I started to write this blog post on Dec. 30, being triggered by one too many proofs of AI madness. Meanwhile, I found one more video to discuss, so I’m rather glad I didn’t complete it in 2025.

1🤖

My brain is tired of this AI-mania. My soul is very, very tired. We, as a society, were already irreparably fucked-up even before LLMs enraptured us starting with the public release of ChatGPT towards the end of 2022. Three years later, we’re so much like a hamster running in a wheel, that my heads tends to explode. Am I living in a dystopia?

2🤖

GitHub is an endless source of projects. Here’s a random one I stumbled on by chance: llama.cpp. As described on SourceForge, “The llama.cpp project enables the inference of Meta’s LLaMA model (and other models) in pure C/C++ without requiring a Python runtime. It is designed for efficient and fast model execution, offering easy integration for applications needing LLM-based capabilities. The repository focuses on providing a highly optimized and portable implementation for running large language models directly within C/C++ environments.” There is also whisper.cpp, a port of OpenAI’s Whisper model in C/C++. What’s not to like, after all? At least, it’s not Rust! Look what a nice GUI you can have!

But it can be worse. Take Ginto AI: “A PHP/Composer Project — Built with PHP 8.3, managed via Composer, following modern PSR standards.” What for? Here: “A powerful local AI Agent that runs entirely on your machine. Works with OpenAI-compatible APIs and leverages the fastest inference engines on the planet: Groq and Cerebras.”

Is there anybody under the age of 75 who is not trying to do something using AI? Anyone? (I don’t even know what Groq with “q” is, and Cerebras doesn’t ring a bell, either.)

I can understand sheeple using AI chatbots instead of Google or Wikipedia, because I use up to 12 myself. I can understand people using AI-assisted integrated development environments, because this is what the hype stresses on: AI can write and debug code for you. (It mostly creates bugs, though.)

But I feel that my head is exploding. Will life ever go back to the old normal? (Nope, I’m afraid it won’t.)

3🤖

From one Derek Jones (there are many of them in IT):

Unrelenting talk of LLMs now infests all the software ecosystems I frequent.

  • Almost all the papers published (week) daily on the Software Engineering arXiv have an LLM themed title. Way back when I read these LLM papers, they seemed to be more concerned with doing interesting things with LLMs than doing software engineering research.
  • Predictions of the arrival of AGI are shifting further into the future. Which is not difficult given that a few years ago, people were predicting it would arrive within 6-months. Small percentage improvements in benchmark scores are trumpeted by all and sundry.
  • Towards the end of the year, articles explaining AI’s bubble economics, OpenAI’s high rate of loosing money, and the convoluted accounting used to fund some data centers, started appearing. Coding assistants might be great for developer productivity, but for Cursor/Claude/etc to be profitable, a significant cost increase is needed. Will coding assistant companies run out of money to lose before their customers become so dependent on them, that they have no choice but to pay much higher prices?

With predictions of AGI receding into the future, a new grandiose idea is needed to fill the void. Near the end of the year, we got to hear people who must know it’s nonsense claiming that data centers in space would be happening real soon now.

4🤖

Random shit:

5🤖

Here’s the first video that I want to briefly comment on. It’s in Romanian, but it has genuine English closed captions, meaning that there is a transcribed (not auto-generated) Romanian transcript, and a real translation to English. So just enable the CC. 📽️ DACĂ OAMENII AR ȘTI CE VINE… – IF PEOPLE KNEW WHAT WAS COMING…, a talk between an influencer called George Buhnici and Antonio Éram, creator and CEO of Netopia Payments.

Selected portions of the English transcript follow (slighty corrected).

