turn off the power from some cities and then we'd have enough.
Jeff Jarvis [00:39:42]:
You know, stop this podcasting crap and we can have more AI.
Paris Martineau [00:39:46]:
Every podcaster needs to give up their mics so that the data centers can have more power.
Louis Maresca [00:39:52]:
There's lots of corporations that are actually saying that they have to cancel their contracts because they're trying to put them in towns that just don't have the power and the grid.
Paris Martineau [00:39:59]:
Craig M on the Discord is saying it's more efficient to push cooling into a small space versus having to cool a big warehouse and blow air. I agree, but I just am— I am concerned as to whether or not, like, the space constraints of having the, the area where the heat is evaporating from be at the top would pose structural issues if the top is blocked by another box that has to be cooled. It just seems like it's A bit of a structural nightmare. Sorry. Yeah.
Jeff Jarvis [00:40:29]:
And I was thinking you could take over some of the old Bell Telephone COs, central offices, like in New York. It's one that's, that's a high-rise with no windows. Um, but you're right. Then you'd have to vent out.
Louis Maresca [00:40:41]:
Vent out. Right. Right. Well, you just mentioned the shopping mall. I mean, I feel like that's kind of poetic to take over old shopping malls and just dump large, obviously put container data centers in them. Uh, this specific one though. Right, right. This specific one is, as you said, in San Francisco.
Jeff Jarvis [00:41:00]:
So there's a mall on Market Street. Is it Westfield Bonito, or what'd you call it? Westview?
Benito Gonzalez [00:41:06]:
Yeah, Westfield or Westgate, one of those. It's something like West something.
Jeff Jarvis [00:41:09]:
I think it's Westfield. So the big mall company is Westfield. Anyway, there's a mall on Market Street in San Francisco that's been derelict for some time now.
Paris Martineau [00:41:18]:
The San Francisco City Center Mall.
Jeff Jarvis [00:41:21]:
Yeah. Which, and by the way, putting them— the same thing happened in New York. You try to put a mall in the city, in the center of a city, it doesn't work. But anyway, it's been vacant now. And there's been debate about what to do with it. And the latest version is that it will be a data center. And then there'll be some kind of show around that, like, look, here's all your data. And the one city that I think can't complain about this should be San Francisco.
Jeff Jarvis [00:41:44]:
Given that they're doing it to the whole rest of the world. Though San Francisco being San Francisco, they'll find ways to complain.
Louis Maresca [00:41:52]:
Right. I think every city finds ways to complain. So
Jeff Jarvis [00:41:57]:
yes, I
Louis Maresca [00:41:57]:
think Seattle is the same. So although they need this, they need to probably do something similar to their malls because there's lots of malls in their areas, whether it's like strip malls or whatever, they're just empty and they really need something in them. So this just might be the next option, especially with all the corporations around them. Yeah. I mean, as long as they have the grid, the power. Yeah. It's pretty cool.
Jeff Jarvis [00:42:20]:
Just put these containers in there, put the DSX pods moving in the neighborhood near you. We talked about a story some months ago now, 3 or 4 months ago, of a builder who was going to put little versions of this in every house that he built in—
Paris Martineau [00:42:33]:
A data center in every house?
Jeff Jarvis [00:42:35]:
Like outside, like you're— Basically look like your air conditioner compressor outside, and then that becomes a distributed data center.
Paris Martineau [00:42:44]:
I mean, why do they not do that instead of having it centralized?
Jeff Jarvis [00:42:47]:
I think they've got to change their views on this. They're just not thinking creatively enough about this. I think they could have taken over any given mall that was already dead and derelict and falling apart and say, well, it's not falling apart anymore, and we have tax revenue for you. Boom. It would have been fine. I don't think people wouldn't have had the— they could have gotten away with this before NIMBY stuck.
Paris Martineau [00:43:08]:
I do think that that poses issues in areas like, I mean, this mall, this proposed data center in a mall in SF is in the middle of SF. I think that that is going to— anyone who lives near there is going to have issues with that, given what we know about how data centers impact the surrounding area, quality of life.
Jeff Jarvis [00:43:29]:
Not necessarily. As long as you, as long as you have power, you're not, you're not using, um, uh, gas turbine jets, then you're not visible. If you can get the power in and if you're not water cooling it, I don't think it's a big deal.
Benito Gonzalez [00:43:46]:
The thing about this mall though, the thing about this mall though is it's kind of like a community central, central thing to the community because, uh, there's a subway that runs under there and that mall used to have have, uh, I think SF State has classrooms in there, stuff like that. So it's kind of like a community center, and that might be it. That might be it.
Jeff Jarvis [00:44:05]:
But there's all that. But you know, when I was in San Francisco, Benito, I was really— and it's a city I used to live in and I loved, um, it's really sad. There's, there's on the corner of 5th and Market, there's a space that was a department store many, many, many years ago before I lived there in the '80s.
Paris Martineau [00:44:19]:
This is— that's the— that's what it's talking about. That's the corner of 5th and Market.
Jeff Jarvis [00:44:24]:
Well, there's, there's 2. There was one that was— has been vacant for literally 35 years. So what the hell?
Paris Martineau [00:44:30]:
It's in between 5th, 4th, and Market. Yeah, that's the place we're talking about.
Jeff Jarvis [00:44:34]:
Yeah, that's the place. Yeah, that's, that's the one. That's the Westfield. Um, you know, up the road there's a, there's a Ritz Hotel. I mean, it's trying, but it's, it's bad. Um, when I went into North Beach, which is the great old Italian neighborhood, empty storefronts, empty restaurants, all over. This is the wealthiest city we have going now, the greatest inequity, and they can't figure it out. It's pathetic.
Louis Maresca [00:45:00]:
Yeah.
Jeff Jarvis [00:45:00]:
And if any city should take data centers, it's San Francisco. If any city will complain about it, it's San Francisco. That's what makes the city the city.
Benito Gonzalez [00:45:07]:
Yeah. You lived in San Francisco long enough to know that the city is really terrible at zoning. Yeah. It doesn't know what it's doing.
Jeff Jarvis [00:45:14]:
Oh yeah. Yeah. And that's the reason that because they kept out housing, it's the reason it's so criminally expensive because it's all NIMBY. Anyway.
Louis Maresca [00:45:28]:
I think we should, I think I'd love to take, to switch just a little bit. I want to, I want to go back to one of the stories because there's one that kind of caught me and it's a little bit research heavy, but it's the AI interpretability story that we were reading. Um, I think this is a big one just for their overall research. I can give a quick summary of it because it is kind of dense. This summary is a little bit helpful for it. Researcher Nick Cammarata, he's one of OpenAI's previous researchers. He is big on interpretability. Essentially, it's fundamentally understanding how AI models think internally.
Louis Maresca [00:46:03]:
A lot of times, we don't actually know. They're black boxes. Being able to understand how the individual neurons represent certain unrelated ideas will actually help not only from training, but also understanding how models think, and then also understand why they think the way they think, whether they think badly or if they go in a particular direction. He has a different theory now, which is shifting to what the old theory was. The old theory was that we don't actually— it's kind of a spaghetti mess inside there. And they don't— we don't actually know why the model chooses the way it does. But he thinks, based off of his current research, which again, very heavy, dense paper that he wrote here, but he had just— he thinks he understands why, uh, the models eventually— he can explain why the models actually think the way they think. And he has some ideas and theories on how to start studying that and building tools to go in and actually individually understand that, which again, I feel like will almost kind of change change the landscape, if you ask me, on how people really understand models and train models.
Louis Maresca [00:47:05]:
I don't know, what do you guys think?
Jeff Jarvis [00:47:06]:
So as I read this, and I didn't read, you know, I'll be honest, like half the world, I went to Gemini and I said, summarize this for me, explain this for me. But there's a really good example in here that made it clear to me. The presumption before was these neurons are flashing at these things that seem to have nothing to do with each other. And so as I understand the theory, it was kind of saying, well, I've run out of space, so I'll stick this idea over there, and I'll stick that one over there, and they have nothing to do with anything, and it just kind of piles up, and ergo, it makes it— there's no way to figure out the logic of what's going on in there. That was the old assumption, probably badly summarised, but I try, folks. But then he found that instead, neurons— as neurons grow, they sprout offshoots. I'm reading from his paper. The way a tree branch sprouts sub-branches.
Jeff Jarvis [00:47:51]:
For example, one neuron detected I-Y, And then it might detect a name with IY in it. But then the next thing it does is, so it does DIY, and then it says, do it yourself, and do it yourself has no IY in it. So, if you look at it in one way, that makes no sense. Why is that there? But if you can track that logical sequence, which he believes he can, then you start to say, ah, this is how it's operating. And if we can get to the point where it is readable, it is, it is legible, then that opens up a lot of possibilities for understanding how they work, controlling how they work, auditing how they work, regulating how they work. I think that this is very early days. This paper is a week old, and it's just a different way to look at it. But if this is right, then the excuses that the AI boys have.
Jeff Jarvis [00:48:44]:
Oh, it's so complex what we made. We can't possibly tell you how it works. We certainly can't stop it from going rogue and destroying the Earth. So give us all this power, please. Now, if you can have this kind of legibility, I think it changes the expectations we should have of the AI boys and how they operate.
Paris Martineau [00:49:01]:
What do you guys think was the smoking gun here? I mean, what led it from— this is quite a shift from the dominant position in the field. What do you think led Nick Camaretta to kind of make— to zag where others had zigged here?
Jeff Jarvis [00:49:18]:
Well, I'm asking Gemini that question right now because I can't remember.
Paris Martineau [00:49:21]:
We could just do a podcast where we read Gemini.
Louis Maresca [00:49:24]:
Yeah, I think, I think for me, the cat
Jeff Jarvis [00:49:26]:
and car thing.
Louis Maresca [00:49:27]:
Yeah, the interesting thing is a bunch of researchers here all thought this, and so they kind of just like left it on the table and walked away and, and said, okay, that's— we're just not gonna be able to understand it. And for some reason, obviously something triggered on him and he went back and did some research on it.
Jeff Jarvis [00:49:43]:
So there's the— I'm going to read from Gemini.
Paris Martineau [00:49:46]:
What does Gemini have to say about this? And what model of—
Jeff Jarvis [00:49:48]:
Cat-car neuron, previously cited as classic evidence that— so why is a cat and a car in the same neuron? Why, if each concept comes up, why does that neuron flash? That made people say, there's no logic to this. There's no sense to it. It's inscrutable. We can't figure it out. Instead, it is a V-shaped neuron that responds to pointed features like fox faces, cat faces, car fronts, flames, and turns off for round features.
Paris Martineau [00:50:14]:
It fires for cat ears slash whiskers and car fronts because they have a geometric V pattern.
Jeff Jarvis [00:50:21]:
Interesting. So, the software was seeing a pattern that we just couldn't intuit, but it's there. And once you know it's there, then that changes things. So the canonical 2019 circuit where a car feature appeared to spread across unrelated dog neurons turned out to be a misinterpretation caused by reward hacking in early feature visualization tools. Visualized properly, the neurons were detecting shared poses across animals and vehicles rather than stashing arbitrary features. Cool, huh?
Paris Martineau [00:50:54]:
I love that it all comes back to cats.
Jeff Jarvis [00:50:56]:
It does. Yes, it does, as it should.
Paris Martineau [00:50:59]:
Cats, cars, and not dog neurons.
Louis Maresca [00:51:03]:
Right, right. Yeah, I mean, obviously it's early days, like Geoff said, but I do think that this will change. Obviously, you're going to start having researchers build tools to kind of dig in and debug into what models are doing and try to understand it better. I think that's where we'll start to see some additional improvements happen. So,
Jeff Jarvis [00:51:20]:
uh, well, the key to this too 2, to me, is this is going to require a lot of work. It's going to be important work that has real benefits. And so this should be about funding universities. This work shouldn't just occur with the AI boys and their companies. This has to be open research.
