UL NO. 423: AI is Becoming Like Reading

into it here. So, yeah, doing a whole bunch of new stuff with the desktop swapped out Kitty as the new terminal got rid of alacrity, which was cool for a hot minute there, but I think that lasted about a week until I realized it had a lot of problems. Someone recommended Kitty in the GitHub project for alacrity, and I ended up switching over. And I just like a better for lots of different reasons. And both of them,

in my opinion, are better than I term two. Both of them have text configuration for for configuration, you don't have to go into like settings with a UI as oxide. You buy for a window management SCT for managing shortcuts for your buy sketchy bar for this new bar. You can see it up here and uh, Sto for syncing files. Pretty cool stuff. And we're going to be talking about this a whole bunch during a UL meeting on Thursday. So that is coming up this Thursday if you are

a member. And yeah, I haven't been this excited about a desktop configuration. Like I don't want to if I start flipping around right now to show you it's going to become a whole video and we don't have time for that. So I'm just going to get into the show. But I'm more excited than I've been for a very long time about desktop configuration and just optimization. I'm just I don't want to leave my home keys, right? I want to vim all the things right. And that means

like screen management. I'm looking at three screens here. I've got a, you know, the Apple Z1 in the middle and I've got two LG five KS, so it's six K in the middle, two five KS on the side. I can move between them, I can move between desktops inside of it inside, you know the Mac OS management of desktops as well. And most importantly applications and windows inside of the interface. But again, I don't want to

talk about it too much. Already have. All right. So fabric now supports clod opus and opus is actually the first model that I've seen that is actually better. I've seen tangibly be better than GPT four. So I didn't think that was going to happen before GPT five came out, but it did. Opus is actually better in numerous cases that I've actually physically seen, and one of them is actually finding hidden message, which I talk about right here.

So this it's actually much better on this particular thing. Yeah. So this is essentially how you use it. So you do PB paste or whatever the input is. You take any input from stdin or standard in and then you send it into fabric extract ideas, and then you set it to Model cloud three opus and you give the model number. And we're actually about to make it. So you don't need the model number. You could just send it to cloud three opus and they'll use the latest one.

Also added a new pattern. That's basically how you use cloud but added a new pattern called Extract Predictions. So I'm going to basically send this content from whoever. And it's going to rate including my own stuff, and it's going to rate like how certain I am about the,

the prediction and also what the actual prediction was. And when I said it would happen by and I'm going to collect this for a group of people and then retroactively look back and say, you know, I'm not sure like how good they were at doing this, because it looks like all these things are wrong. And with automation plus AI, you have the ability to just like, go get everything they ever said and look at all their different predictions and see how valuable they were. And fine

hidden message is now more effective. And it actually comes back and says, here's a really cynical view of what they said, a normal view and then like a favorable view. So that's pretty cool. I wrote a couple of new essays this week. AI is already becoming like reading. Kind of depressing. Not going to get into that. AI is worse if you think it's someone who's fault, somebody fault. So yeah, that one a little bit about framing essentially.

All right. Security. So had a Google engineer indicted for stealing AI trade secrets to benefit China and basically took over 500 files related to TPUs. Essentially tensor processing units is what it was. And he actually lives not too far from here, a few streets from here actually, like a five minute drive and, uh, yeah, pretty, pretty sad.

And what I talked about here was basically, if you don't want to like, go crazy with what's going on with like, how this keeps happening with Chinese employees because China is overreaching. China loves espionage. I mean, this is just common knowledge. And they love to tap into their, you know, people who are deployed all over the place. Right? And it doesn't I don't know the exact details of this, but it's not like they're actually deployed like a military unit.