Eram: I decided to take a social break. As a result, I deleted all my accounts. Social media… I stopped giving interviews, I stopped attending events. I felt that given the changes happening right now in society—economically, socially, technologically—I needed to take a step back and try to learn new things, to understand. To understand what is happening as much as possible and, of course, to let others speak or act as well.
I had come to believe that my speech had become very pessimistic, I was repeating many things, and I thought it would be better to step back a bit to recalibrate everything, to understand, to educate myself. And that’s what I’ve been doing lately.
…
After the pandemic. Right after the pandemic. But that wasn’t the trigger.
…
Practically today, let’s say, if I were to envy someone, which I don’t, but if I were to, I would envy all these young people who are now and who have access to things that we couldn’t even dream of at the moment. I mean, we didn’t even… Yeah, we didn’t dream of them. This whole AI thing, being constantly connected, your phone no longer… I mean, I see people endlessly scrolling instead of…
…
I ride the subway and see everyone scrolling. Sure, I also use my phone, you know, but at least I’m watching vlogs, YouTube videos, stuff like that. You know… A lot of people do that, you know.
I would believe those I see on the subway don’t do it. What do they do? TikTok, basically, and Instagram, like… Brain rot. Total brain rot.
But let’s get back to the phone, you know? Right now, this phone is a school. You have the AI, you have an internet connection, you have storage, you have a camera, you have… Everything. Basically everything.
I mean, you can write apps with applications that create mobile apps. I even tried one a few days ago, I made something, okay, I’m not very proud because it’s almost illegal. Specifically, at a stoplight, I started the app and said I wanted it to make an app for me that does this and that, and by the next stop, I had something functional. Okay, within parameters, but how long does it take to have an MVP?
Well, this thing, I don’t know, it used to take 5 programmers, a brief, explaining it, 20 meetings, months of development to get to…
I mean, guys, we’re living in a turning point, you know, and…

A lot of crap follows, with business stories and how great Mexico is, among other things. Many, many things of very little relevance to you, even life philosophy. But let’s get to the AI. Fast-forward to minute 89:30. Of course, this brilliant businessman is completely clueless when it comes to Bitcoin:

Buhnici: Let’s talk a bit about crypto then, okay? About blockchain, about bitcoin. So, are you a bitcoin maximalist?

Eram: Yes, as would say…

Buhnici: Is there only Bitcoin?

Eram: There is no second best. Michael Saylor. So, there is no second best. It is the only finite thing. The finite unit of measure we have besides time.
You see, humanity, from a financial perspective, didn’t have a method to measure a finite quantity. We have it now, mathematically. And it’s a matter of trust in mathematics. This is what many, many people don’t understand. That you have a system that is finite. When you have a finite system, it’s like having a measurement system. When you have a measurement system, you can accomplish things. Human society evolved tremendously once we agreed upon some units of measurement to be globally accepted.

Buhnici: We started from shells.

Eram: We started with shells until someone realized they could gather more shells than needed. And then we moved on to bartering, we moved on to things.
But you realize that maybe you don’t have a cow or goats to buy shoes, and then you’d have to wait until you had a kid to sell, and you’d die of exposure because you didn’t have clothes or shoes or other things.
That’s when we arrived at a system that is valid, universal. Bronze… Pebbles, I don’t know what else, and eventually we settled on gold because gold has a fantastic property: it doesn’t rust, it doesn’t get destroyed, you can’t destroy it. Gold is unique and hard to obtain. Yes, it’s hard to obtain because gold is very cool.
…
But if everyone starts mining now, like during the Gold Rush in California, when everyone went searching for gold, the price of gold dropped. And if you remember, historically, there was a king from Africa who converted to Islam and went on a pilgrimage to Mecca, and along the way, he threw gold everywhere. Basically, for two whole months, they were throwing gold, throwing gold everywhere. Gold became so common that the market crashed, leading to inflation. There was a total recession due to the surplus of gold.
…
The deposit that would suddenly double the amount of available gold. What does it mean doubled? It means the value has been halved.
That’s where Bitcoin comes in. With Bitcoin, you can’t… You can’t increase the number of Bitcoins. I mean, theoretically, you could increase it. In practice, it is impossible. It is… zero. And then you have a unit of measure. Finite. 21 million.

Buhnici: Why is it decreasing these days? When we recorded, it dropped from around 120, went down a bit to around 80.

Eram: It’s dropping because…

Buhnici: …someone sells.

Eram: Exactly.

Buhnici: Who’s selling?

Eram: Here we can discuss market manipulation strategies, but that’s irrelevant what is happening in at the moment.