Paris Martineau [00:51:36]:
Okay, one brief thing before we get to the break.
Jeff Jarvis [00:51:37]:
We have that moment of the show when Paris laughs at something. Now we have to know what it is.
Paris Martineau [00:51:40]:
It's true, which is that, you know, I made my joke, it all comes back to cats. And some— I typed that into Gemini as I'm doing it, kind of as a bit to myself. And it gave a response which I found very pleasing, which I know is probably just the model glazing me. But in the world of AI research, yes, it almost— yes, M dash, it almost always comes back to cats. There's a running joke in computer science that the entire internet, and by extension the AI models trained on it, is just a giant machine designed to produce cat and produce and process cat pictures. Historically, that's not far from the truth. Really? truth. 2012, the Google Brain milestone.
Paris Martineau [00:52:18]:
One of the first major breakthroughs in deep learning happened when Google connected 16,000 computer processors, let them watch YouTube videos, and the system spontaneously learned to recognize a cat face without ever being told what a cat was. 2017 to 2020, the cat and car era, which we just talked about in 2016. Oh wait, I guess the 2017 to 2020 was when researchers first tried to look inside these networks, and they were convinced that the AI was too messy to understand because, Like you just said, it fired for both cat and cars. And then today, where Nick Camerata's paper kind of figures that out and is like, there's more logic to the cat firing than you'd think. It's— and then it asks at the bottom of all this, quick questions if you have time. How was the tone of this response? Just right? Too playful or too formal?
Jeff Jarvis [00:53:09]:
Tell it you preferred to— if it would have taught you in limericks.
Paris Martineau [00:53:12]:
I would have preferred that all of the responses were in limericks and prods. Yes. Love it.
Jeff Jarvis [00:53:18]:
Love it. No, so that's a fascinating paper and I think it's going to have a real impact, but I want to see this happening in the open. Yeah. I want to know what the implications are in the open. I want the money and the resources to go to universities to do this because I think that's where it has to happen.
Louis Maresca [00:53:35]:
Agreed. Agreed. Well, I think that actually segues into the other article that we have, which is actually the security cybersecurity article that we have around open open code, obviously, uh, Agent Harness, Coding Harness. They suffered a breach recently that some people actually reported they— their charges exceeded $1,700 overnight. They had agents running roughly 48 hours with large numbers. A lot of open code developers said that their security tools actually didn't even— weren't even willing to go and do the investigation in this, but they eventually received some kind of specialized cybersecurity model that had— that gave them access to that, and they were able to go and review what was going on to see what's happening. I think this is a perfect example of like where, you know, these companies are like churning these kind of harnesses out, but they— and then having customers jump on them and pay all their, all their fees and get onto their subscriptions, but they're not really necessarily, uh, verifying them at scale, and they're open to these attacks. I think this is really something I worry about when comes to harnesses and overusing AI today.
Louis Maresca [00:54:40]:
So I don't know what you guys were saying, you're using Muse and some other things. Are there some guardrails? I mean, not guardrails, but there's some governance around them to how much they can use?
Jeff Jarvis [00:54:50]:
Muse does have a budget thing. Yeah, but I'm not going to hit it yet.
Paris Martineau [00:54:54]:
Yeah, wait, what is the budget? I have seen no guardrails basically on any of it.
Jeff Jarvis [00:54:59]:
Do your harnesses, Lou, go out to the
Louis Maresca [00:55:03]:
When they, when they use the, the Frontier models, it's heavily governed. There's a heavy government. Like, there's a— I, I have tokens that it counts tokens and token usage, and then it will shut down and use a local model if it, if it uses too much.
Jeff Jarvis [00:55:17]:
Are they good boys and girls? Do they pay attention to the rules?
Louis Maresca [00:55:20]:
They do. They do, actually. It's a, it's a hard rule in the harness itself. So, I mean, the calculation might be incorrect, but I like, I like to be safe safer than sorry. So, um, but it— sometimes you can actually see when it swaps, it routes differently because it'll be thinking, it'll be thinking very clearly, and all of a sudden it'll come back with something very weird and obtuse. And, and, and, and so you're like, that's Sonnet 4.0. Yeah, that's something. That's right, exactly.
Jeff Jarvis [00:55:46]:
Who was the company that, that Accenture was token maxing like crazy? And then they— the story that was, um, a few days ago that they now limit their employees to $29 a month in tokens, which you can't do much, I would imagine, right? $29 a month in tokens? That's very low.
Louis Maresca [00:56:06]:
Yeah, that's very low. Yeah.
Paris Martineau [00:56:07]:
I mean, this does seem like the comeuppance for all these companies making a critical part of everyone's performance review how many tokens you've used. Of course, people are going to end up using hundreds of thousands of dollars tokens if you tell them that that's how they're going to— how their performance bonus is going to be determined by.
Jeff Jarvis [00:56:26]:
This is Yahoo Finance. Fixed-price contracts cover 65% of Accenture's bookings, meaning unmanaged AI token costs land directly on firm margins rather than clients. So a lot of them were spending between $500 and $1,000 a month, and they're now limited to $29, which means, okay, screw that AI stuff. Or give us some open-source models here.
Louis Maresca [00:56:47]:
Right.
Jeff Jarvis [00:56:47]:
It's
Louis Maresca [00:56:48]:
actually an interesting topic because Jensen actually has a great quote from way back in the day where he says, I'm fine with paying $100,000 a month for tokens for people, especially if they're, you know, 100 times more productive. But I think that's the key, is should we be like monitoring tokens when it really should be another measure like productivity, shipping something reliable? Like, I feel like companies should focus on those things versus the the token side of things.
Jeff Jarvis [00:57:14]:
All around, tokens seem a bad measure. Sorry, Paris, go ahead.
Paris Martineau [00:57:17]:
I'm just— exactly what you're saying. I mean, tokens should be weighed only as something against your productivity and your output in comparison to what it was like before you were spending that much on tokens.
Jeff Jarvis [00:57:30]:
The other big story of the week that's kind of related, if I may, is the PHP story.
Paris Martineau [00:57:37]:
Yeah. I mean, that's what I've been following over the last 24 hours, and I've got a couple of stories about that down online. Plus lines 172 through 174. We're going wild now that Leo isn't here.
Jeff Jarvis [00:57:49]:
Plus lines 123 to 126. So just take that, Leo.
Paris Martineau [00:57:53]:
Yeah, you know, just in case you guys want to really go crazy out there with being confused. Yeah, it seems that Sol, Opening Eyes Sol, I'm forgetting what version it was, found this issue in PHP that was a really fundamental critical bug. I mean, he wasn't the first to ever find it. Uh, researchers on PHP had kind of noted this some years before, but they're like, it's way too complicated for anyone reasonably to have taken advantage of. So it kind of ended up on the back burner, but with—
Jeff Jarvis [00:58:31]:
Oh, so it was known before?
Paris Martineau [00:58:32]:
It was known before. Yeah.
Jeff Jarvis [00:58:35]:
Ooh.
Paris Martineau [00:58:36]:
Um, let me see here. Yeah, it, uh, the Wall Street Journal spoke to some of the people people who'd done it before. And, uh, where was it? They had a really interesting thing in here. I'll find it later. But basically, from my recollection of reading this article, they had— they were aware of this, or at least some people were aware of it, but they'd kind of down-prioritized it because they were like, it is really convoluted to take advantage of this. It's probably not going to happen. But with the advent of these new models, they realized it's actually quite easy for anyone to take advantage of. And it a critical patch that needed to go up right now.
Paris Martineau [00:59:14]:
And as a result, I don't know, is this related or completely unrelated to the TLS stuff?
Jeff Jarvis [00:59:22]:
I don't know.
Louis Maresca [00:59:25]:
I mean, really, it also begs it, like, why are not all these libraries going back and looking at all the CVEs that have come back in the past and just going and running scans against all databases?
Jeff Jarvis [00:59:34]:
They've got to. So, so, so the foundation, the PHP Foundation put out a memo about all this. They said on Tuesday the 20th, they got these 2 reports circulating on online forums. But they came up with the wrong release time and the wrong score. No technical details were leaked. In response, the PHP project brought its pre-announcement forward and it moved the release of the new models or the models, the new software from October 22nd to October 21st. Distributors were still packaging the fix, so releasing immediately would have left many users without an easy way to update. Moving the release date forward by one day was the best compromise.
Jeff Jarvis [01:00:19]:
However, as a result of the leak and pre-announcement around 8 o'clock California, one of the security firms that had reported the issues mistakenly published a video that showed their proof of concept working against a number of targets and revealed revealed a few additional, though limited, details about the vulnerabilities. So that's a lot of this too, is they had to rush into it, and it is urgent, and they put out software that they're telling everybody, put it up now, and if you can't put it up now, take your site down. It's that bad.
Louis Maresca [01:00:51]:
Yeah, 70% of websites that have— I mean, that's a pretty big number for the number of installs that PHP, at least this particular PHP to Shell, has. The
Jeff Jarvis [01:01:02]:
vulnerability of open source free stuff, man. Everybody uses it.
Louis Maresca [01:01:04]:
Yeah, yeah. I mean, that's the thing is any of these high usage libraries out on the web today— I mean, this was GPT-5.6 even, which is, you know, not that powerful of a model compared to Fable and some of these other models— to be able to find these, these exploits. I would think that they're starting— they should start running it against their libraries.
Jeff Jarvis [01:01:21]:
What triggered this as far as this urgency? Do you know who found it or Or who brought it back to— it was already found, but who brought it back into the consciousness?
Paris Martineau [01:01:31]:
So the Wall Street Journal reports that Ryan Austin, a researcher at, I think, the University of California, uncovered the flaws with help from GPT-5.6 Sol, one of OpenAI's most advanced models, during an audit of PHP's source code in August. OpenAI funded that and the other audits as part of an effort by AI companies to help open-source projects address software vulnerabilities now being uncovered with all of these systems. The class of bugs, the Wall Street Journal reports, had been hiding in PHP source code for years, and a security researcher had flagged it as a problem area decades earlier.
Jeff Jarvis [01:02:06]:
Oh, decades.
Paris Martineau [01:02:07]:
But the flaw had been considered too difficult to exploit and was largely forgotten. AI found a way, though. Austin figured out how to control the code that was running in the computer's memory with a few— as few as a dozen requests to a server running PHP. On smaller systems, the hack takes just seconds. On more complex ones, it takes hours. On August 11th, uh, Duong, who I think is one of the main maintainers here, uh, called PHP— or someone from OpenAI, I guess— called PHP's Dutch and said, there's a serious issue here. The guy says, I have a moment of, oh God, what are we doing now? How do I communicate this? And then you calm down You let the adrenaline flow through and you get to fixing. Yeah.
Paris Martineau [01:02:57]:
And I guess that it seems that simultaneously researchers from Google's Wiz cybersecurity team reached out with new exploits which they'd discovered independently.
Jeff Jarvis [01:03:10]:
Oh, so there were 2 different tracks.
Paris Martineau [01:03:12]:
There were different tracks and there were findings from OpenAI's research team too. So then, and the Wall Street Journal summarizes it as, AI is enabling well-meaning security researchers to find thousands of software bugs, overwhelming projects like PHP with a flood of issues that need patching. But PHP2Shell, which is this issue we're talking about, was viewed as a priority, and AI's proficiency at hacking prompted worries that criminals could exploit it on many websites before they could be patched.
Jeff Jarvis [01:03:39]:
And because it's everywhere.
Paris Martineau [01:03:41]:
Yeah, and it was leaked before the patch date, which really what is what set a lot of this, uh, into— from it being a critical security event to being a DEF CON 1 situation. I mean, a fair number of websites were just down.
Jeff Jarvis [01:03:57]:
Yeah, that's what they were—
Paris Martineau [01:03:58]:
that's what they were recommending is basically they were saying like, take your websites down. Someone—
Jeff Jarvis [01:04:02]:
government in Europe just took their site down.
Paris Martineau [01:04:04]:
Yeah, multiple governments just took their sites down.
