Transcription
Check it out. Coming up on 30,000 subscribers, I show Speed is now facing a lawsuit after a viral live stream where he was seen hitting and shaking a humanoid robot. And if you're wondering how you can sue somebody for this, because there's only one human involved, it's the robot company suing for the repair costs.
One thing I haven't thought about with self-driving cars is when a woman who ordered a Whimo found an actual person in the trunk hiding out in there. "Are you in the trunk?" >> "That's right, man. I'm trying to figure this out. Like, I'm Yeah, this won't let me out." This >> "Why are you in it?" >> "What in the What in the world is that guy doing in the trunk of a Whimo?" So many questions.
Dan clearly seems to have put up an article on Medium with I guess technically no title. I mean, no one that I see here, but it's fascinating nonetheless in the body of the text because he argues that the universe isn't random chaos, but it shows deep numerical structure in its fundamental constants. We'll dive in.
Also, we'll look at Glenn Bochard's thoughts about why we only have artificial intelligence in the form of artificial special intelligence right now. And he predicts that humanity will not cause its own extinction. So, that comes as a relief to me.
The Neuron had a really interesting breakdown of how Microsoft and Google are kind of going different routes in their attempt to be the first to AGI. And considering it's Mustafa Solleman and it's Demisabus who both co-founded Deep Mind, it's sort of interesting. They come from such a similar place and they've taken these different paths and when you get your head around it, it's not quite clear which one's going to be better. So, we'll have to see how that plays out. But, it's fun to talk strategy.
Look into Alex Shapiro's blog about how he reverse engineered a billion dollar legal AI tool and found a 100,000 confidential files.
New research coming out of the Pew Research Organization. Turns out 62% of single women have given up on men. So, if you're curious, what does a smart single feminist woman want from a man today? I'll break it down for you. Or at least I'll tell you what Linda Carroll told me, cuz it's above my pay grade.
Control over the algorithm of any social network has been this huge pet peeve of mine forever. I need an algorithm that doesn't tempt me. So, this is a step in the right direction. It's not perfect cuz it's Instagram, but we'll talk about it.
I'll share a story about how Google's healthc care AI made up a body part. And turns out doctors didn't even notice. Yeah, I called it a typo. Experts say it demonstrates the risk of AI in medicine. It's a little of both. But now it's time to hit that hype button. If you do that for me, it would really mean mean a lot.
All right, quick message to all my creators out there. If you've been intimidated by animation or 3D work, you're going to love this. This is Domo AI and it basically gives you the power of a full animation studio in your browser. Now, let me show you why these tools have exploded this year. You can take any idea, Cyberpunk City, Fantasy Battle, Coffee Shop Story, whatever it is, and Domo AI will turn it into a fully animated video. And not those loweffort, blurry AI clips that you see on some random Tik Tok account. I'm talking about clear motion, expressive character movement, and textures that look handpainted. But my favorite feature is the videotovideo magic. You can shoot a 5-second clip on your phone, upload it, and instantly transform yourself into an anime protagonist, a comic book hero, or a Pixar style character. It's wild seeing all the comparisons. And for creatives who are building brands or clients, this will save you hours. You can prototype visual ideas, storyboards, concepts, all without touching Blender, After Effects, or any of the heavy software. Domo AI has also been rolling out new models constantly, which is why the quality keeps jumping up every few weeks. Domo AI is a suite of AI powered tools improving at a rapid pace. And if you want to try it out, use the link below for a 10% discount and create your own animated video today. And thanks again to Domo AI for supporting the channel.