It's more like, hey, I could tap you on the shoulder at any given moment. Somebody named Andrew Bustamante talked about this in extreme detail. At one point he's like, look, this is the biggest and scariest spy service in the world because they consider all of their, all of their citizens to basically be an extension of the spy network. So at any given moment, they could. Just tap you on the shoulder and say, hey, it is your duty

as a Chinese citizen to help China. And that does not mean that anyone's going to respond to that, right? I absolutely don't believe that people have been here for a long time, have loyalty to the US, that they're going to do that. Right. And that's why I talk about this. It takes a lot of courage and wisdom to simultaneously realize how bad Chinese government espionage is without giving in to racism, right? And a really good answer to this is to have a really good insider threat

program that looks at behavior rather than characteristics. But the problem here is that those programs are really difficult to build and run and maintain, and so they tend to only exist in big companies like Google. And this was actually Google where this this was discovered. So the question is like where is it also happening but not being discovered? Big story. Midnight Blizzard is basically pounding on Microsoft. And that is a Russian group. Qnap alerts more Criticals. I mean,

never under any circumstances put a NAS online. It's like like I say here, it's a perfect storm of the most critical data that you don't want to lose or have stolen. And then the worst possible code just absolutely nasty. And thanks to Plex Track for sponsoring North Korean spies hacked into South Korean chipmakers stealing designs A flipper zero device was used to break into a Tesla, but it was a little bit of like kind of a stretch here. First of all, they didn't have to use a flipper zero.

They could have used anything that just stole Wi-Fi because it was like a fake Ssid. And then you had to have multiple things sort of stacked together. You had to be right next to it, like a bunch of stuff that didn't make it super serious, which is why

I think Tesla didn't take it that serious either. Scammers are increasingly using AI to mimic the voices of loved ones, and my advice here is to let your most vulnerable family members and friends know that scammers can now fake voices and everything, and that if they think you're in trouble, they should try to get in contact with you or get in contact with someone who knows that you're okay, someone around you, in your circle and in scammers. Do, I think, try to do this? I know this for

a fact that they do. Not all can do this because this takes time and effort. It it's what we call in security world. Expensive to try to determine what your current state is, but the more advanced the attack is, the more money they would get from it, the more likely it is to happen. But the point is, scammers do try to figure out how to do this when they know you won't be able to get in contact

with them. Okay, if you're on a vacation or you're, you know, whitewater rafting or something and they know you're on the river, again, this is less likely. But if they know you're on a river, you don't have cell phone connection. Well, that's the perfect time to call grandma. And here's the problem. People post that they're whitewater rafting. People post and say, I'm going whitewater rafting. I will be unapproachable, unreachable for five hours. See you when I

get back. Well, thanks for mentioning that because now I can call grandma and say, hey, they had a rafting accident and the whole family needs surgery and we need $80,000 to get the best doctors. And guess what? Grandma immediately calls you and gets no answer. Or even worse, the phone is out of service. Holy crap, now they're freaking out. And then they use these tactics to just make it more and more pressure. It's like, oh, the doctors really worried. They said, they have to start now.

I need your credit card. And so that that sense of urgency is really, really bad. So you've got to train the more vulnerable people in your life to like, get ready for this. It's especially bad. And this is kind of the point of this one is like, what if it's in your voice? Okay, it's a voice message from you saying, please wire money to this account. Otherwise these kidnappers are going to kill me, okay? And it is. It sounds exactly like you. I mean, that that is

not in the future. That's like right now they can make something sound like anyone. So you've got to prepare your family and friends for this US sanctioned individual and entities behind predator spyware. I like the fact that they're going after people like this. Cloudflare has a new firewall for AI and looking at protecting yeah, basically AI from security threats. And my my analysis here is so nimble. These guys, they they basically seep into all the cracks.

They're slowly becoming the internet. I honestly believe this. Cloudflare is quietly becoming the internet, right. If Google like gets rid of Gmail and YouTube out of sheer stupidity, which is not out of question, and Akamai gets bought by whatever Johnson and Johnson and boom, suddenly Cloudflare is like the internet. Brian Krebs has a really cool analysis of radar, a data broker that sells all sorts of American data, and they've got ties to Russia and all sorts of badness.