Buhnici: Because of the zoom out?

Eram: Exactly.

Buhnici: Let’s speculate anyway.

Eram: What?

Buhnici: I would like to speculate. There are theories that China is behind it, with some pretty big wallets, through which they can sell to decrease people’s interest in stable coins.

Eram: No, no.

Buhnici: Based on dollars.

Buhnici is, obviously, even more retarded. BTW, in cryptocurrency and trading, “zoom out” is an investment mantra that means taking a broader, long-term perspective rather than focusing on short-term price fluctuations. “When in doubt, zoom out” is meant to encourage long-term thinking during market turbulence.

Eram: I don’t believe that. I think we are simply very… So, in order for you to increase the value the capitalization of these bitcoins, you need to invest money, right? So… Ok, you never invest 100%, you invest around 20%, it grows, the value of that 20% can double, but we’re getting into some financial stuff that doesn’t…

Buhnici: Let’s explain briefly.

Eram: It doesn’t, but it absolutely doesn’t make sense. What is happening right now is as follows. What is good to know. One. I believe that around 70%, 60-70%, definitely 60, maybe not 70, is retail. So, there are individuals who hold Bitcoin. The rest, from those that exist and can move, the rest are in institutions. They are institutionalized by holders, meaning in vaults, in the books of companies like MicroStrategy, and so on.

Buhnici: BlackRock and others.

Eram: And at BlackRock, they are in ETFs. ETFs are financial instruments. So you can have custodial, like MicroStrategy does, or you can have ETFs. The thing with ETFs, I’ll talk less about MicroStrategy or we can discuss it later if we want, but the thing with ETFs is that they are like bonds. Meaning you can buy and sell them, right? They are financial instruments.
This makes people who own these ETFs demand various things like wanting to sell them, wanting to reduce them, BlackRock and others, right? I know, chip in, ETFs, they can’t say, no, don’t sell me now, fool, because it will grow. They have to execute orders, it’s not like they still have shares, right? ETFs are like a fund or a share, right? It’s just that it’s tied to something. You might have an ETF on multiple shares in a bundle of, well, for those who know, I don’t need to explain it now, and for those who don’t know, they can google it and they’ll understand. Or ask ChatGPT or whatever.
The idea is that what’s happening right now is that a lot of money is coming in and going out. Depending on the need for capitalization in other segments, things happen happens one way or another.

Buhnici: So there are people who sell.

Eram: There are people who sell who say sell, and you have to sell. But the most beautiful part is that when someone sells… Often, not most of the time, but often, someone buys. Now, if you look at the liquidity level in the Bitcoin market and see that liquidity is low, it clearly means someone is buying. At this price. If you look, for example, like three weeks ago when it was high, someone who had been holding Bitcoin for 10 years, bought them at $1000, sold them for $1.7 billion, and sold them, you can guarantee that now they bought them again and marked a profit.
That is, he sold at 100,000 and bought them at 80,000, it’s simple. On the difference, they added another 20% to the portfolio, which was marked as…

Buhnici: Or they were left with a quarter of a billion in pocket money.

Eram: Exactly. This is how it works. Is it market manipulation? In a way, yes. But, you see, things settle down. There will be fluctuations like this. There have been fluctuations before. This drop of 32% or whatever it is, is not the biggest, it’s not the most aggressive, and it certainly won’t be the last. Yes, now we’re talking about going from 120,000 to 80,000 to 25,000. When it drops from 2 million to 1,800,000, we’ll comment the same way, I can tell you right now that we’ll comment exactly the same way.

OK, let me say that this is not market manipulation. This is how markets work. If we’re not happy with that, we need to find a better Marx and a better fix to this situation. Jump to 1:33:30:

Eram: People need to understand that murky times are coming in the financial system. I’m not that Nostradamus but they will happen for that the entire system is built on a foundation which is not stable, and you know very well what that means.
On top of this, there are some movements in the labor market that cannot be quantified and will not be easily quantifiable. We are completely unprepared from an educational, social, economic, and cultural standpoint.
Alarm systems keep going off, but at the same time, we look at people who provide us comfort or give us comfort and think, wow, if that person said it will be fine, then it will be fine. Which is a mistake.
On another note, there are also many who beat the drum and say, will the apocalypse come? The apocalypse will not come, but what will come is a paradigm shift absolutely unique in the history of humanity.
This combination of financial overcrediting along with the declining education and this aspect of Machine Learning and algorithms, which will effectively change how we think and interact, it has already changed how we interact, but how we will think and work, how we will live.
That is, the progress will be logarithmic, not linear. Logarithmic.
And I don’t think we are capable of coping with changes.