And now, let's get back to all of these crazy AI updates. And we'll start with another this is a more interesting attack on robots. Like we've seen a bunch of people demonstrating how resilient they are by pushing them around, hitting them with sticks. It's never been a good look. It's always kind of given me the ick even though I sort of know it's not the same as doing it to a human. And there's just something culturally that bugs me. Obviously, I'm not the only one. A whole bunch of people online thought that I show speed was kind of over the line when he's challenging RZbot, getting all pissed. I don't know if he's just joking or acting or it's part of getting attention or he actually was like just triggered. He hurt the robot and they thought it was kind of uncalled for. So they're suing him. The robotic company who owns Rizbot was like, "Hey, you need to pay for the damages." As it says here, the cases sparked a wider debate on responsibility when creators and social media personalities interact with these high value robots, especially when these moments are broadcast to millions online. What do you think? Do you think he's influencing people to be more aggressive than they should be to robots? If he was just yelling at a brick wall, I don't think anyone would have suited him, right? If that's what you think it is, it just looks like a human, then maybe that's okay in your book. Or maybe it's entertaining, interested in your thoughts. But yeah, even if sort of logically, I know it's sort of a robot. It says, "Is this what we're teaching robots? I've never been so embarrassed to be human. I realized how Einstein succeeded. He succeeded because he read the Bible." Okay. By the way, I plan to study all other religions. Nice, bro.
All right, so here's another weird one. Whimo self-driving cars, although they don't have drivers. So, in theory, I understand how a person could have snuck into the trunk and not been seen, they're also highly surveiled. You have AI that can see all sorts of people outside of the car so you don't hit them. And I'm sure there's a camera in the car watching passengers to make sure the right people get in the car. So, this seems like a solved problem. Now, this probably happens once. This woman gets in a Whimo, finds a creepy dude like curled up in the trunk with no explanation for what he's doing there. He said, "They put me here." "They didn't put you there, bro. Like, you put yourself there." It's just such a crazy thought, you know?
Okay, so next up, let's dive into some of Dan Clearly's work. And by the way, just cuz I can see the actual domain and the the URL afterwards, it says the universe might be a construct, and I can show you why. So, I'm pretty sure that's the title of this article, even though for some reason it's not showing up for me. Point is, fascinating stuff in here. He is arguing that the physical constraints of nature are not random, but they are basically an uncanny symmetry when expressed in certain ratios. I find this stuff really fascinating and it's almost as if they follow a hidden blueprint rather than drifting by chance. So by examining how constraints like Planck's constant mass, gravity, length, time, charge, and energy deviate just slightly, but in a symmetrically uniform way, like a symmetrical form of unity, and sometimes they're just tiny bit off. He suggests that this balance isn't messy drift, but it's harmony embedded in the universe's structure. And it's from these dimensional relationships familiar equations like E= MC² fall out naturally. And that prompts the question about whether or not the universe might be a kind of computation or a construct that's sort of rooted in minimal logical rules rather than chaos. And if so, to me that kind of hints that we're in a simulation. It feels to me if we were to build some kind of simlike characters in a video game and they started studying their own environment, they would see some pretty interesting almost symmetries from the way that we are computing all of the matrix multiplications that create the 3D environment for them. Obviously, I'm just making a big jump there to say it's just like because we're in a simulation, but there is a fascinating symmetry to this universe and just the tiniest bit of asymmetry to describe some of the stuff like mass around us. But you can see here at the end of the article, this isn't just mystism. It's math. It's precision. It's the symmetry beneath the chaos. And it might be the closest we've come to reading the source code of reality. Now, this doesn't mean there's God or a programmer, but it may mean that reality bootstraps itself from first principles and that there's a mathematical seed that everything else blooms from. So, like our mathematical universe, Max Tegmark would say, maybe that's everything. And he says maybe zero implies one, one implies symmetry, primes imply structure, structure implies constraints, and constraints are what's needed to imply for a constant, right? So a constant would be something that has constraints on it, you know. So what do you think? Do you think the constants we use aren't arbitrary leftovers of the cosmic dice roll, but instead they're part of a blueprint or not? Leave it up to you.