Jeff Jarvis [01:04:07]:
So, Lou, I'm curious about your perspective about this because you're in the biz. You're an actual expert as opposed to us.
Paris Martineau [01:04:13]:
Yeah, you could— you should have stopped us from talking like dummies about this at any point, Lou.
Louis Maresca [01:04:18]:
I'm actually listening. This is great stuff.
Jeff Jarvis [01:04:21]:
So if— because you said a few minutes ago, gee, you'd think that they'd be putting their software against these models to find this stuff. And when they do, when they get around to it, when they do it, and somebody's got to fund that work, and good for the AI companies to be funding some. Do we get to the point where we can presume the software is going to be nearly bulletproof, that at least to the capability of current models, that it's possible to get to that? Do these moments disappear? Not completely, but pretty much?
Louis Maresca [01:04:56]:
I think they thin out. I think they thin out. I think the biggest challenge here, though, is you're going to get these adversarial— you know, they're going to get hackers are going to go off and they're going to use open weight models and they're going to find ways. Right. They're— it's just a race. So they're going to continue to find things and they're going to continue to exploit them as much as they can. And then of course corporations on the other side of the house are going to have to fix them as fast as they can. So it's definitely a race and see who can use the most compute to find the things.
Louis Maresca [01:05:19]:
I do think that companies need to be doing this themselves. So they have their software bill of materials, like they need to go in and all their software is built on something, scan it and do it yourself. Like if you, if you're not wanting other corporations to do
Jeff Jarvis [01:05:32]:
it. Yeah. But let's take the case of PHP, though, because PHP, you know, or Apache, or any of these things that are open source that are out there, who's responsible for that? And the foundation doesn't have the resources, I assume. In this case, OpenAI paid it. You know, there's another story this week we'll probably get to, all these companies trying to act like they have halos over their heads. Oh, we're going to put our money in a fund to do good things. Well, one of the good things they should do is be funding this kind of security work across the board and make it a, you know, an urgent project to beat the bad guys and, and use these models in this way in an organized fashion. And if you want to regulate this technology, Washington, that's one way to do it, is to, is to put on kind of a security tax on them to make sure this work gets done.
Jeff Jarvis [01:06:22]:
Is that insane?
Louis Maresca [01:06:25]:
I don't think so. I mean, somebody's got to pay for it.
Jeff Jarvis [01:06:27]:
Yep. It's got to get done. It's got to get done. We're vulnerable.
Louis Maresca [01:06:30]:
Yeah, I mean, this— that these types of stories always kind of throw me because it's like these things have been out for so long. I'm trying to find the CVEs that were out, but it— they're not new. So
Jeff Jarvis [01:06:42]:
our friend Craig Newmark, of course, is, is very big on cybersecurity. And I think that people presume cybersecurity has this narrow definition of, you know, don't use the same password. It's this kind of stuff. It's making sure that your systems are bulletproof from what you can protect them from and prioritizing that and paying for that and putting urgency on it and getting it done. Craig, tell me if I'm wrong. No, I think you're absolutely right.
Louis Maresca [01:07:19]:
No, you—
Paris Martineau [01:07:21]:
are you guys familiar with what's going on with the TLS? Because this is not the only cybersecurity major issue that happened today. There was also a whole issue with where I think 8 different compromised TLS certificates were discovered, which had a whole other cascading effect.
Jeff Jarvis [01:07:40]:
Oh, do tell.
Paris Martineau [01:07:43]:
Listen, I'm not probably the person to explain the nitty-gritty of this. My high-level understanding is similar but very different issue in that it is a cybersecurity issue. It involved the discovery of 8 different compromised TLS certificates with like private keys so weak that they could allow basically anyone to mint trusted browser and app credentials for any site on the internet. And it kind of culminated in by sometime this morning or yesterday morning on Tuesday, a bunch of browser makers and different apps and programming languages had blocked 8 of the intermediate certificates. But I just, I felt like it was a, I mean, I don't think we have any evidence on this case as to whether these issues were found by AI, but it feels like a time where suddenly we're having all these different issues come to light. All at once. Lou, do you have any understanding of this issue or anything?
Louis Maresca [01:08:38]:
Yeah, actually, as you were reading it, I was trying to get up to it. But what it seems like is the cryptographic keys that were integers below 256, they were actually able to create automations to go out and solve these tough problems, and they were able to get key recovery from that.
Paris Martineau [01:08:54]:
Oh, I saw someone on the internet— who knows if they're right— describe it as, for a non-technical person, it's as if someone found out that that all of these different certificates, the password was just the number 4, and that's not good. And I was like, that makes sense, thank you.
Louis Maresca [01:09:12]:
Totally makes sense.
Jeff Jarvis [01:09:13]:
Yeah, this will be a massive cleanup job, bigger than Heartbleed, Filippo Valsorda, a cryptography engineer, cryptography maintainer for the open-source Go programming language, told Ars. There will be a very, very long tail of IoT devices, Docker containers, forgotten Linux machines, unmaintained CLI tools that shipped with their own copy of the roots... Heartbleed, since he brought that up, is the name of a critical vulnerability in the OpenSSL cryptography from 2014.
Louis Maresca [01:09:44]:
I think the biggest thing to call out here is it's not that the AI models are getting really good at solving math problems. It's just they're branching out and spreading it out and spreading it out to multiple agents and being able to automate it all. So I think that's the key here. You can't stop that.
Jeff Jarvis [01:10:01]:
Well, we may have new jobs for mathematicians who are worried about their unemployment.
Louis Maresca [01:10:05]:
That's right.
Jeff Jarvis [01:10:06]:
Go work on this, guys.
Louis Maresca [01:10:10]:
I think actually that gets away to the other, um, uh, recent, uh, call-out, um, which was the— around the— was it PHP? No, it was the other one. What was the other one that was in here? The cybersecurity Hmm, I can't find that one. Well, we could talk about privacy. I like this one about Google and Google Maps that recently— Google Earth, actually. It was actually court-ruled around aerial photos around Google Earth.
Jeff Jarvis [01:10:48]:
The Germans. Yes. The Germans. The Germans.
Paris Martineau [01:10:50]:
Is this related to the AI-generated Google Earth photos or something different?
Jeff Jarvis [01:10:54]:
No, this is a little bit different. So, so a bit of time machine here. here, uh, before your time here, Paris, I made fun of the Germans because they went crazy over Street View and they insisted on the right to blur the photos of their buildings taken from a public place, a public view.
Paris Martineau [01:11:14]:
Is this why all the rich people in the U.S. get their places blurred and no one else does?
Jeff Jarvis [01:11:18]:
Yes, because Google had to change the rules and all kinds of stuff. And the joke at the time was a waggish friend of mine said renamed his country Blormanie. So Street View just left Germany entirely. But then it finally, I think, basically came back in some form or another. And now the latest is, is that Google Earth has aerial views. And a court in Germany said, nein, nein, you can't do that without permission, which is kind of ridiculous. But that's where we are. So if you go to— I'm going to line 200.
Jeff Jarvis [01:11:55]:
I wrote about this in the first instance in my book, Public Parts, in which I talk— I think we talked about this, Paris— going to a sauna in Germany where in the sauna the Germans are naked and mixed. Right. And my conclusion was that it made me realize that the Germans care deeply about about the privacy of everything except their private parts.
Paris Martineau [01:12:19]:
And of course, you have to have a section of that in the book, Public Parts.
Jeff Jarvis [01:12:24]:
Yes, yes. So yeah, so now I don't know, this has been languishing forever. Google Earth has been there forever. But now Google gets accused of spying from doing this. A guy came in and commented when I made fun of this on the socials. A guy said that he, um, uh, there was a very tall crane briefly in his neighborhood. So he got a camera on top and he took pictures of the neighborhood. And then he thought he was doing the neighbors all a favor and giving them photos.
Jeff Jarvis [01:12:53]:
And they went crazy on it. You know, why didn't you tell me so I could have cleaned up my yard? You can't— there's no pleasing people, man. There's no pleasing Germans.
Louis Maresca [01:13:01]:
Right. It's like in the United States, you'd have to actually prove, prove that there's some harm being caused by this.
Jeff Jarvis [01:13:07]:
Yes.
Louis Maresca [01:13:07]:
It's interesting that in Germany, they're like, eh.
Jeff Jarvis [01:13:10]:
And there's arguments that this comes from World War II, that it comes from the Stasi. Okay, but it's also my German grandfather-in-law came over in 1923, and he was like this. If we'd say something to somebody, he would say, why do you tell anybody anything? They don't need to know that. No one needs to know anything. No, no, no, no, no. It's just Germany. It's the German character, I think. They don't like to be public in this way.
Jeff Jarvis [01:13:38]:
So that's a pain in the butt for Google Earth. If you want to try to use Google Earth and Google Maps and try to understand where you are, what is it? Are they going to have to turn it off now because one schmuck doesn't want his backyard barbecue picture taken?
Louis Maresca [01:13:55]:
I mean, this goes back to having to, you know, animated or, you know, having some
Jeff Jarvis [01:14:02]:
Somewhere in the Googleverse, there's some lawyer and executive scratching their heads tonight trying to figure out what the hell to do. First thing they're going to do is appeal it.
Louis Maresca [01:14:13]:
Right.
Jeff Jarvis [01:14:14]:
They might win. Can we talk about the Anthropic regulatory capture?
Louis Maresca [01:14:19]:
Yes, let's go for it.
Jeff Jarvis [01:14:21]:
So, because both Paris and I are, I think, hot on that one. Paris, do you want to describe it first?
Paris Martineau [01:14:25]:
No, you go for it, Jeff.
Jeff Jarvis [01:14:26]:
So Politico had a story revealing that Anthropic had— I'm trying to find my post about this— that Anthropic came out and proposed a law about it. Line— was it line 162? Do you have that quote from it about the—
Paris Martineau [01:14:47]:
Yeah, let me find it here. It is— I'll just Ctrl+F50. The bill would let government block a model's release only if it poses a foreseeable imminent risk of more than $1 billion in damages or of killing or seriously injuring more than 50 people. The risk must stem from the creation or release of a bio— of a biological weapon, the model evading its developer controls, or a cyberattack. The language mirrors that found in California's AI laws, including SB 53, which was signed into law last year. It is basically Anthropic is— its representatives are pitching Capitol Hill lawmakers on a draft bill of AI safety legislation that just so happens to have a very unique definition of AI safety, which is that you can do up to $1 billion in harm and kill 49 people. But if you do $1 billion, if you do But if you do a full $1 billion or $1.1 billion or kill 50 people, then that's a no-no.
Jeff Jarvis [01:15:57]:
Ah, the hammer's coming down then.
Paris Martineau [01:15:59]:
How do you decide that it's 50 and not 49?
Jeff Jarvis [01:16:01]:
That's ridiculous. And this was based on some language in a California law, but this is, this is a perfect example of regulatory capture. We're going to write, we're going to write our own law to regulate us. We're going to set this bar really high, uh, and so we're going to reduce our liability and risk to zero. And we're also going to throw in things that are going to be difficult for our competitors to deal with. So they're not going to come in because we're going to increase the cost for them. This is regulatory capture. I made— I used Nano Banana to make a little poster here, movie poster of this on line 99.
Jeff Jarvis [01:16:35]:
If you can see that, Benito.
Jeff Jarvis [01:16:39]:
Yo, I should have warned you.
Paris Martineau [01:16:42]:
And here we are. You know, I think that it's beautiful.
Jeff Jarvis [01:16:46]:
No, that was not it. No, I'm sorry, it wasn't. Did I do the wrong link? Oh, I'm an idiot. I'm an idiot.
Paris Martineau [01:16:53]:
Listen, I think that you're trying your best, and that's okay.
Jeff Jarvis [01:16:56]:
Uh, but you know, I'm putting this into the WhatsApp.
Paris Martineau [01:16:59]:
This better be the most worth it.
Jeff Jarvis [01:17:01]:
It's not, it's not. It's so much of the stuff.
