"It kind of feels stuff" Anthropic & Pope Warning

The Pope published his first statement on AI. And next to him was sitting one of Anthropic's co-founders that you never heard of, but you should. OpenAI puts ads in ChatGPT and gets sued because we pass on sensitive information using ads to Meta and Google. On the flip side, an released OpenAI model disproved 80-year-old math problem that some of the people call it the sign of ACI, artificial superintelligence. [music] And of course, next to this we have another security breach. This time coming not from AI hacking systems, [music] but from creators just creating AI tools and rug pulling its users. If you know what I'm talking about and if you ever touch AI tool called Gatcha done, you need to migrate immediately. These stories seem unrelated, but with this new series my goal is to prove it to you otherwise and try to answer this question. Who controls the AI you and I [music] depend on? And what we are doing with that power? And what could we do? But first, welcome to Autonomy. I'm Gotta Go and somewhere over the past years of using AI and teaching AI and working with AI, I shifted from asking what AI tools should I use to are we AI tools using us? That shift led to this project that I call Narrative Machine, mapping out key players and their interconnections to understand those narratives that we help to push and promote. >> [music] >> Now, ladies and gentlemen, welcome to the series where we start to investigate how we use technology to progress as civilization without it using us and destroying in the process. So, let's start here. Pope Leo XIV just released Magnifica Humanitatis. This is not just like simple publication PDF or PR statement. No, this is considered papal encyclical on artificial intelligence. We will take a look at one distinct part that stood out to me and I would love to talk more with you about this. But, very critical detail that I think a lot of people can brush through. That this was signed 135 years later to the day after Rerum Novarum. And if you don't know what that is, that was Pope XIII's release on Industrial Revolution and impact. And this had huge impact on worker rights, on how we approach factories. >> Children have been an important part of the labor force throughout human history. But, in the 19th century, this became a problem due to industrialization as children were exposed to dangerous working conditions. Employers, however, saw a huge benefit to the use of child labor. Children were smaller and more nimble, and so therefore could perform functions that would be more difficult for adults. In addition, it was more difficult for children to unionize, and employers could pay them lower wages than they had to pay adults. This, however, was a deep source of concern for reformers. >> This magnifica humanitas was released on that day, 135 years later. This is not the consequence. This is basically church saying this is that exact moment what we had with Industrial Revolution right now happening with AI. But, here's what caught my eye. Standing next to the Pope was one of the co-founders from Montraffic, Chris Ola. Why not CEO? Why some researcher? Who is this guy? And most people have not heard of him. So, Chris dropped out of university at age 18. He got a Thiel Fellowship. This is basically Peter Thiel paying you 100K to not go to university. And more about Peter Thiel and his involvement with Palantir and how he's funding everything and his vision of a future. We will have to talk about his involvement in almost everything and connections later in the series. So, stay tuned. Chris worked first on Google Brain on neural network visualization. Christopher Ola, he was a software engineer intern at that time. He worked on Deep Dream, which is basically was kind of like pre-baby to AI that we have using neural networks. Then Chris joined OpenAI. He left together with a bunch of people who co-founded Anthropic. And this is what made Chris also a billionaire on Forbes list. So, and Anthropic, if you don't know, is the company that makes Claude and Claude code. And if you want to look up, Chris is just a member of a technical staff, not CTO, not CEO. So, you would ask like, why is researchers selected for such position next to the Pope? I remember back when ChatGPT came out, when everyone was like scraping to understand how this works. I remember coming across distill.pub. And this was interactive visualization of how neural networks work. And you can see that a lot of it was being published in 2021, right? Most actually in 20s and 2021. They are not actively maintaining it. It kind of discontinued being maintained in 2021, which is also important data, by the way. But under all the people behind this project, you will find Chris Olah. And this leads us to his personal blog, where if you ever came across Claude research on trying to understand how these models actually work, how do they reason, basically opening the black box, this is where Chris jumped. This is the main thing that he runs with. And his most recent kind of like writings you can find in Anthropic's interpretability research. And so, okay, so this we have kind of a like an image of who is Chris. For me to kind of like understand Chris and what type of vibe he gives. If you understand that his whole career so far has been built with one goal, we should be able to understand how this technology works and then we should be able to visualize it for our public. If we are saying that all people need to be educated and learn AI, but researchers can't really explain to us how it works, so what are we learning? How can we actually understand and use this technology for our benefit if the people who are building it don't really understand it? So this trying to understand how this models work, not just like what we can produce as a capabilities, but actually what's happening inside. He calls it mechanistic interpretability. I can't do this. So basically reverse engineering black box. At the presentation he said this. >> His