And Krebs does a great job as as always taking that down. And Russia, setting up fake news sites. Can't wait to actually start consuming these things. I'm going to start consuming. Tons of online content and actually doing multiple filters on it. And one of them is going to be looking for propaganda. So I'm hoping to detect these types of networks before anyone else does. China is increasing its defense budget by 7.2%, and this more than doubles

the military budget under Xi's leadership. Technology for AI is a cold calling AI service. You have got to check this thing out. It is the craziest thing and I've got links here to the video. I don't know, I might start doing a longer version of this where I can actually play the videos, but this, this one here. New voice demo. This one actually has a female southern person calling and you can hear the accent. It's really endearing. I spent like, whatever, 12 years in Georgia, got my

girl from there. And like, the accent is just amazing and it just kind of makes you open up and like, listen a little bit more, which is the whole key to a cold call. And anyway, this thing is it's insane. It's the coolest. Like sales automation thing I've ever seen. And it yeah, it uses AI. All right. So it's been giving AIS different IQ tests actually this matrix IQ test. But they're giving different LMS the matrix IQ test. And Claude three just broke over 100 for the first time.

And 100 is like the average human IQ. Apple has quickly shifted from a passive stance in AI to going crazy with it, and can't wait to see what they do in September. Apple podcasts now have auto generated transcripts. Research is saying that models will soon be able to optimize prompts better than people, so prompt engineering might be just going away. I think that's kind of right, but at the same time, you can't just give it total garbage, right?

I don't see some random person be able to just bark idiocy at a model and have it say, oh, you were referring to Feynman's third principle. Indeed. Let us proceed. Don't think it's going to be able to do that, but if you do voice something that's halfway decent, they might be able to like, clean it up and be like, I think you meant this. Global trust in AI is falling, waning, dropping off with a significant drop from 61% in 2019

to 53% now. And this is why I wrote one of the essays that I was talking about up above France just put the right to abortion directly in the Constitution, and somebody set a new rule capping credit card late fees at $8. This is going to cost banks a lot of money. A lot of people are basically saying, hey, look, they're just going to find ways to get that money back with other fees, but it doesn't matter. I like the fact that the government is actually pushing back on

this really cool, raw image. Look at this. This is insane. This is why I want to do a video podcast instead of a audio. I mean, look at this. I've already got the Hubble version of this, but this is the new one from the Webb telescope. Look at this thing. This is ridiculous. Every single thing you see is a galaxy, not a star, a galaxy. Actually, there might be some stars that that happens sometimes, but yeah, really cool. So I'm going to basically order this on a metal print

like I have for the Hubble. One single dose of LSD shows promising results for treating anxiety 48% remission at 12 weeks Sweden is now a NATO. New York has rolled out National Guard troops to protect its subways. That's disturbing. And Scotus ruled unanimously to keep Trump on state ballots. Yeah, I consider what Colorado and some other people tried to do to get him off the ballot as a Brexit move, and I call it a Brexit move because it's the

type of thing that you try to do. You get it done and then you immediately regret that you did it right. You don't take people you dislike off the ballot. You don't do that. It's not American. It's a bad thing. And like I said down here, the moment you do it and it actually works well when you're not in control and it comes time to put up your own candidate, the opposition is just going to do it to you and it's going to get worse. And pretty soon you

have Banana Republic. It's not a real democracy. And here's the other point. If somebody can be elected and you find them so detestable, but they actually get elected like they they legally are allowed to run in the country votes for them. That's the game. That is the game. And I do not like Trump. I'm not a Trump guy. Doesn't matter. And we're not going into politics here. But he's not my guy. The point is, if the country votes him in and that very well may happen, then

that is American. That is American. For somebody to vote in in a free election, of course, which I think it will be. And it's like, that's what that's what America wanted. So either change the country or leave. I mean, those are my options at that point. All right. Ideas and analysis. I was troubled with Harare's analysis on Colbert. So he basically said, down here. He was talking to Colbert and he basically says, we have no idea what to teach kids growing up. Now we have no idea

to tell them what kind of skills to have. And we used to be able to tell them, oh, you know, shoot a bow, you know, ride a horse, you know, farm wheat or whatever. And now we don't know what to say. And I just don't think that was good. I mean, he's he's the best. I got massive respect for him. And I also want to give him, you know, some props because like I say here, the lights are bright, right? I mean you're encouraged to give like these little soundbites.