Paris Martineau [01:17:03]:
Jeff, you just posted a link to your Blue Sky account. You didn't post a link to anything. You just posted a link to Your Bluesky account.
Jeff Jarvis [01:17:11]:
Oh, shoot. Oh, well, I did a poster, License to Kill 007 Anthropic with Dario Amadei pointing the gun.
Paris Martineau [01:17:19]:
I can't even find it looking. Oh, no, I can.
Jeff Jarvis [01:17:21]:
You found it?
Paris Martineau [01:17:22]:
Yeah, it's fine.
Jeff Jarvis [01:17:24]:
It's fine. It's not as good as the, as our chat. They'd make fun of me when I'm trying to emulate our chat. Never mind. That was a failure.
Paris Martineau [01:17:33]:
I'm glad that we got to all go back and forth on it for a bit.
Louis Maresca [01:17:38]:
I would love to know how this kind of relates to Anthropic's call out of obviously slowing things down for security purposes, right? Like, this doesn't seem like a, a bar that causes them to slow down.
Jeff Jarvis [01:17:48]:
It's ridiculous.
Paris Martineau [01:17:50]:
I mean, I think this is also what happens when— if we were to give Anthropic the benefit of the doubt, which I don't think we should. No, no. Um, this is what happens anytime you get Capitol Hill lobbyists involved on codifying some sort of policy position like It's of course never going to be like the law should be that anyone that causes any harm, financial or physical, should have the full weight of the law brought down upon it. It's like, well, if you kill 51 people and do a billion dollars of damages, then maybe it's a problem.
Jeff Jarvis [01:18:27]:
Then bad on you. Um, yeah, I'm naive about this stuff. because I don't do Washington. When I was working on some of the protectionist legislation around the news industry, I was amazed how much lobbyists do fully write the legislation and it goes letter for letter into legislatures. I thought there'd be at least a little more. It's like treating your AI writing and you try to fake it a little bit and throw in some of your own stuff. stuff? Nuh-uh. And so we should not let these AI companies write this regulation, damn it.
Louis Maresca [01:19:08]:
Well, speaking of AI companies and also writing some regulation, I think there's some more about that because I think they're also trying to donate their way through some things. Well, yeah, we were talking about a little bit about regulation, but obviously some of these companies are, you know, they're attempting to kind of flip the coin a little bit. And, you know, although it seems I'm going to give them the benefit of the doubt. There are some dueling donations going on, but some interesting stuff here. Anthropic is setting aside 5% of its shares for public benefit, which is roughly around $75 to $100 billion. And then of course, Elon Musk on a different approach, he's promising every American, every child in America— Child. Yes, I have to be very specific. specific, a SpaceX share.
Louis Maresca [01:19:51]:
So it's about $150 today
Jeff Jarvis [01:19:53]:
after a million-for-one split. Yeah.
Paris Martineau [01:19:57]:
I do think it's very funny how a lot of news agencies reported these as equivalent announcements when Anthropic's was in a flashy press release, a lot of different, you know, information about it. And Elon Musk's was like basically a reply on Twitter to a bunch of people patting Anthropic on the back for that, being like, well, I'm gonna give a share to every child.
Jeff Jarvis [01:20:22]:
Take that.
Paris Martineau [01:20:23]:
Take that. And people are like, yeah, Elon Musk did one too, and that's really important for us all to know. And it's like, how many things has Elon Musk tweeted about that he hasn't done?
Jeff Jarvis [01:20:33]:
This Anthropic thing strikes me as a bit of effective altruism BS. So they say that they're going to contribute 5%, uh, with an additional 2% added at the board's discretion every 50% increase in the company's value How's that? Above a set baseline, whatever that might be. But then it's really unclear what it's going to do. It's a public benefit fund that is going to be part of an LLC, not a 501. They say, well, okay, we won't use it to lobby or do political things, but still, there's no clear view of how the money's going to be given to what. It'll have increased flexibility and scale enabling us to pursue our mission in new ways. For example, we might explore large donations to traditional nonprofits or direct distributions to specific populations. What the hell does that mean? One example is labor market impacts.
Jeff Jarvis [01:21:30]:
Anthropic has already committed $200 million to studying and addressing this issue. Okay, but who's doing that? How, how public is that? Is that really independent of you? And what does that mean? What does that mean? What does that do for anybody? Because of the way it's structured, the more Anthropic grows, the larger the share of the company will be dedicated expressly to the public good. Well, who's defining the public good here? If we go to the EA bozos these days, that means you don't care about things going on today. You care about the 10 to the 54th future computerized human beings, you cultist fools, you. And so, no, pay your taxes. Be specific about what you're going to do. to ameliorate the damage that you might bring. Make public commitments about how you're going to affect the economy and the environment.
Jeff Jarvis [01:22:17]:
This is meaningless. It's completely meaningless. But it got them what they wanted before their IPO— lots of publicity.
Louis Maresca [01:22:25]:
Right. I feel like it's almost a lift, trying to get a lift on their IPO to say, hey, if our price is higher, we're going to spend more money at being— donating more money.
Jeff Jarvis [01:22:35]:
Yeah. And then Reuters tried to compare it, to your point, Parris, with other things. So the OpenAI Foundation with a 26% stake in the not-for-profit— so that's the foundation that owns OpenAI. So that ain't— that's nothing. The foundation will initially focus on a $25 billion commitment in 2 areas, health and curing diseases. Well, that's the work that they're doing anyway, and trying to get good points for anyway. Then there's the Gates Foundation. Well, we know the complications of that, but that's his own personal money.
Jeff Jarvis [01:23:05]:
He's giving it all away. Then there's Warren Buffett's pledge, same thing, though he pulled it back from Gates. Then Mackenzie Scott, let us finally say there's one person who's doing it right. And she's done the Giving Pledge. And then the Ford Foundation. Yeah, because Henry Ford was a jerk and he gave away his money. Okay, fine. The Lilly Endowment and the Novo Nordisk Foundation.
Jeff Jarvis [01:23:25]:
So Anthropic gets thrown into some of these things that are actually proven philanthropic goods, as if it's done anything yet, and it hasn't. And, and what's the accountability here? Where does the money go? Grrr.
Paris Martineau [01:23:37]:
Speaking of Anthropic, uh, do we want to talk about the S-1 that just leaked?
Jeff Jarvis [01:23:45]:
Yeah, give us some details.
Paris Martineau [01:23:48]:
So this is, um, from the Wall Street Journal, from, uh, Anissa Gardizzi, a former colleague of mine. Um, I've also got the link, it's the bottom one. In our spreadsheet of myth and lore. Basically, it's the draft prospectus for their initial public offering. Anthropic said that it had recorded a loss of more than $80 billion in the first half of this year, a total that largely reflects an accounting adjustment related to its financing agreements, which basically just means that, like, suddenly it's counting the things that would be associated with that loss. loss. And at that, $80 billion is a lot more than what it had said that it lost, like, as far as a net loss basis, um, last year, which was $13 billion. Anthropic laid out a sprawling list of risk factors, including uncertainty over how it will prevent its AI models from causing harm and protect itself from those liabilities, from sophisticated cyberattacks to possible blackmail by uncontrolled models.
Paris Martineau [01:24:51]:
Blackmail. Yeah, blackmail. Who would have guessed? The takeaway is from the Wall Street Journal that computing costs are growing fast, revenues are increasing even faster. Although it says Anthropic generated around $16 billion in revenue in the last 6 months, uh, and that it during that period spent nearly $9 billion in computing capacity and infrastructure. Where, I guess, then is the rest of that $80 billion in losses coming from? I'd like to know.
Jeff Jarvis [01:25:19]:
I think, well, the question is, how are they booking the commitment, the huge number of commitments they've made to Amazon? And I don't know if Microsoft too, and Google and such for hosting and chips. I bet that's not even booked in that.
Paris Martineau [01:25:36]:
Yeah. I mean, Anthropic, it says Anthropic needs, it's on the hook to pay around $450 billion in cloud computing agreements over the next 7 to 10 years. And the bulk of those deals are non-cancelable. It has spent $111 billion on Google Cloud computing contracts, $110 billion with Amazon Web Services, $31 billion on Microsoft, quote, regardless of actual usage. It's also agreed to spend up to $86 billion renting chips from Nvidia, or Nvidia chips from smaller AI cloud providers.
Jeff Jarvis [01:26:06]:
You know, to answer your question from The Wall Street Journal, the vast majority of that loss, $2 billion resulted from accounting charges. Don't you love that? Tied to changes in the value of convertible notes and other instruments the company issued to raise funding.
Paris Martineau [01:26:19]:
What does that mean?
Jeff Jarvis [01:26:22]:
They had convertible notes and
Paris Martineau [01:26:26]:
they converted and they were bad.
Jeff Jarvis [01:26:28]:
I'm guessing so. They owed more. You know, they gave away more value. I don't know. That's the kind of a thing that is opaque to most mortal people.
Paris Martineau [01:26:42]:
Well, the next point, I think, really goes to the point we were talking about earlier with regards to regulatory capture, which is that Anthropic sees damages caused by AI as a great financial risk. Roughly a third of the draft prospectus was dedicated to risk factors, which is not usual in these sort of things. Risk factors is normally like maybe like half a page or like a page if you're crazy. Roughly a third was risk factors, including the possibility that misaligned models cause real-world harms. They said that models can show self-preserving qualities to avoid being shut down and can generate responses that resemble blackmail in an effort to convince humans to allow them to continue operating. In some cases, Anthropic said it bears liability for certain harmful model behaviours, meaning it'd be on the hook to pay damages, such as for stolen data or security instances caused by its models. But the company warned that, quote, harm that could result from a catastrophic AI-related incident might be so large that it would be unable to obtain insurance to cover potential damages. Which is, I guess, why it's pushing for a bill that would give it— sell it, get it off the hook.
Paris Martineau [01:27:47]:
Unless—
Jeff Jarvis [01:27:47]:
license to kill 49 a year. Yeah, yeah, almost as many as Florida executes in a year. You know, it's, uh, Sorry.
Louis Maresca [01:27:58]:
I mean, I've said this before. I think I said this even on Twitter. I think, I think definitely think Anthropic is playing 4D chess. Like they are going from all different directions to kind of manufacture ways to solve their problems. And I think this is one of them.
Jeff Jarvis [01:28:12]:
Yeah. I don't think they're as clever as they think, but they're cleverer than the policymakers and the media. It gets here. Yeah, there's— I'm not a doomer. I think that's all BS. I think that's part of the cult. But there is real current harm and a failure of their accountability to worry about that current harm. The other financial story that got me, I've been thinking about this, is Larry Ellison's borrowing spree to bankroll Skydance.
Jeff Jarvis [01:28:52]:
I went on the socials. I think that Larry Ellison could single-handedly cause a crash.
Paris Martineau [01:28:59]:
And for what? Wow.
Jeff Jarvis [01:29:00]:
Exactly. So he's borrowing like crazy to build out for hyperscalers. And if one of them goes under, Then he's screwed. His son has bought this picket of poke of entertainment. It's right now worth less than half. Last I looked earlier this week, there was a story that, that, that the combination of Viacom and Paramount or Paramount and Warner Brothers Disney were less worth half what they were on the market before he bought them. So that's already sinking down. That's gonna be bankrupt.
Jeff Jarvis [01:29:37]:
I predict, uh, and the equity holders lose all of their value and the debt holders control it. And then you look at how Ellison has, um, what's the word you use when you collateralized the stock he has in all these other companies? That's going to cause a rush on them. So I think Larry Ellison is putting the entire economy at risk Around data centers and Skydance.
Louis Maresca [01:30:06]:
Yeah, there's actually a lot of diverse companies in there too. There's Paramount, there's Tesla, SpaceX. She's putting a lot of risk in there.