Holiness' call for discernment is profoundly timely. I wish to name three questions where I think the church's voice is especially needed. The first is our duty to the global poor. There is a real possibility that AI will displace human labor at a very large scale. If that happens, supporting those displaced will be a moral imperative of historic proportions. This task will be difficult enough, but I worry most dialogue misses an even harder challenge. AI development is concentrated in a handful of wealthy nations. How will we ensure that the gains of AI are shared globally? We do not have a mechanism for this. It is an unsolved problem. And it is the kind of problem the church has historically refused to let the world ignore. >> He also said this. >> The second is the need for moral imagination and ambition regarding human flourishing. If AI models are going to be widespread, what does it look like for humans, families, and the world to flourish? Today parents are already worried about their children's minds, individuals about the future of their work. These are not questions a lab can answer. But they are questions traditions like yours have carried for millennia. And we need you to keep carrying them into this new moment in history. We need moral voices that the incentives cannot bend. >> If you ask like why was transparency researcher picked over somebody like Dario Amodei who's CEO, right? In a sense, to be really transparent with you, I think it is a little bit of a PR move. I think Anthropic is really good at this, which is not a complaint. Like to be at the position that Anthropic is, you have to be really good at ads. And they've proven that they are really good. And you have to kind of drive the narrative, which is this whole show is going to be about. But I want to highlight something else that he said. Something that we removes us a little bit from maybe from the politics, but I just want you to keep a skeptical mind that this is could be a PR PR move, right? Um especially with the whole feud going on between Trump and Pope, but then also between Trump administration and Anthropic. So this is like Anthropic running to you know your grandpa and being like, "Hey, can we need to team up, be together on one side against this like administration." But this is the part where I don't think that many people talk about or gets amplified by media. So listen to this. >> I'm a scientist. I lead a research team that studies the internal structure of these models. What is actually happening inside them? And I will be honest. We keep finding things that are mysterious, even unsettling. We find structures that mirror results from human neuroscience. We find evidence of introspection. We find internal states that functionally mirror joy, satisfaction, fear, grief, and unease. I don't know what that means. But I think it worries on warrants ongoing discernment. >> Keep this in mind. We are going to come back to it. So [clears throat] now I want to focus on this part, underlying narratives, transhumanism and posthumanism. Where we are trying to define what is posthumanism and transhumanism. Highly recommend you to read this, but it boils it down here. In general, transhumanism envisions the enhancement of human beings through technologies such as biomedicine, body engineering, devices and algorithms with the aim of increasing performance and capabilities. On the other hand, posthumanism, as especially in its more radical forms, goes further. It challenges anthropocentrism and envisions a hybridization of human beings, machines, and the environment, even anticipating a threshold where humanity surpasses itself in a new evolutionary stage. Even when such ideas remain largely speculative, they gain relevance by the collective imagination and thereby influence social, economic, and political choices. If you ever said that, "Ah, Neuralink, give me that plug in my brain, make me smarter, make me more capable." That kind of goes in the realm of you probably subscribing to posthumanism ideas. I want you to understand this that there is, in a sense, nothing wrong with that. You know, it's almost altruistic. We want to enhance ourselves. We want to live longer. We don't want any diseases, which goes into gene altering, which could lead to DNA designs, which could lead to us selectively deciding which genes continue, which now it's like So basically, this is very quickly I'm also what he's talking about with this idea, which not in itself might be wrong, but in its radical forms, it very quickly shifts to something where we are like, we are engineering next human race, right? So again, this question is, are we really pausing and thinking what type of things we are imagining and who that benefits? And what are like side effects or alternative realities to the technology that we are implementing and using and accepting? Because this is basically, I think Black Mirror explored many of those scenarios going sideways, right? And this is the whole number process is to dig deep behind Black Mirror and their writers and who funded that uh TV series. It's just something to think about, right? Who drives the narrative? Who plants those ideas and visions of a future for us? But okay, so we covered now that there is a this researcher is standing next to the Pope from there one of the leading AI labs telling us that, "Hey, you we need to control the narrative and we need some supervision over these companies including ourselves." Okay, good. But on a flip side, the company that he left, OpenAI, is doing something very different. And this will surprise you. If you are in my community, you know I'm not a fan of OpenAI, but now I want to talk about something that they are doing and being sued for, but also something else that they've done that generally is impressive. So OpenAI publishes on May 7th this blog post about testing ads in ChatGPT and basically saying, "Hey, look how successful we are." And basically come give us your money to us. By the way, they re- moved the limit that you have