So so I'm not like throwing shade, but I just wish he didn't come out and say, we have no idea what to tell people because we actually do write clear thinking, clear communication, understanding the past, understanding the merits of various arguments, how to disagree, philosophy, critical thinking. All these things are more important than ever and more obvious than ever that we need them in the future. So it's kind of the opposite of what he's saying. So

that's what I basically said in this tweet. All right. Been drawn back into the Stoics lately, been reading a whole lot of Marcus Aurelius, and really got a whole bunch of stuff from Ryan Holiday's Stoic store. And you should check that out. And oh, yeah, we're doing our mid-month on optimization. I think I talked about that already, and I should go sign up. That's the link to go sign up Daniel missler.com/premium. And yeah, this newsletter was

super fun to do. Really fast, more creative than ever. Just it's going great. Don't really need to say too much about that. All right. Discovery obsidian as a graph database for Rag. This is really cool because people put so much effort into obsidian. Being able to apply AI to it is super, super cool. Male in the middle automate spearfishing. Enough said. Junaid Islam outlines a five step method for cutting cybersecurity budgets. This was a pretty cool

write up. During World War Two, America fought against damaging rumors with rumor clinics in newspapers and magazines. I think we should look and do this right. We should look into this so debunked lies by fact checking and publishing the findings, helping to maintain morale and unity. Um, pretty cool. I get Julia Evans, the surprisingly complex world of Jets head, revealing its multifaceted roles. I am embarrassed by how often

I realized I don't know anything about Kit. I just too many times per month, I just delete the whole I'm like, I'm confused, delete the whole repo, re clone it. And that's just sad. I need to get class. All right. It's getting harder to tell humans apart from bots, but not because bots are getting more smart or more like humans, but because humans are acting more like bots. Oh that's disturbing. How to start a home lab? Really good primer here by Hayden James. And this guy Chen is all excited

about soft skills. He says 80% of failures at Amazon are due to soft skill issues, not technical ability. And it says that he got a job at Amazon even though he wasn't technical because of this skill. I'm not sure it was actually him that said 80% of failures, but I wrote that anyway. All right. Tolkien intensely disliked Dune because Tolkien wanted to talk about inherent acts of goodness, and Dune talks about consequentialist morality of action judged by

their outcomes. I like it, I like it, and I also like this word. I wrote a piece about this a while back. Deontological. The morality of something depends on the effects that it has as opposed to intentions. Pretty cool concept. Someone reminisces about the simpler, less polished writing they used to do. They express a longing. This is on Reddit, by the way, a longing for their earlier, unrefined work. Feeling it had charm that their current writing lacks.

This is absolutely true. I've experienced that in the past, and Kate Hall shares how anyone can learn to be more energetic. This is pretty cool. It's over on the site every, which is a really cool site from Dan Shipper and the recommendation of the week. Trust your routine. If you spend a lot of time putting together a routine that helps you feel good and positive, then you got to follow it. You got to follow your routine. Whenever I'm out of routine, I'm not feeling good. I'm like, ah,

it's got to be something else. It's got to be. I got to figure out what it is. Is a supplement or something? No, it's it's not a supplement. It's not something special. You're just not on your routine. Right. What's the routine? Something like this very Huberman esque sleep, sun exercise, clean food, walking. Talk to your friends and family. Like be social. It's the basics and there's a reason you wrote them down, so go do them. And the aphorism of the week. Choose not to be harmed and

you won't feel harmed. Don't feel harmed and you haven't been. Choose not to be harmed and you won't feel harmed. And if you don't feel harmed, then you haven't been. Marcus Aurelius. Unsupervised learning is produced in. By Daniel Miller on a Norman 87 AI microphone using Hindenburg. Intro and outro music is by zombie with the Y, and to get the text and links from this episode, sign up for the newsletter version of the show at Daniel miller.com/newsletter. We'll see you next time.