Now, logarithmic is better than linear. I guess he was thinking of exponential. Well, he probably didn’t need math to make business.

Then, he misinterprets the social and individual effects of TikTok. His brains might be rotten already.

Buhnici: We as a group.

Eram: We as a group. And here I thought of a theory, which might be stupid but it is what it is, it either gets validated or not, the fact that we are exposed to things that last 10 seconds, like TikTok, is maybe to get us in sync with things that will be very fast. I mean to think very quickly. Or to…
the brain’s capacity to be.

Buhnici: But it doesn’t ask you to think.

Eram: It doesn’t ask you to think, but it asks you to access information quickly. Imagine that, I don’t know, 10 years from now, you’ll start implanting something where you’ll have a brain-machine interface. Meaning you won’t need to interact with a screen anymore, but you’ll…

Buhnici: For a long time, I believed that too. It doesn’t work. From a biological standpoint, we don’t have the capacity to process more information. Do you know how much reaches the cortex? Dragoș [Cârneci, a neuroscientist,] told me. Bytes per second.

Eram: Bytes, but how do you train a muscle? Do you lift weights?

Buhnici: Yes, you need to build it up. And if we speed it up a thousand times, it’s still not enough.

Eram: But it’s a start.

Buhnici: 8 kilobytes per second would be hundreds of times faster than any human.

Eram: It’s true. But what happens if you multiply 8 kilobytes by a million? Of interconnected people. Because, you see, human intelligence isn’t measured by processing speed. It’s not about speed, it’s about the connection. And maybe by creating a brain-machine interface, you’ll be able to interconnect many neurons that will outperform any Nvidia or TPU chip.

Buhnici: And why would we need this?

Eram: Because we will evolve into a future level of the human species.

Buhnici: Whether we like it or not?

Eram: It’s inevitable. Do you really think that the Neanderthal man or they said, hey, we don’t want to leave the cave. We want to stay here. And the others said, hey, we want to walk on two legs instead of four. Let’s go, look, there’s a mountain over there, let’s see what happens.
There were some. In the end, those who went to the mountain or walked on two legs or made the first fire won and so on. It’s inevitable. Evolution is inevitable. Clinging to things that hold us back, how beautiful it was, how good it was, is irrelevant. It is what it is now.
Look now at you or other people, the process of seeking information, no one searches Google anymore. It’s foolishness. It’s a waste of time.

Nobody is opening a book anymore!

A talk on houses, housing, prices and regulation follows. Then, at 01:41:20…

Buhnici: And so this brings me back to the discussion about crypto and artificial intelligence. How do you connect Bitcoin to artificial intelligence and this future? Because I look at the future of our generation and I don’t think we’re all gonna make it.

Eram: Definitely we are not going to make it. That is clear. I mean the generation, what follows is a… Everything will be based on one thing. Energy.

Buhnici: Just energy.

Eram: Bitcoin is energy. AI is an energy consumer. Progress is driven by these two things.
So, we need energy. If I were, I don’t know, in the Romanian government, I would stop everything else. Like, literally everything else. And I would create a strategy where in two years we invest all resources. All, absolutely all resources. We build, I don’t know, five more nuclear power plants, wind farms, transportation infrastructure, hydro…
Everything. Energy. There should be storage, there should be abundance, there should be… There should be abundance.
The country with the highest energy production will be the country that leads the world. Right now, for example, America has a big problem. At the moment, all countries have an energy problem. Except, not by much, but less so for those in China.