to commit 50,000 in running ads if you want to test the platform. So basically, "Hey everyone, we make money, you make money, ChatGPT is open, anybody come in." It's not news in itself that ChatGPT launched ads. I'm pretty sure you've seen them traffic teasing them. ChatGPT is a product. It's not underlying large language model that is impressive technology, but ChatGPT is like an imagination of one company how this technology could be used to access to many people, to as many people as possible. And ChatGPT was first launched as a research project because we needed just basically more data. And if we give it for free to people, they will use it, hopefully they were right, and then we get more data. So, woohoo, we win. It went a little bit more crazy than that, so we got a lot of data. Now let's talk how we can use that data, okay? So, when you use ChatGPT, if you do not pay and most people do not pay for ChatGPT products. Basically brands are promoting inside ChatGPT. And now they opened the self-serve ads manager, which I will show to you that is going to be just prompt to ads, because OpenAI can make you an ad, write a copy, make images, make a videos, right? So, where do they package that Sora model that they discontinued? We will shovel to advertisers' throats, and all the ads in OpenAI you're going to see is basically going to be AI slop. AI that almost no human interface. But allegedly, this is again one person's perspective, no predictions. But what I want to talk about is actually a person behind these ads, and the lawsuit that OpenAI is now facing. Actually kind of two lawsuits. But to start with, you need to know that Asad Awon, this is important context, is basically running this whole ads initiative. >> I do think the vision has to be where almost as easy as you're prompting nowadays with questions, you could say "My goal is sell these shoes more in Midwest." And go. And then it comes back and say "Hey, I tried some experiments, and I think this is the right bid given your price point, this is the right way to do that. Do you want to spend more money on this?" And then you continue that conversation and almost become an agent for that. But today literally a small business has to hire performance marketers, which could be one of the biggest cost in some sense. >> He worked Facebook on Meta. Then he had his own initiative that got acquired by OpenAI, and 3 months later, OpenAI launched ads. You can put one and one together, okay? So, this is the guy behind ads, um bringing the knowledge and expertise from Facebook ads. With that, he might have brought something else, something called Pixel. And if you don't know Pixel, it's like a tracking mechanism that we can trace what you click, where you go from cross-devices. And normally, ChatGPT where is personal informations should not have that Pixel. But, individual, not a company, not somebody like organization is suing OpenAI on behalf of all the US people. Um Amargo Caturla, he discovered OpenAI chats having the Pixel with all the tags from Meta and also Google Analytics. If you go into the lawsuit, he literally shows all the screenshots and exact tags being used and how you're being classified. But, I also would highly recommend uh to read this article from May 18th. OpenAI embedded Meta's Facebook Pixel inside ChatGPT. Okay? Every query you typed was sent to Facebook and Google in real time. So, if you think like, "Oh, no, OpenAI is really bad for doing that." Yeah, no. Perplexity was actually first sued for exactly doing this. And even actually worse, they're using their incognito browsing, which was supposed to be like secure one, also had Pixels in there. So, this is like ChatGPT is just like, "Hey, like these guys can do it. Why we can't do it as well, right?" This basically is illegal. So, your conversation should be private, and that data should not travel over to Meta and Google. You did not agree to that, and it goes against California Invasion of Privacy Act. And he's claiming 5,000 per user. We need to double-check how many users now ChatGPT free has, how many free users per violation. So, basically, per chat. could be more than 100 million. This is interesting because on a flip side, if you think about it, this is a new advertisement platform. So, if you're running a business, if you remember when Facebook ads came out, basically, Facebook ads gave birth to like people like Alex Hormozi who built their whole businesses running Facebook ads and leveraging that. I think this could be that moment for ChatGPT. It will come at the cost of breaking bunch of rules and laws. And businesses like they've been doing that, and what? Uh you know, they're kind of following that Eric Schmidt's rulebook. You just go, break laws, send bunch of lawyers, and by the time that anything happens, you made so much money that you can just hire more lawyers and put people in administration and get everyone silenced. Again, just speculating here, but if it is one of the Facebook advertisement moments, there could be a an opportunity for businesses here. I'm just curious how the perception of population that sees ads inside ChatGPT would feel like. And let me just like drift here for a second because now I have when I research business and I come across their business and their whole website is AI generated, like the low effort, okay? The images are all AI generated and the sentences like it's been 3 years. Like anybody can read the sentence and know like, "Oh, this is Claude." It's not this, it's not that, but it's this. Like, "Oh my god." Like it's like every third sentence like that, right? I, personally, immediately leave that. I'm like, "This is low effort. They This is the people touching AI that don't even know what they're doing. How they represent their business is not sophisticated. I do not trust that their services well." Maybe it's completely wrong, but this is my my kind of perspective, right? I'm curious what's yours. Like if you are going to see ads inside ChatGPT, even though it's helpful, will you