Jeff Jarvis [01:30:14]:
Yep. And lending from 5 banks: Citigroup, HSBC, JP Morgan, UBS, and PNP Paribas. And we know they're all too big to fail. Plus, when— if Skydance goes bankrupt, and I'm just taking bets on the over-under on that right Um, who takes it over? The, uh, debt holders. And who are the debt holders? Oh, I don't know, Saudi Arabia owns Hollywood.
Paris Martineau [01:30:42]:
Whoops.
Benito Gonzalez [01:30:43]:
Well, the Skydance play, I think, is more closely related to Elon buying Twitter. It's a culture play, not a money play.
Jeff Jarvis [01:30:50]:
Oh yeah, yeah.
Benito Gonzalez [01:30:52]:
Like they want to control the conversation.
Jeff Jarvis [01:30:56]:
Yeah, but yes, I, I think that's their motive. But they also bought parts of a dying industry. And they're going to put out 30 junky movies, banal movies a year. And I don't think it's going to have the effect that it has. And look what they've done to CBS. And it's a cluster F already. CNN is going to be the same. The Journal says that Ellison is one of the single biggest borrowers in the world.
Paris Martineau [01:31:27]:
It is wild. It's crazy to have that as a sentence when you have OpenAI and Anthropic.
Jeff Jarvis [01:31:33]:
Yeah. You know, and I remember back to Rupert Murdoch famously risked his whole empire when he overextended himself with borrowing, but all he risked was his empire. He could have lost Fox and News Corp. In this case, he's risking everything his tentacles are touching. Those banks are affected. The other companies' value in the market is going to be affected. The entertainment industry is going to be affected. He's just a— I watched The Dropout, which is about Elizabeth and the Larry Ellison scenes.
Jeff Jarvis [01:32:08]:
Do you remember those, Paris?
Paris Martineau [01:32:10]:
I don't.
Jeff Jarvis [01:32:11]:
She's on his boat and he's screaming, just get the money, get the money! And she's screaming, get the money, get the effing money!
Paris Martineau [01:32:17]:
Oh, I do recall those.
Jeff Jarvis [01:32:18]:
Right, right. And it's just, it made Ellison just the worst possible jerk. And he hasn't been presented that way in culture in a while. And watch out, folks. This is what AI could be doing. Because there's a Jenga mountain here. And one set of the Jenga pieces of wood are the hyperscalers and AI. Another set is the entertainment industry.
Jeff Jarvis [01:32:44]:
And you pull enough of either out, and the whole friggin' thing collapses. How's that for cheery?
Louis Maresca [01:32:51]:
It's cheery. Maybe we should go to this cheery story. There's some other ones, right? Some small— there's some, a couple, uh, entrepreneurship stories in here. I think there's one, uh, I'm not sure if Leo dropped it in here. Idealab.ai.
Jeff Jarvis [01:33:04]:
I did put that in there. That's a friend of mine, Bill Gross.
Louis Maresca [01:33:06]:
Yeah, I'd love to hear about that one because that's a—
Jeff Jarvis [01:33:07]:
So this is really interesting. So Bill Gross is a phenomenal entrepreneur. He, uh, started Idealab where he has created more than 175 companies. Phenomenal. Now, some were badly timed, Pets.com. Others were very well-timed, Cooking.com, he sold to Target. And you can name a lot of exits. Bill's a wonderful, wonderful guy, very generous.
Jeff Jarvis [01:33:35]:
And I was talking to him a few weeks ago. This is the announcement just kind of came out today in the Wall Street Journal. So now I can talk about it. And so Bill's model has been to see a problem, then imagine a company that could solve that problem, then hire a few engineers to build what that is, to invest some angel funding in that. And then if it shows traction, then he can invest some more and others can invest and it becomes a company and then it goes off on its merry way. Or if it doesn't, it doesn't. And that's that. But not too much is lost.
Jeff Jarvis [01:34:06]:
Right. And so Bill put in— he realized one day that journalism and academic papers are problem statements. It's a really interesting insight. And so he took a ton of news and a ton of academic papers, and he put it into a model. I think it was Claude, and said, find me the problems here that these things are writing about. And then he said to it, imagine companies that can solve those problems. So now he has a site at idealab.ai where he has 1.2 million business ideas for free, giving them away, no strings whatsoever attached. He wants to see people start things.
Jeff Jarvis [01:34:44]:
That's the way Bill is. And some of them he may invest in, but no difference. He doesn't care. He doesn't care if you ever tell him about it. He'll just go with God and start stuff, solve problems. And so you can go in there. And the other thing he did, if you keep scrolling a little bit lore down there, the periodic table of business functions. He also split apart what businesses need.
Jeff Jarvis [01:35:06]:
So things like building websites and payroll and lead qualification and so on and so forth. So he looked at these functions of what companies need and said, here's where they're being met already in the market, but here's where they're not being. That's also business opportunities. So Bill has put out there 1.2 million business ideas, and anybody, any of you can go for free, find a business and say, hell, I want to start. That's a great idea. I want to start that. Or that's not so good, but inspired me to do something else. And if you want to go to Bill and say, I've got an idea to how to do this, he might invest, he might not.
Jeff Jarvis [01:35:43]:
But that's the Idealab model. So he thought that AI would eliminate his method because it made it so easy to find the problems and solutions. but very characteristic to Bill Gross, he instead decided to make that available to all. So it's kind of cool.
Louis Maresca [01:35:58]:
Yeah, really cool. It's actually really— you actually go in there, tell what you're good at, and then you choose— you could choose an industry, it looks like, some kind of sector. And it actually— it's like a fit model, says you're fit for this particular idea, 57%, uh, Apprenticeship OS for AI Era Work, or Human-AI Operating Architecture Studio, all Different leadership flight simulator. That's interesting. So there's some really awesome ideas in here and they kind of connect you up with some of your skills. So if you want to go try to build it.
Benito Gonzalez [01:36:26]:
Okay. So I'm a, I'm a triage nurse trying to be a creator for up to $10K under 3 months. Let's see. Let's see.
Jeff Jarvis [01:36:33]:
This is you, buddy.
Benito Gonzalez [01:36:34]:
Well, no, that's not actual me. I just chose stuff.
Jeff Jarvis [01:36:37]:
No, I know that. I know that if you're a triage nurse, you're wasting your talent.
Benito Gonzalez [01:36:41]:
Nobody, nothing. Okay. Well, we got to find something easier.
Louis Maresca [01:36:46]:
Your prompt might be too long.
Paris Martineau [01:36:49]:
Yeah. The prompt might be too long. It's basically 2 sentences.
Jeff Jarvis [01:36:52]:
Uh, make yourself— make it— you're— I'm a podcast producer wanting to, um, uh, solve the production problems of, uh, podcasters. See what it says.
Paris Martineau [01:37:07]:
I make a lot of insightful comments about the roles of CEOs' lives versus I'm a podcast producer looking to, looking
Benito Gonzalez [01:37:13]:
to solve podcast problems.
Paris Martineau [01:37:14]:
Looking to solve podcast problems.
Jeff Jarvis [01:37:16]:
Podcast production problems. There you go.
Paris Martineau [01:37:17]:
That's great. That's really good.
Benito Gonzalez [01:37:19]:
All right. I want it.
Jeff Jarvis [01:37:20]:
It's gotta be cheap.
Benito Gonzalez [01:37:21]:
Oh, no. Let's see. Let's see what's, let's see what we can get for $10 million. Let's see what we can get for $10 million. Okay. In a year. Let's give it a year.
Paris Martineau [01:37:27]:
Yeah. That casual amount of money that every podcast producer has.
Jeff Jarvis [01:37:32]:
Let's see if it comes up with anything here.
Benito Gonzalez [01:37:34]:
Oh, hey. Oh, hey.
Jeff Jarvis [01:37:36]:
Adaptive production OS for independent podcasts. Capital of only $150,000.
Paris Martineau [01:37:41]:
I think the issue with the, uh, trauma nurse was that she only had $10K because all of these seem to take $100K or more.
Jeff Jarvis [01:37:51]:
Radio to podcast revenue engine, transcript-first podcast production workspace, podcast revenue intelligence.
Benito Gonzalez [01:37:58]:
Oh no, no, this, this isn't real, guys. This, that's not— this isn't— this one isn't real. They can't do this, right?
Jeff Jarvis [01:38:05]:
What do you mean?
Benito Gonzalez [01:38:05]:
This is my whole job. This is my whole job.
Paris Martineau [01:38:10]:
Oh, no!
Jeff Jarvis [01:38:11]:
Or actually,
Louis Maresca [01:38:12]:
I just hope the bar is so high. Look at the bar. That's pretty high. Nobody's going to do that.
Jeff Jarvis [01:38:16]:
Or Benito's going to grab this one. And the problem is, we've got to tell Leo he lost Benito.
Benito Gonzalez [01:38:19]:
I don't have $75,000.
Jeff Jarvis [01:38:21]:
Leo's going to go off and become rich.
Louis Maresca [01:38:25]:
You've got to wonder how they compute the capital to start.
Paris Martineau [01:38:30]:
What model is pumping out these numbers?
Jeff Jarvis [01:38:33]:
Yeah. Yeah. I mean, what I saw when it had 6 million businesses, some of them were ridiculous. They made no sense at all. So he's put on more factors on the prompt here to make it better. But yeah, some of them are gonna be ridiculous. But it's very cool. And so, you know, I've gone to somebody at Stony Brook and said— Bill's already talked to ASU, Arizona State, and talked to Stony Brook and said, wouldn't it be great for students to be able to imagine themselves in these positions and maybe start something.
Jeff Jarvis [01:39:03]:
I mean, I ran a program in entrepreneurial journalism and I had students start businesses. And I just— I should plug her. I just saw one of my graduates last week, the author of The Butterfly Season, Leah Korsgaard. You might like this, Parris. Very cool. She started Zetland, which is a digital newspaper in Denmark, and they've expanded to They're expanding to Norway, they're in Finland, they're going into Germany, I think, and Austria and so on. And so she wrote a book about trying to find every butterfly species in Denmark in a year, in a season.
Paris Martineau [01:39:38]:
Oh, wow.
Jeff Jarvis [01:39:39]:
It's a really cool book. And she's really cool. Anyway, she was one of my entrepreneurial students. And she started Zetland. And it's very successful. It got bought by a big media company. It's continuing. Yeah, anybody can be an entrepreneur.
Jeff Jarvis [01:39:50]:
Not easy.
Louis Maresca [01:39:52]:
Right. It's not, especially now in this, this climate, it's very hard.
Jeff Jarvis [01:39:55]:
Well, then again, you can do a lot of stuff with AI that you couldn't have done before, right? If it requires coding, if it requires analysis, um, AI can do a lot.
Louis Maresca [01:40:08]:
Yeah. I mean, if you think about it, I just, I just posted this in chat. I mean, if anybody who can just know how to prompt something can build something, build an app in 30 minutes. I, I basically built louis-m.com in 30 minutes and I didn't even look at it. I just let it build it while we were talking here.
Paris Martineau [01:40:25]:
Is it live?
Louis Maresca [01:40:26]:
It's live. Go to it. Yeah,
Paris Martineau [01:40:27]:
I'm sorry,
Louis Maresca [01:40:28]:
I'll probably take it down. It probably will be taken down because I didn't put a lot of capacity behind it. But,
Jeff Jarvis [01:40:33]:
but, uh, actually, Lou sees wealth and he's gonna— I
Paris Martineau [01:40:36]:
like that it immediately has something on the front page that allows you to break up your image into this And I didn't tell it to do
Louis Maresca [01:40:45]:
that, but yeah, I guess—
Paris Martineau [01:40:46]:
Kid-hearted curiosity, engineering degrees as enabler.
Louis Maresca [01:40:51]:
I gave it very little information too.
Paris Martineau [01:40:53]:
It's also a classic Claude-ish thing.
Jeff Jarvis [01:40:55]:
Ideas are better out loud.
Paris Martineau [01:40:56]:
It's got solid text, some bold text, normal text, and then italicized text in a different color.