Next up, let's talk about artificial special intelligence. I keep want to say AGI but no special intelligence is kind of what we have now. So this is Glenn Bochart's article. He argues that AI systems that we interact with today are not general intelligence. They are specific for certain roles. Machines with humanlike broad intellectual understanding is not what many of these AI systems do yet, but instead it's what he calls artificial special intelligence. This is a narrow form of generality. And he explains that despite rapid improvements in AI speed and task performance, current systems lack the holistic awareness, the deep reasoning, and the truly universal understanding that we would expect from AGI. You guys want to go look at the AGI countdown really quick? Oh gosh, I never even know anymore. Do you think it's still at 96? I hope so. Yeah, still there. Getting closer. Any updates? Yeah, Gemini 3 Cloud 4.5. Now, of course, I'm sure he would argue with Allen's conservative countdown to AGI because Borchart points out that AI today can surface information and mimic analysis, but is does not genuinely understand or integrate all knowledge that's available online. It simply processes patterns with narrow domains. And he contrasts this with the idea of AGI, which would be aware of everything ever on the internet, something current AI clearly isn't. His view extends to how intelligence itself is defined as a response to environmental changes. And since current AI doesn't truly embody that across contexts, it remains quote special rather than quote general AI, it can outperform humans in all sorts of specific things and data processing. What we were just learning about ending bench seems to be able to analyze things so much better than humans in some ways, but yet just drops the ball, especially on on context problems. I have a super broad concept of intelligence. Now I am starting to look at slime molds and and mushrooms and crows and dolphins and squids and anything that solves problems. Sometimes even automat has a simple rule that seems to create patterns. We think of what we have as humans as kind of the end all be all. I think it's the tiniest tip of the iceberg and there's so many other ways and it's going to shock us how many new ways of being intelligent we're going to discover in our lifetime. So yeah, even human, even if we can get it to be as general as a human, it might still be narrow if you want to take the broadest view of it.
All right, so next up from the newsletter, the neuron, they had a kind of interesting article in here, a little part where it talked about Microsoft versus Google. And I kind of forgot how interesting it was that if you go back more than a decade. This is even before OpenAI showed up on the scene. You really had one leader in the whole world, the team that was DeepMind started by Demisabus and then sort of acquired but left alone by Google to pioneer so much of the research that we use today. They were the ones just doing it before everyone else and they were funded enough with the kind of geniuses that actually cared about this problem when no one else did. Now there were two co-founders. One of them now works at Microsoft and has access to lots of compute sort of in a way competes with open AAI but they also partnered also has this history with his competitor now over at DeepMind funded by Google and has access to all sorts of things. So these are two extremely smart people who have kind of came from the same leaf, the same root I guess you would say and they have a bit of a different vision of how to command the massive amount of resources and influence they have. So Microsoft and Google DeepMind are chasing AGI with radically different philosophies. Deis Habibus at Google is treating AGI as a scientific breakthrough tool, something that can solve fundamental problems like fusion, superconductors, and a true physical reasoning. Now, his strategy splits effort between scaling compute and doing deep research, aiming for systems that actually understand the world and can generate new knowledge, not just remix human data. Meanwhile, Mustafa Sullivan at Microsoft is focused on building AGI as an economic engine, practical, controllable, and deployable. Now, he's less interested in quote winning AGI and more focused on rolling out reliable AI agents that replace apps, generating real economic value, and keeping humans legally and operationally in charge. So, it feels like he's very practical. It feels like at Microsoft they're just saying build AI into tools that people will use. They'll use them. We'll have a feedback loop that will help us grow and then that'll kind of give us the lead in the long run. It's almost like safe super intelligence which doesn't even do a product. You know, it's almost in stealth mode, but they're just going to try to straightshot it. They say, "I don't want to waste time collecting feedback and trying to build a product. I just want to get to AGI and then it solves everything else." So, you have these two ends of the spectrum. Which one do you feel like is a better kind of route to apply this stuff nowadays? If someone put you in charge at Meta and said, "Hey, we're like redoing