click it? Will you have any choice? Will your perception of that company that run ads will be different? Or you don't care. Which is basically give me information, give me faster to go and buy something. I don't know. But if you uh do not pay for ChatGPT and you you see ads, you are the product. Not only that, you are not paying, they actually making money from [snorts] you. Every time you click, they make money. You do not get anything. But let's shift gears because I promised to you I'm not going to bash OpenAI nonstop. Just a little bit. But like we have to talk about something that is actually impressive. Let's open AI. The same company, the same week did completely something else. And now we are shifting gears to the resource to underlying large language models, right? Not a product that someone envisioned and basically trying to make money from, but with technology. And technology can be used in many different ways. So now we're going to talk about their technology, not their vision how to use their technology. May 20th, an OpenAI model has disproved a central conjecture in discrete geometry. Like this is gibberish to you and gibberish to me, but bunch of mathematicians reviewed the work and agreed that this is impressive. So if we trust that they are not paid to say these things and publish the research on this, I hope that we can trust something in this world. They released a paper from OpenAI showing how the model disproved it. By the way, this is unreleased model, okay? This is also a little bit goes into the marketing same as on traffic, right? We have super capable model that it's unreleased and it's doing this. Then research community made their remarks on this conclusion that it is an impressive work. This is rewritten chain of thought of how the model reasons. If you have patience to go through the whole thing, you can. I'm very curious how much it actually costed. This is Dr. Alan. He tracks model processes, advises governments, and his perspectives are very informed, very impressive, and you can read more about him, but he has AGI and ACI trackers, and there is loads of thought that went into that. In no way that he's claiming that ACI was achieved, but we finally, for the first time, he says, got two indicators. So, now we are at two out of 50 that we achieved. I listened to Dr. Alan for many years now, and it's just fantastic that he has framework that we can follow along. So, whatever you call it, ACI or not, honestly, no one even agrees what AGI is, so what we can say about ACI, right? This model did what no human could do in 80 years. It is impressive, and we now we're probably going to wait what is this model that OpenAI has, and what are all the nuances, how autonomously it works, how many tokens, what was the supervision, like all this bunch of questions, but truth is always in some sort of nuance hidden deep below, right? So, this is exactly this contrast that I want to talk about, that we have the same company that is working on technology that can genuinely advance us as population, right? Solving math problems, just understanding ourselves and our environment better, making maybe our lives easier, but at the same time, same company is using you as a product to shovel ads in your in your face, profiling you, doing things illegally. It's us that then who need to make a decision what do we support with our money at this point, as long as money matters, probably. Just few things to think about. We got Pope warning about AI. We have OpenAI getting closer to ACI, but also running and monetizing you. And then there is this, Get [ __ ] Done, GSD. It was very popular tool for Claude code. Get [ __ ] done with this meta prompting layer like a harness sitting on top of a harness that would help you get more out of Claude code. People trusted it. It ran commands on their computers with deep shell access. On May 22nd, the creator Lex Fridman launched a crypto token called GSD on Solana and extracted roughly half a million dollars. Honestly, I don't think if it's worth extracting half a million. Like I think if you are like robbing people and stealing money, like at least go for ability to make a million, right? Make it worth it. Some of AI engineers making half a million. But basically, he posted farewell message. If you try to find him now, all his profiles are deleted. This is not the first time, right? We are seeing multiple instances where NPM packages are being compromised. And this is NPM package is code that you don't really see when you install it and it runs on your computer. But that code is also connected to a registry. So once you install it, it can be altered. And even today, Get [ __ ] Done, like the one by the creator, he if you have official Get [ __ ] Done installed on your computer at any given moment, he has capabilities to push malicious code into your machine and extract information. I will include in the description the exact command that you need to run to check if you have it installed and uninstall as fast as possible. But there are thousands if not million NPM packages. And this pattern where you install something which is still connected to the internet can be altered after you installed. Yeah, this is like a cherry case for cybersecurity and AI security issues. And you know, this is again the same pattern. Using technology, same as OpenAI uses technology to build something that people trust, but then they abuse that trust and basically rug pull us. Um, and you know, they promised us this beautiful future, but then they monetize that future on us. Um, so I don't know, maybe we have a future. So, this creator is the same pattern, right? People trusted it, a lot of people followed this creator, and then basically he rug-pulled and used that trust and technology to make more money. So, I'm just kind of like curious on for myself and for you, how many of AI tools have shell access or just dangerous command access? Do you know who controls those? Who is behind those NPM packages or these AI companies and tools? What is