Louis Maresca [01:41:03]:
Right. And I just thought that was really interesting that I gave it very little information. I went out and did this in 30 minutes and it's it's viable.
Paris Martineau [01:41:10]:
It is.
Louis Maresca [01:41:11]:
So, and I have no idea what that does, so we'll have to try that out later. But there's some stuff in here that has some interesting things that it decided to add to it. I didn't stop anything, so whatever it does is— What was your prompt? I'll have to share it. It was, it was my default prompt in the beginning of like, you know, make sure you build it this way. But then, uh, it was just build me a profile site that, you know, that also talks about hosting podcasts and, you know, and GoWebGround and the— and Twit and so on. So it was like, I'll have to share the whole prompt, but, um, but as we were doing commercial break, I wrote it.
Benito Gonzalez [01:41:46]:
So you didn't give it any info about you? Like you just said scour the internet about you?
Louis Maresca [01:41:50]:
No, wait, I told it to go look on the web about me. So that's why mostly the Twit stuff's there, my guess is, because there's not a lot about me anywhere else. So, so yeah, I thought that was really interesting. Um, I did scan it really quick before I posted it, just so people know that there wasn't any like, you know, naughty stuff in there or anything like that. But, uh, but yeah, it's pretty, pretty cool. That's— I did— that's the point of this whole entrepreneurial thing is like really, it really enables folks to go and do what they, you know, come up with an idea and they don't have to have the skills to do it. And this even posted it. This is a free tier in Azure that it's running on.
Louis Maresca [01:42:24]:
If you— as long as you don't go know how to go create an account. It used all the commands and everything and deployed it for me too. So I didn't have to do anything. So I think we, you know, the— I hate to say it, but the Squarespaces of the world and stuff like that might be, uh, doomed as well.
Jeff Jarvis [01:42:39]:
Oh yeah. Yeah. Well, I mean, if, if, uh, Adobe and— sorry, Microsoft, uh, can be replicated. I saw, I saw a lawyer on, I think, I think it was on TikTok talk, lecturing people and saying, just because you say that you didn't— you reverse engineered the thing, you're not violating perhaps copyright, but you probably are violating the terms of service in each case, and you're going to lose in court, folks. So, but I think this replicating phase, I could remake Adobe Suite, is fun and cute. But the really interesting thing is, what can you make that's entirely new? And make it fast and make it good with these tools.
Louis Maresca [01:43:23]:
Right. There are a lot of—
Paris Martineau [01:43:24]:
I just put in my Claude, design and make a fun website that I can launch and share with the I Am listeners tonight while on air. Use your best judgment. Ask me no questions and I'll let you know what it comes up with.
Louis Maresca [01:43:36]:
Yeah, I would love to see that one. Yeah, mine was, I have this like standard, you know, coding prompt that I stick in there. So It's pretty long, but I think it tends to have them go follow. And I think mine, I used Copilot, so I told it to go, you're limited to like X number of subagents or whatever to go do it. I think it would probably take it longer if it was one single model and agent, but I'm just curious to see how many tokens it used. I'm gonna have to—
Jeff Jarvis [01:44:02]:
Yeah, I was gonna say, what did it cost you exactly?
Louis Maresca [01:44:04]:
I don't know, I have to find out.
Jeff Jarvis [01:44:08]:
So this is a future for young people today, but meanwhile, LaGuardia High School students staged a walkout in protest of AI. Keep this stuff away from us. I don't know if that's a very New York thing or if that would be true across the country.
Paris Martineau [01:44:23]:
I just, I need to stop looking at the Claude thing cuz I got distracted because it, one of its thinking things was planning out bingo win logic. What does that mean? I don't know.
Louis Maresca [01:44:35]:
I think, I think you're right though, Jeff. I think there's, there's a lot of people that are are either afraid of it or they're going to— that it's going to take their job away or it's going to take their, you know, the things that they're studying for. They're paying hundreds of thousands of dollars to go to college for, and they're worried it's for naught.
Paris Martineau [01:44:49]:
Yeah. And I think they have every right to be worried about that.
Jeff Jarvis [01:44:52]:
Gee, Mom, I always wanted to be a PhD in math. No kid. Can't you do something better? Go be a plumber. Build a data center.
Louis Maresca [01:45:02]:
I mean, it's tough. I have a son who's 16. He's a junior in high school, and he's really struggling to figure out where he's head next because he wants to head somewhere where there is some room.
Jeff Jarvis [01:45:15]:
Yeah, I think we talked a few weeks ago that computer science majors are down 53,000 in the US. Now math majors, who knows what happens there, right? And that's a surprising space. I didn't think that'd be— mathematics would be hit early. But then it doesn't really— If you're intellectually curious, Right.
Paris Martineau [01:45:36]:
What you study in college doesn't really matter. It's more about just learning to think, which I'm also worried about children not being able to do that because suddenly they can just ChatGPT everything. Mm-hmm.
Louis Maresca [01:45:48]:
That's a huge—
Paris Martineau [01:45:49]:
How old are your kids?
Louis Maresca [01:45:51]:
Uh, they range. I have a 5-year-old all the way up to 16.
Paris Martineau [01:45:54]:
So every 3 years. How have they had to contend with the fact— I mean, how are they contending with the fact that And now I guess if you haven't done your homework and it's 10 PM and you've got to turn something in at 9, you could just ask ChatGPT and you have an essay.
Louis Maresca [01:46:12]:
Yeah, they— I have had to contend with that. I think it's helpful I can block those websites though from my local network. But, uh, but I think—
Jeff Jarvis [01:46:21]:
But Dad, you use AI.
Louis Maresca [01:46:22]:
I know, I know. The, the— it's funny because my son had a project he had, and I'm gonna, I'm gonna actually like, uh, get him in trouble here. But he, uh, he had a project that he had to do yesterday, which was— I think it was a, a, it was a mousetrap vehicle where he basically had to build a mousetrap vehicle that propelled it forward. And I think he— it was Opus or one of the ones out there that he went and used. And he got an idea and he built it. It was like CDs as the wheels and a straw and a couple things. And he built it and it was like an hour and he And he, and he, uh, took it to school this morning and it was just built all by AI basically. So I wish it was actually designed by him, but, but, uh, it looked pretty cool.
Louis Maresca [01:47:03]:
But
Jeff Jarvis [01:47:05]:
now you learned things, right? Did he learn from the experiment, from the, from the exercise?
Louis Maresca [01:47:11]:
He, you know, he tweaked, you know, I think the interesting part is he tweaked it and he had to figure out the physics behind it.
Jeff Jarvis [01:47:16]:
Yeah.
Louis Maresca [01:47:16]:
So I think that's the part that's, I guess that's one of the important parts, but yeah. Yeah, was unfortunately not his idea.
Jeff Jarvis [01:47:23]:
Well, I won the 8th grade science fair because I made electronic bongos.
Paris Martineau [01:47:31]:
Say more.
Jeff Jarvis [01:47:33]:
Well, it was in— but it was just something in Popular Science magazine, and it was obvious, and I got it out of there, and I just built them, and it won. My father sold resistors for a living, and if you're trying to understand what your father does, and he sells these little tiny plastic pills with wires coming out of them. You have no idea. So the one way to try to bond with your father is to figure out what resistors are. So we use resistors, and I got free resistors and all that. So they were electronic bongos.
Paris Martineau [01:47:59]:
Did you have fun building the electronic bongos?
Jeff Jarvis [01:48:02]:
Well, let's see, this is the— this is a— this is representative. I was trying to make the case for the amplifier, an old— an old radio. My father came down watching me do it, and he said, and I quote, you're so clumsy you couldn't stick your finger up your ass with 2 hands. It was rather lopsided.
Paris Martineau [01:48:20]:
My version of that is one time my dad and I were looking for a bonding— I might have been at middle school or something. We were looking for a bonding activity. We decided to make like a little model plane or something, and midway through we're both like, this sucks. We're both so bored. We don't have the dexterity to be maneuvering all these tiny parts. Let's just go get ice cream or something. Well, see,
Louis Maresca [01:48:41]:
you did ice cream together. That's great too. Yeah, there you go. It was great.
Paris Martineau [01:48:44]:
There's other things that you can do other than maneuver tiny parts into a tiny something. I'm sure that works quite well for someone.
Louis Maresca [01:48:54]:
Right.
Benito Gonzalez [01:48:54]:
The thing with model building is one of you has to be good at it already because that is not an easy thing to do.
Paris Martineau [01:49:01]:
I don't know how people are building model ships in a bottle. That's crazy to me.
Jeff Jarvis [01:49:06]:
Nor do I want an AI for
Louis Maresca [01:49:07]:
that.
Jeff Jarvis [01:49:07]:
No, I don't. Yeah, no, no. Paris, there went your career as a civil engineer. Could have been different.
Paris Martineau [01:49:15]:
Yeah, could have been all different.
Jeff Jarvis [01:49:17]:
Yeah.
Louis Maresca [01:49:18]:
We have fun with these model rockets, you know, being able to fire rockets up in the sky. We, in fact, you're right about that, Benito. I actually had to learn about how these things work before I had to bring my kids into it to build these things.
Jeff Jarvis [01:49:32]:
Yeah.
Louis Maresca [01:49:33]:
Even the kits.
Jeff Jarvis [01:49:35]:
Yeah, even the kits. That's true.
Louis Maresca [01:49:36]:
Yeah. Yeah.
Jeff Jarvis [01:49:37]:
It's okay. AI is not going to destroy everything, just some things.
Louis Maresca [01:49:41]:
Right. Right.
Jeff Jarvis [01:49:43]:
That's all.
Louis Maresca [01:49:44]:
There's some medical stuff in here that I think is pretty interesting because I'm not in this space. I always find it very interesting around some of the biotech industry startups that are coming into play here. This one being Radical Numerix. They have a medical milestone by using AI to design, using AI-designed viruses to treat human patients. Viruses are called bacteriophages, infect and kill bacteria rather than human cells. Very interesting stuff.
Jeff Jarvis [01:50:12]:
What could possibly go wrong?
Louis Maresca [01:50:14]:
Right, right. They have their own trained model called OmniOne that actually helps design viruses. Yeah, this doesn't sound like it could be safe. We don't have any patents yet, they say, but we're early on in our AI to help design and treat specialties for those specific person infected.
Jeff Jarvis [01:50:34]:
I hope so. I think this is the kind of stuff that AI is good for and should do.
Louis Maresca [01:50:38]:
Very, very, yeah.
Jeff Jarvis [01:50:39]:
A different way to attack a virus. Then the next one is a legit story. It's in The Guardian, but it really struck me of like Elizabeth, something Elizabeth Holmes would do. A new AI tool can predict cancer survival by a single patient selfie.
Paris Martineau [01:50:58]:
Can it though?
Jeff Jarvis [01:50:59]:
No, no.
Paris Martineau [01:51:02]:
Yeah, it can't.
Jeff Jarvis [01:51:04]:
It can't. Plus, I look old, so I don't want it to.
Paris Martineau [01:51:11]:
Yeah, what is the—
Jeff Jarvis [01:51:12]:
Custom built by experts at Mass General Brigham.
Louis Maresca [01:51:15]:
Right.
Jeff Jarvis [01:51:16]:
Right, that's, that's a legit place.
Louis Maresca [01:51:18]:
Yeah, I go there myself.
Jeff Jarvis [01:51:20]:
Um, one of the most comprehensive, uh, demonstrations to date of how AI analysis of a selfie can help assess assess a patient's health and potential guide their medical care. It was trained on data from more than 40 million facial photographs across a range of ethnicities, as it well should, and ages. They published it in Lancet. It can estimate a person's biological age and predict the chances of survival for cancer patients from an ordinary photograph of their face. So if we want to get paranoid here, the insurance Somebody uses it on you and says, oh yeah, you're not worth it.
Louis Maresca [01:51:56]:
Not fit.
Jeff Jarvis [01:51:57]:
You're doomed. Give it up. Forget it.
Paris Martineau [01:52:00]:
Yeah. This is concerning.