everything, but here's all the money and resources you need to compete with Mustafa and Demis, how would you position yourself to be different? Or would you just try to do the same philosophy, but execute faster?" It's so weird. The negotiating, the executing, the creating value on behalf of the people that give you money, but also knowing that you're going to be displacing jobs. There's going to be accountability. there's going to be somebody legally responsible for a world that changes really fast as machines eat the real world and where should the companies be and how should they be held accountable and and if they make I guess too much money is it unfair does it need to be reinvested in some way that helps the general public a lot of interesting questions and they have two very different philosophies so it'll be interesting to see who wins or Mustafa would say it's not even a win-lose thing there just there's going to be all sorts of winners in different ways he's more product focused a little bit closer to Sam Alman
All right, next up, let's dive into Alex Shapiro's breakdown about how this AI legal tool was just just not secured properly and what he found. So, Alex is a security researcher and there's this AI platform for legal work. It is called Filevine and he discovered that there was a critical API vulnerability that exposed nearly 100,000 confidential files from a law firm instance. So that included attorney client privileged communications, HIPPA protected records, internal memos, and court documents. And because of an Amazon Web Services API endpoint that required zero authentication, it could return full admin credentials to a box file system with just a simple JSON request. Now, Shapiro responsibly disclosed the issue to Filevine in late October, and then the company confirmed and patched it by November 21st. And then he published his findings on December 2nd. So, you know, he did the right thing. Hopefully, nobody got anything leaked. But the idea that there could be these errors like a billion dollar AI tool that's used for sensitive legal work had just a wide openen API with no authentication. I mean, nuts, you know, like I can't expect there to be no mistakes. But, you know, humans finding this stuff is going to be rare. AI can just go out on the internet all day long and find any weakness and the instant it finds it. It could patch it if it's doing it for the right reasons or it could exploit it if it's doing it for the wrong reasons. And it just everything is going to go so fast soon. And we're, you know, I hate to say we're in a race condition, but we kind of are. We kind of aren't, but we kind of are. And it underscores a broader industry problem where every startup is just like rushing to kick AI out and like get new users and so many people want to use it that they're trying new products that aren't, you know, super well vetted and they're startups and they don't have enough resources to build this slow and carefully enough or they're not using AI on their own end to actually like secure things. So yeah, just remember due diligence is important. Just cuz something has like an AI feature, it doesn't guarantee that it's protected or the personal data is in place. In fact, AI just seems so loose. Every time I use it, I think, ah, you pretty much like remember this forever. I got a feeling somebody else could talk to an LLM at some point and like pull it pull it out the the information I fed it or at least some generalized version of it and maybe even the specific data if it's not secured, right? OpenAI had some history issues with their growing pains. you know, bigger companies obviously have more robust security built out and they can kind of implement AI inside of some sandboxes where they've already secured it and that's sort of comforting, but still there's just so many new security issues and I don't feel like any LLM is just truly unbreakable.
So, the Pew Research Center just put out a report. Turns out nearly half of US adults say that dating has gotten harder for most people over the last 10 years. But even more shockingly, 62% of single women have given up on men. I mean, I got to admit, I don't feel like we're doing that great of a job as a gender, but like, you don't have to give up on us. The Guardian asked single women why they're not dating. And one woman named Danielle said, "Too many men have a problem with incel rhetoric." And that woman after woman repeated the same sentiments, incel, red pill, manosphere. But this Medium author, Linda Carroll, is arguing those guys are not the main reason 62% of young women have lost an interest in dating. It's the other guys who have been affected by them. 86% of men say they're stressed out because they're struggling financially and can't be a good provider like women expect them to be. But the TLDDR is that there's a disconnect between what many single men and many single women currently want out of a relationship with a notable percentage of women choosing independence over traditional dating. a bunch of guys in my audience that like want to probably date women, you know? So, I feel like health, longevity, relationship stuff is still kind of important sometimes.