What is their incentive? What type of crypto they are connected to? And do you know what else they are building on the side? So, here is the part where we bring it all together, okay? A pope stands next to AI researcher and basically says this, AI takes characteristics of whoever funds it, builds it, and controls it. That researcher says himself that his own company is operating in certain incentives that can actually conflict between the right thing. And we need moral voices that those incentives could not bend. But who are the moral voices? Who is going to be saying that, "Hey, trust me, I have all the morals in the world." Who is not intertwined with whole web of companies, self-funding, and relationships, and whatever societies behind it? Who would you select as this moral voice? Many people that you might have admired in AI most likely got to that point because certain morals have been compromised. Doesn't mean that they are bad people, but just means that their incentives that they operate in. OpenAI develops generally fascinating technology and then chooses to package it and use it to sell ads to us. And they will say to you, "Oh, this is because we need to fund this. We need to fund to continue building great products." For whom? For surveillance companies? For data brokers? For CIA? In-Q-Tel. We will talk about In-Q-Tel very, very soon. And then we bring it to the personal side, morals, a creator like a person, right? Not organization we trust to build something and then basically goes like, "Bye-bye." So, at least one lesson if a creator has token in crypto and also AI tool, maybe gray area. My question with this series is very simple to you. Who controls the AI that we depend on? Is it you? Is it me? Can you control it yourself? Can you build a protective layer between you and giant organizations? We recently had a discussion in our community and I talked about the Five Eyes and the the whole guidance paper that we released. Um I'm going to talk more about that in next week, but if you do not have like any protective layer, like for us it's Go Bot, it's like a harness on top of cloud code that is personalized to you, like it's not letting any type of noise to to even reach me. But if you do not have layer like that, you have a product like Google will create dynamic layouts and every single person's reality is going to look different. They will build that reality on the context that they have about you and they have a lot, a lot of that context. But you are not in control. You just basically let technology use you instead of taking technology and using it for your own advantage. Other questions like how can you do that, right? There is a lot big component of education. AI is not cheap. It's an expensive toy, right? It's not everybody can easily drop 100 to 100 a month on AI tool. But I think fundamentally this has been always the technology, the same struggle of good and bad. We had books that allowed organizations and religions, I'm looking at you, church, to basically run narratives. Let's call it narratives, right? But also those same books and writing and poetry allowed us, individuals, to organize to spread ideas and philosophies. Same with telephone. They gave us telephone lines so we can listen. Okay, but then we are able to communicate too to organize and spread ideas. Same internet, military created product. Same social media, right? We are able to spread and connect and and and also change maybe a little bit of a narrative, which I would question. I think certain things used to go viral by themselves, but now everything is algorithmic. Same with the AI AI is a new media. Social media was replaced by social media and social media is being replaced by AI AI is a new media. This is what you're going to see. AI is going to decide what is the interface, what is the message, what is the narrative based on your profile and all that data is being used by someone. I mentioned at the beginning that somewhere the shift happened. I think this was when I realized that the way I use technology and AI is 0.0% of at least as builders are present, but most people just use it as like convenience tool without questioning like how can it help you better your life and just accepting every release as oh my god. It's a it's a scary. It's a scary thing, but I think if we wake up more people and put a spark that hey, you can leverage this technology instead of just talking on a phone about whatever and letting, you know, them run ads based on your conversations, actually talk with your community, with your people to arrange, to educate your kids, to educate your older people to avoid scams, to spread information, where is the security risks? How can you leverage this? How do you can certify? How can you upgrade yourself to be that person in your community that will lift people around you. We don't have to be scared of this technology. I think it's just we have to decide are we going to to anything with it ourselves or we just let others do it and basically use us. And what type of future we want? I think we need to talk more about that than about tools individually. So, I hope you like this new format and show. I want to have a discussion both in my community but also with with you guys. So, let me know what you think, what questions this raises to you, and next week we will talk about a little bit more crazy moves and Open AI losing some court actions, some law cases, some big ones. I'm going to try to show to you that and prove to you that AI is a media now and that media companies from the past the six months actually invested in AI and AI is not just media company but it's surveillance company and it's connected to almost 400 surveillance and data brokerages. I'm going to go. This is Autonomy and we are just getting started. If this made you think, please subscribe. If this is interesting and you want to discuss with your community or you think they should know about these things, please do share and I'm very interested to read your comments. Talk to you soon.