Louis Maresca [01:52:02]:
That's a little concerning.
Jeff Jarvis [01:52:03]:
It is concerning.
Louis Maresca [01:52:04]:
Yeah. Because then they're saying we won't pay out because we didn't think you were fit and you did it anyways.
Jeff Jarvis [01:52:11]:
Well, that's the fear. So it focuses— the various parts of it focus on distinct parts of the face. F-A-H-R, The measurement of face age focuses on the creases that run from the sides of the nose down to the corners of the mouth. I don't have any.
Paris Martineau [01:52:27]:
So what happens if you've gotten cosmetic surgery?
Jeff Jarvis [01:52:32]:
Well, then you cheat death, Paris.
Paris Martineau [01:52:34]:
Oh, then you'll never die. That's great. That's what I always thought would happen.
Jeff Jarvis [01:52:38]:
The forehead just above the eye sockets. No wrinkles here. I can't get rid of them.
Paris Martineau [01:52:46]:
This is why we'll never die.
Benito Gonzalez [01:52:50]:
All right, there we go.
Paris Martineau [01:52:55]:
You know, you see someone do something silly and you realize it has to be the show title or the show card. It does.
Jeff Jarvis [01:53:01]:
Well, last week we did Elizabeth Holmes' eyes,
Paris Martineau [01:53:03]:
and I looked crazy, and you
Jeff Jarvis [01:53:05]:
two looked crazy, and Leo barely opened his eyes. He wasn't into it.
Paris Martineau [01:53:09]:
That's why he didn't like the movie. Yes, committed to the bit. I do Do you think, Jeff, that tomorrow when we record our thing, we should all wear black turtlenecks? Can we? Do you have that technology? You're not wearing a turtleneck right now, Jeff.
Jeff Jarvis [01:53:22]:
I have a black turtleneck.
Paris Martineau [01:53:23]:
Okay, great.
Jeff Jarvis [01:53:24]:
I wear black.
Paris Martineau [01:53:24]:
But you know, tell Micah. And also I think, 'cause we forgot to say it before, it's gonna be streaming on all Twit platforms. You can watch it live in the club.
Jeff Jarvis [01:53:31]:
Even though it's club, we're gonna do it live.
Paris Martineau [01:53:33]:
youtube.com/twit, on Twitch, on Kick, wherever you get your streamable Twit products.
Benito Gonzalez [01:53:40]:
It'll also be released to public. It'll also be released to the public once it's done. done. So it's not going to be— we're not holding on to it for 30 days usually like we do for the Club Twitch shows.
Paris Martineau [01:53:47]:
Yeah, no, it'll be the next day. So whenever you're done watching, because I mean, the thing is, uh, it's actually had $15 million— made $15 million in revenue already despite the fact that it's only been released to like 800 theaters worldwide. But it's getting a broader release, I think, starting Thursday. So once you've had a chance to see
Jeff Jarvis [01:54:08]:
So here's the question, folks. Which one of us is Ebert and which one of us is Siskel?
Paris Martineau [01:54:14]:
It's a great question. There's only one way to find out. It'll be watching the podcast tomorrow.
Jeff Jarvis [01:54:18]:
Mm-hmm. Or one of the two, one of the two old guys in the, in the theater.
Louis Maresca [01:54:22]:
Mm-hmm. Um, the Muppets.
Paris Martineau [01:54:25]:
Muppets. Yeah.
Jeff Jarvis [01:54:25]:
Yeah. The Muppets. What are the names? I forget. Yeah.
Louis Maresca [01:54:27]:
Oh my gosh. Yeah. I don't remember.
Benito Gonzalez [01:54:30]:
Stadler and Waldorf.
Louis Maresca [01:54:32]:
Yes.
Jeff Jarvis [01:54:33]:
Stadler and Waldorf. Yeah. Which one of us is Stadler and which one of us is Waldorf? You decide.
Paris Martineau [01:54:38]:
Okay, back to the study. They say image quality, lighting, and facial expressions can still affect accuracy. Okay, cool. So if I'm frowning in the pic, in the selfie I take, you're like, sorry, you're gonna die earlier, and therefore you don't deserve to have your cancer covered by insurance?
Jeff Jarvis [01:54:57]:
I made up the insurance part.
Paris Martineau [01:54:59]:
I know, but like, that is of course what the— if this becomes a viable use case for it, of course that's how it's going to be used.
Jeff Jarvis [01:55:06]:
Yeah.
Paris Martineau [01:55:06]:
And I mean, the difference in these sort of smile and laugh lines, if I'm sitting here stone-faced versus if I'm frowning, is quite great.
Jeff Jarvis [01:55:18]:
Well, also, people who smile a lot have more wrinkles, I think,
Louis Maresca [01:55:23]:
right? Yeah, true.
Jeff Jarvis [01:55:24]:
People who are grumpy like me might have fewer. Paris, you're just so young and fresh-faced, we don't know how it's going to turn out for you yet.
Paris Martineau [01:55:32]:
I mean, I do have Quite large smile lines if you're looking at me closely because I smile all the time. That's great.
Jeff Jarvis [01:55:38]:
Yes,
Louis Maresca [01:55:39]:
it's a good thing. I think it's interesting here that they, you know, any type of time a model needs to do judgment— obviously they trained it on 40 million images— judgment is still, you know, there's still room in there for interpretation. I think with when you define it based off of data like test results or something like that, you have a better success rate than just based off judgment. So this is very interesting that this is all judgment-based.
Jeff Jarvis [01:56:04]:
But there's also— there's other factors that can lead to it, and many of which can be racial, demographic, how you lived, what you ate your whole life, that may have an impact on your face. Or you worked outside all the time, you know, those kinds of things. And that may not have an impact on your health otherwise, but these correlations are dangerous.
Paris Martineau [01:56:26]:
Yeah, and the study on authors describe it as, these models provide a non-invasive, inexpensive alternative that can flag patients unlikely to tolerate aggressive therapy, improve resource allocation, and quality of life. Basically saying that, like, yeah, you should use this to decide who probably shouldn't get the most aggressive forms of cancer treatment. Yeah. Because we think that they're probably not going to be able to make it work. That's crazy.
Jeff Jarvis [01:56:56]:
Yes.
Louis Maresca [01:57:01]:
All right. Well,
Benito Gonzalez [01:57:03]:
actually, I got one, one thing about that. Actually, this story is that it also assumes that there is a middle, a normal, uh, state for people. And does that exist?
Jeff Jarvis [01:57:12]:
Yes.
Paris Martineau [01:57:12]:
Yeah. Which is—
Jeff Jarvis [01:57:15]:
no,
Paris Martineau [01:57:16]:
a wild assumption to make, assuming that there's a default amount of not about to die from cancer, that that person should receive the default amount of treatment, but anyone on either side of that should either have their treatment reduced or improved, I guess.
Jeff Jarvis [01:57:34]:
It's also like society thinking there's a default definition of beauty. Yeah.
Louis Maresca [01:57:40]:
You know, right. It's time for picks of the week, which I love. I'd love to hear you, uh, your folks, what you have Paris?
Jeff Jarvis [01:57:51]:
Paris always goes first.
Louis Maresca [01:57:53]:
Oh, Paris goes first. I'm sorry.
Jeff Jarvis [01:57:54]:
Paris, it's probably—
Benito Gonzalez [01:57:54]:
You muted Paris.
Jeff Jarvis [01:57:55]:
Oh, but Paris then mutes herself and we don't hear it. Either her camera's out or her microphone is off.
Paris Martineau [01:58:00]:
I muted myself in case I needed to cough and I did not. And instead I was just talking as if you could hear me when you could not. Yes, I do. So I've got a couple quick things. One is that Craig, Craig, Craig Newmark is doing an AMA. New Mark! He's doing a Reddit Ask Me Everything tomorrow, October 22nd at 11 AM. I've included the, uh, link to his AMA announcement, and I don't know, he's, uh, ask him anything if you want. Just always shout out to friend of the show.
Paris Martineau [01:58:38]:
Um, my other thing is that I don't normally Do you guys ever watch the Subway Takes videos? Have you heard of this?
Jeff Jarvis [01:58:45]:
Once in a while. Have you seen them?
Paris Martineau [01:58:46]:
It's this guy who asks basically a celebrity what their take is when they're on the subway. It's kind of evolved from being lightly SpawnCon to kind of all SpawnCon stuff now. But it was very interesting. A video came across my desk this week of Conan O'Brien at Subway Takes, and the person that popped off was not Conan O'Brien, but just a random woman on the subway named Debbie who happened to walk in the shot and went completely toe-to-toe with Conan on improvising. And I don't know if we can play the clip, but, you know, are we allowed to play the clip? I, uh, can they copyright? We'll do a
Benito Gonzalez [01:59:24]:
little
Paris Martineau [01:59:27]:
bit now or in post.
Benito Gonzalez [01:59:31]:
All right, there.
Paris Martineau [01:59:32]:
Here, we'll do it in a sec.
Benito Gonzalez [01:59:33]:
Oh, I can't play the sound. Oh yeah, no, no, no, the sound is what's gonna kill us.
Paris Martineau [01:59:37]:
Oh, the sound, the sound is what's gonna kill us. Okay, never mind. So basically—
Jeff Jarvis [01:59:40]:
But you get the dynamics of it.
Paris Martineau [01:59:42]:
This woman just randomly walks into the shot and Conan is like, does a little bit with her and she out-riffs him in just a beautiful, like, 25-second bit that I'd really recommend you watch. It really delighted me. She matches and then bests him at like an improvised riff, and she's just a random lady on the subway. And everybody was like, wow, what a phenomenal improviser. What's her deal? And she's just a woman who owns a bar.
Jeff Jarvis [02:00:12]:
This is New York.
Paris Martineau [02:00:13]:
It is New York. And so, Vulture did an interview with her, How Subway Takes First Spinoff Star Outwitted Conan, that kind of gets into it. I mean, the interaction begins when Debbie, which is her name, gets permission from Rama, Kareem Rama, who's the guy who does subway takes, to even just walk through the shot. And it spirals from there. They do a wonderful interview with her where you kind of learn more about her, her sense of humor, where she comes from and what she does. She just owns and operates a bar called Confessions Bar. And yeah, I don't know. I found it a very delightful little thing that has happened this week on the internet, and I think you will too.
Benito Gonzalez [02:00:59]:
Well, she owns a bar. That's kind of the same job as Conan, right?
Paris Martineau [02:01:03]:
Basically. Yeah. I mean, she does say that she, um, kind of operates like a karaoke night and some sort of like little events at the bar. And she says, even at the comedy shows we run, I'll go on stage and host before the comics come on. I just ad-lib and the audiences are cracking up laughing. I host karaoke nights and everyone who sings gets a gag prize, and I insult every singer. I talk about how bad their breath smells on the mic and they get get a prize that might be breath spray, toothpaste, or mouthwash. I tell them the more you drink, the better you sound.
Paris Martineau [02:01:35]:
Anyway, she's a delight and would really recommend watching. It's— it maybe happens 18 seconds into the video that I've shared. It's towards the very end of the Conan segment.
Jeff Jarvis [02:01:47]:
And nobody better than Conan. He's a great spot with rolling with punches with people.
Paris Martineau [02:01:52]:
I mean, it is. And it was just— it was just a delightful Cool little thing to see.
Benito Gonzalez [02:01:57]:
Link in the show notes.
Paris Martineau [02:01:58]:
Jeff, what do you got?
Louis Maresca [02:02:00]:
Yeah, cool. Thanks, Paris. What do you got, Jeff?
Jeff Jarvis [02:02:04]:
So I gave up some of them for the good of the show already. I was going to do the German thing, the German paradox. I was going to do the mall in San Francisco. Um, this one is a little, um, sad. So Foursquare, do you remember Foursquare?
Paris Martineau [02:02:19]:
Oh, do I.