All right, so here's some good news. Actually, you finally seem to have some control over your algorithm on Instagram. this is truly what it seems like it is and we got actual open access to algorithms and there was a real marketplace for them and you could you know sort of purchase them or vote with your attention in ways that were dictated by your choices it would be such a big thing we don't even have that option but then if we had the option then we could maybe start blaming people like well you chose to like get thirst trap so you deserve it you know right now we don't have that choice and we're kind of forced onto these platforms because of the network effect so it's not like we really can just go to any network cuz not everybody's on every network. So when Instagram introduced a new AI powered feature called your algorithm, I was pretty interested. It gives users a transparent look and some real control over the algorithm behind their reals recommendation. I want to give them kudos. As much as I got a problem sometimes with how these social networks have been built, this is something we should we should show thank for. Like everyone go use this because it'll show meta that like the users actually do care about controlling their environment a little bit. Users can now see what topics Instagram thinks they're interested and based on your watch history and your engagement. And then you can add or remove topics to influence what content shows up in your feed in real time. So, the aim here is to make the algorithm like a little bit less opaque, more personalized. It responds to your growing concerns about automated content curation and echo chambers. Now, it's not a fully open thing. There's an interesting play here. Meta can also take, you know, the changes you're making to help them serve you better ads to help them learn more about what people want. It's not just the interest of marketing like selling an ad to you. At least now you can kind of say, well, you know, I understand maybe a gambling ad it would make the most money, but I'm not into it. And they're going to say, oh, well, we want him to actually keep being on the platform, getting the content that he wants, so maybe we'll try to give him whatever he wants and make as much advertising as we can there. So, it's I'm not saying they're doing it out of the goodness of their hearts. I'm not all about it, but this tool needed to get built. I'm glad somebody built it. I'm glad it's a start. And Meta said they have plans to extend the tool beyond reels to other parts of the app as it rolls out more globally. So take note, use it. I think it's powerful. I already did go in and adjust my topics. I'm trying to be smarter. Trying to get less garbage, but you I just don't need like war fights, news, thirst trap. Like I'm just like Jesus. Like just chill. Got I've been trying to get off short form, but it keeps sucking me back in. So yeah, let's actively shape the algorithm's behavior as a thoughtful group of citizens. Okay, I know it's a big ask, but maybe it's a small step towards a safer future, a better future. Yeah, maybe even safer. Maybe it's a step towards alignment in the broadest sense, but you know, not really, but kind of.
Okay, so I wanted to show you this article cuz it's it's a little bit older. This was like from May, so it's not like exactly the newest version of Gemini or anything, but I came across it and I've been making this joke about Baslier ganglia. I know it sounds like something someone really smart would say about the body that could be real, but I just never heard of it before. But it's not. It's hallucinated. Basil or ganglia. If you hear somebody being like, "Oh, my basil or ganglia is like so tired. I feel like it's not working." It's I didn't get this at first, but it's because it's a joke. It's a hallucination that made it into an actual medical thing that came from Medge Gemini. So, now that it's kind of like popping up a little bit, but Basley or Ganglia, if somebody says it, they're like trolling you. Okay. And I just learned this. So back in May 2024, there was a research paper. It claimed that uh it could this Med Gemini model could interpret medical scans, right? So doctors were, you know, it there was an AI that confidently identified a condition involving the Baslier ganglia, which is a body part that does not exist. Medgemini appears to have mixed up two real structures, basil ganglia and basier artery. Those two things actually exist in the world. Basley or ganglia is the fake body part that Gemini hallucinated. And I'm not saying humans that write these papers don't make mistakes. I probably would be surprised if a human hasn't made a sort of typo like that too. That's a body part that doesn't exist. But it just kind of shows that AIs can hallucinate. And I know we're getting better at them, but still. Basier ganglia, if you hear some jokes about it, now you know. Now you know. Don't want you guys to get trolled on by the young generation. You know, we're getting old or I'm getting old. So, I don't want those kids being like, "Look at that guy. Doesn't even know what basil or ganglia means."
Now, if you don't mind getting your Basley or ganglia focused on the hype button, go to your YouTube app, swipe on the comments, hit that hype button. Help this. Dude, look at how many hypes I got on my last video. You guys used your Basley Ganglia so good. Look, that's for sure a record. Over 200,000 hype points came from 268 actual button presses. Boom. It It like helped my video do so much better than average. Like you'd see at this point I normally would have Sorry, that's engagement. Yeah, you could see normally at this point I would have maybe between 5 what 5.5 and 6.8,000 views on a video. This one's at 10,000 and above. Dude, I got a 100 new subscribers. I made $70. So yeah, man. Those hype points, they really help the channel grow. So, you know, if you don't mind, if you made it this far, Basley or Ganglia, I'm out.