Jeff Jarvis [02:02:20]:
So Foursquare, Dennis Crowley, who's wonderful, he went to NYU's Interactive Telecommunications program, and he created Dodgeball first, which was kind of a where-you-are thing, and then Foursquare, where you announced where you were with people. And Leo was all into this, you know, were you the mayor of this restaurant or the mayor of this bar? And it was a really cool application for a long time, but then it kind of went nowhere. And then it pivoted, and Dennis realized— they realized that they had tons of data about locations. And that pivoted the company. And it had been profitable as far as I was concerned. It was started in 2009. But then suddenly a month ago, Foursquare was acquired, ending its 17-year story. This is in a post by Dennis, by Dens.
Jeff Jarvis [02:03:08]:
The terms of the deal were not disclosed, but some have surfaced online. And the bottom line is the Foursquare acquisition was a $0 outcome for equity holders. $0 preferred, $0 for common. It's bittersweet to see these details surface publicly, said Dennis. At least now there's an opportunity to comment on them. So he's heartbroken about this. Um, uh, he left as CEO and then he was on the board and kind of left major management some time ago. And he said leadership was unable to preserve the value it inherited or create new value during the 6 years.
Jeff Jarvis [02:03:44]:
after that. And so unfortunately, it's gone. And this is what happens to some of these wonderful companies and wonderful ideas. And I always thought that he was a success story. It was a success story because it managed to find a new life in data. But that's that. So farewell, Foursquare. Darn.
Louis Maresca [02:04:01]:
It's a great app when it was around. It was great.
Jeff Jarvis [02:04:06]:
Yeah, it was.
Paris Martineau [02:04:08]:
I was always really surprised. I mean, I remember people using Foursquare. My main recollection of it was, I mean, I had somehow missed that Foursquare had turned into basically an advertising data selling company until, let me see if I can find it. There was a time when I was at Wired Magazine, I was in the pits of working on a totally different story and I got an offer through some channels that I had to come down to Foursquare and meet with Dennis. to talk about what they were doing. Let me see if I can find it. And I wrote a piece— I ended up writing a piece, You May Have Forgotten Foursquare, But It Didn't Forget You. Because basically what happened is I walked into their offices.
Paris Martineau [02:04:55]:
I knew that it was like an advertising tech company, but how they pitched to me, which I guess what they were doing was launching at South by Southwest like a— some sort of map that used all this Foursquare data and like gamified it to show all the location data they had. But how they pitched this to me, a humble wire reporter, is I walk into the office building and he pulls out his phone to show me an unreleased nameless game that him and all of his employees had been working on that basically was like Candyland, but instead of like fantasy locations, it included like place categories associated with all the New York City neighborhoods because Foursquare is in kind of Flatiron. And so I'm reading from the piece that I wrote. You may have forgotten Foursquare, but I didn't forget you. As in Candyland, you move your game piece forward by drawing cards. But in Crowley's version, the cards are the habits and locations of real people whose data has been turned into literal pawns in the game. Foursquare knows where their phones are in real time because it powers many widely used apps. These people are not playing Crowley's game, but their real-world movements animate it.
Paris Martineau [02:05:58]:
So like, he pulled out his phone, was like, oh yeah, Yeah, we're all like trying to play with each other by like picking out someone who's in the bar downstairs of our building, and we're like, we're gonna add that person's data to our pot, and if that person moves into SoHo, then I get 5 points. And I was like, that's effing crazy, dude. That's crazy. You have all that data and you're literally making a game with it just for craps and giggles. And I did write kind of a piece that I think pissed him off a little bit. But I think that's also what the company was trying to do, is I don't think that you have a Wired reporter come over to, uh, report on your company's launch at South by Southwest, which is going to be a hyper-trending app, like a real heat map of where all the users were in Austin based on the data Foursquare had collected them from other things, and then show a Wired reporter this weird second app you made without everyone's location without wanting it to be. And I don't know, I've always felt a little weird about that story because I follow him. I've followed him on social media forever.
Paris Martineau [02:06:59]:
I think he's got a really great, smart social media and blogging presence. And I read this blog post and was sad and heartened to like read his whole eulogy, which I thought was very carefully considered. But I think that the company itself, I don't know, I think it's hard to be like a data location broker type company. and then also be run by such smart, well-thought-out, heart and soul-filled people. It always seemed like a strange mix for me.
Jeff Jarvis [02:07:31]:
Well, when you also— so I interviewed Crowley for my book, the aforementioned Public Parts, which has gotten no mention anywhere for years. And he said, I happen to be a very social person who's part of a very social crew, and I live in a very social city. We're finding that when you opt to share in your location, no matter if it's a library or a bar or a classroom or an airport, it tends to lead to things that facilitate serendipity. If you can check in at the airport gate and realize that a friend of yours is 3 gates down, that makes a connection that wasn't there before. It enables things to happen in a way that you wouldn't be able to do if you weren't participating in these systems. Those are the wonderful days of the internet. We thought it was wonderful. all candy canes and flowers.
Jeff Jarvis [02:08:15]:
Yeah. And he was very public about, you know, his own life. And his idea was that if you're walking down the street and you're bored, that the app should tell you, oh, I know that you wanted to, you know, go to a good deli. This is new here, or your friend's there, or whatever's going on. It could improve your life. And
Paris Martineau [02:08:29]:
I mean, I think that that idea is great in the abstract. And in these early days, it really seemed like something that was going to happen, but it almost like foolishly naive to think about something like that existing today when like, you know, you think of all the examples of companies that have used that sort of data and suddenly are selling it to not even 20 but like 200 other companies and it ends up being used to advertise to you things or have your data be packaged to data brokers that suddenly any private investigator or a stalker could come and take that data, or that a, um, wide-scale police agency-backed search could suddenly purchase that information from data brokers and get around, uh, what some might say is your right to have privacy without executing a warranted search. There are just— I mean, I think that it is an incredible idea. I yearned for the fact that I never got to live in that heyday of the world because it does sound very fun. And I know a lot of people who are still big Foursquare users today because they experienced the ecstasy of that time. But I think that everything that has happened since has shown that it's really hard to trust companies with that much information.
Jeff Jarvis [02:09:54]:
So this is what I wrote in the book. Back in those good days. Mobile is hot too, but I predict mobile will soon be a meaningless word as we become connected constantly and ubiquitously. iPods and other tablets are transitional devices pointing to a future when we are online all the time and everywhere via our cars and TVs and ambient devices in rooms. And oh hell, just go ahead and implant the chip in my head. Yeah, that was the dreamy days.
Louis Maresca [02:10:21]:
Those were the days.
Jeff Jarvis [02:10:22]:
Those were the days. So, so, Dens, uh, sorry about Foursquare.
Paris Martineau [02:10:26]:
Sorry about Foursquare, Dens. You really love it.
Jeff Jarvis [02:10:28]:
You're a good guy. You're one of the good guys out there.
Louis Maresca [02:10:31]:
He's a good guy.
Jeff Jarvis [02:10:32]:
Yeah, he is.
Louis Maresca [02:10:32]:
Very cool. Thank you, Tef. Also, mine's a little bittersweet.
Jeff Jarvis [02:10:37]:
I apologize for that.
Louis Maresca [02:10:38]:
No, no, that's great. I think mine's a little kind of going back to what we were talking about before, which I found this just this week. Um, uh, it's how I manage all my agents. Plus you got all, all my agents internally. It's called Squad, and it's basically a framework that you can use that helps, you know, people who are novice at this just instruct this thing to go out, and it manages all of your coding agents for you, all of your agents for you. You give them personalities, and it also allows for what they call human in the loop or human-directed review. So they basically— and it has too. So it has ways to manage token usage, or, or, um, and it helps you.
Louis Maresca [02:11:19]:
It has a watch mode, so you can actually— it can actually text you or send you notes and ask you questions. Um, and it has a budget, so you have a— you have to have a specific approved budget for specific things. Um, and it's a little bit of setup just by talking to it or even texting to it, but after that you can send it off to do a bunch of things without worrying that you're going to get a $17,000 bill the next day. And it also helps experiment things, and it delegates and tests for you. And so this is like kind of the future, even home, do home things, right? Go figure out— I have to go pick up 5 kids, and I have to figure out the proper route to do all this. And map systems today can't do that. But, you know, these— this system can do it because it kind of sets up a squad and go figure out all the things it can do. And it has a, has a memory, has, um, has— it can work in parallel, um, and, uh, it needs to continue to expand its capabilities.
Louis Maresca [02:12:12]:
It has an econo mode too, which I love, which I let the kids use. Um, so it's kind of a cool framework if you have a chance to check it out. Squad, open source.
Paris Martineau [02:12:20]:
Very cool.
Louis Maresca [02:12:21]:
Yes.
Jeff Jarvis [02:12:22]:
Yeah.
Louis Maresca [02:12:23]:
Cool. Well, this was great.
Jeff Jarvis [02:12:25]:
Oh, you did a good job, Lou. Thank you.
Louis Maresca [02:12:27]:
Thank you very much. This is very interesting. Learned a lot of things myself, uh, this week. So I appreciate
Paris Martineau [02:12:34]:
Yeah, thank you so much for being such a good steward of this.
Jeff Jarvis [02:12:36]:
Yep.
Louis Maresca [02:12:37]:
Yeah, absolutely. Well, thank you everyone at home for joining us. We do Intelligent Machines every Wednesday, 2 PM Pacific, 5 PM Eastern. You can watch the show live, How the Pizza's Made, of course, on Club Discord, YouTube, X, Twitch, Facebook, LinkedIn, Kick, all the streaming services. But you don't actually have to watch live, right? You can always You can always watch the recorded or downloaded audio and video of your choice on the twit.tv website. It's twit.tv/im. You go there right now, or you can go to youtube.com and just find the Intelligent Machine channel. All the episodes are there for you to go and review right now.
Louis Maresca [02:13:13]:
Of course, you can always check out the latest episode that we just did. Now, I think the best thing you can actually go and do is subscribe. Find your favorite podcatcher that you have and go and subscribe to Intelligent Machine. machines, which allow you to get the show automatically every week, audio and video of your choice. That's it.
Benito Gonzalez [02:13:31]:
One thing before you go, one thing before we go, uh, yeah, sure, Lou, is that next week we have Jason Heiner as our host, as our special guest host.
Jeff Jarvis [02:13:39]:
All
Louis Maresca [02:13:40]:
right,
Jeff Jarvis [02:13:42]:
Jason, who's an expert in AI from his publication— I'm suddenly forgetting the name of it—
Benito Gonzalez [02:13:48]:
The Deep Think.
Jeff Jarvis [02:13:50]:
Thank you. The deep think. Deep think. Benito, you know everything, Benito. Everything.
Paris Martineau [02:13:54]:
It's true. All the world's knowledge is contained in one man's head. He's Oz.
Louis Maresca [02:13:58]:
He's Oz. We
Jeff Jarvis [02:13:59]:
never see the man behind the curtain.
Louis Maresca [02:14:02]:
We don't. We don't. Actually, I've never seen Benito until today.
Paris Martineau [02:14:05]:
People are saying— You're kidding me. There he is.
Jeff Jarvis [02:14:09]:
Oh, I like that blue, Benito.
Paris Martineau [02:14:11]:
It's cute. I like all your wires. Why are— what are they doing?
Benito Gonzalez [02:14:17]:
What do you mean? They're connecting things.
Louis Maresca [02:14:21]:
Are you an
Jeff Jarvis [02:14:22]:
operator?
Louis Maresca [02:14:23]:
That's what wires do.
Benito Gonzalez [02:14:25]:
No, that's— it's a modular synthesizer. So that thing makes music. That thing's— it's a musical instrument. So yeah.
Paris Martineau [02:14:31]:
Of course.
Louis Maresca [02:14:34]:
Very cool. Well, thank you guys. Appreciate it. And thank you for everyone, your time. And we'll see you next time on Intelligent Machines. Take care, everybody.
TWiT.tv [02:14:42]:
I'm not a human being, not into this animal scene. I'm an intelligent machine.