Buhnici: Because the country installs more photovoltaics in a year than the rest of the world combined. I’m looking for a news piece, someone, I think it’s you in the group, who posted about that Nvidia conference, Nvidia announced some investments. Yes, I found this. So it says this. “Nvidia will build around 19 AI fabs in Europe,” meaning processing factories for artificial intelligence. And on that map over there, you have you saw it, right? … So “European Union Building 20 AI Factories.” Take a look at that list and tell me which country is missing. So we’re talking about the biggest company on the planet right now, Nvidia, which is building AI processing factories to sell to those who need processing if they don’t build their own. Did you see who shows up below us? Bulgaria. And yet we produce more energy than the Bulgarians, and still, they will have an AI fab and we won’t.

Eram: Yes, the Bulgarians have more nuclear power plants than we do.

Buhnici: And they have a surplus.

Eram: Yes.

Buhnici: And we don’t have a surplus.

Eram: No.

Buhnici: Why?

Eram: Why are you asking me? No I am the minister of energy.

Buhnici: You wanted to say of energy independence?

Eram: No, of energy or whatever it’s called, because they keep renaming it. Right? So what are we talking about? We’re hesitating to build two more reactors at Cernavodă. Two reactors in the context where, I don’t know, it’s the safest thing that exists. The most abundant thing that exists. That exists and produces constantly, right? Constantly. Photovoltaic panels are an abomination if you put them on fields.

Buhnici: On agricultural fields. Yes? And we still do this thing. I’m rabid when I see this.

Eram: Yes, an abomination. You have blocks of flats. Put, damn it, on all the blocks and on all the buildings, put panels and batteries.

Buhnici: On all rooftops.

Eram: Exactly, on all the rooftops. Don’t ask for things and so on. Ok, that’s a discussion.
Nuclear, nuclear, nuclear, and nuclear. This is the only thing. And, obviously, add gas plants, coal plants, do everything, anything, you need to have more, you need to have 1000% energy.

Buhnici: Because it still won’t be enough.

Eram: Exactly.

Buhnici: Why?

Eram: Because until we invent an AI system that is less energy-consuming, no, let me rephrase that, because energy is life. If we want to progress, we need, just like when you grow, when you have a child, you give them food, you give them energy to grow. When you’re an adult, you don’t eat as much as a… If you keep eating, you continue to grow. But you don’t grow taller, you grow wider. You need energy. We are… We are here preparing for the next evolutionary leap. For this, we need energy, and we need a lot of energy.
Look at what happened when we transitioned from ox-drawn carts and horses to coal-powered engines. That was the first technological revolution. Another leap occurred when we moved to gasoline and oil.
We invented this thing. Now we need to prepare for something that is much more planet-friendly and efficient.
The problem with oil is that while it is abundant, we can set up filtration systems, but it’s inefficient because you have to transport it where it’s needed. Whereas with a nuclear setup, you place the plant there and you know it will run continuously for 50 years. You have 50, 20 years, I don’t know exactly how long, but certainly not just two months, to prepare the next fuel supply. That’s why it’s the most efficient.
Forget about fusion, we’ll have that too, probably AI will help us achieve it.

Yes, look at the price of solar. So the price of solar has come down, it’s easier now, it costs you less now to put up a fence with photovoltaic panels than to put up a wooden fence. I’m not honestly trying to sell you cucumbers, you know much better than I do what I’ve been through this.

Buhnici: I built a wooden fence, and the wood cost me more than the panels. If I had used panels only, the architect wouldn’t have allowed it because it would have been ugly.

Eram: Yes, exactly. But this will be irrelevant in the European future.

Buhnici: What many people don’t understand is that even though someone like you comes and says nuclear, energy, everywhere, people look and say, but I have my lights on, I have enough power. What people don’t truly understand about the need for energy? Because MIT recently published a paper that says yes, we can make inference and reduce the cost per token by up to 400 times. By 400 times. But it’s so early that even if we reduce the cost per token by 400 times, we still won’t have enough energy for what’s coming.

Eram: Because many people, due to the educational level, the educational system, do not have the analytical capacity to understand what is being prepared and what is happening now. This thing with AI is not a whim where you say make me a nano-banana and other such things. No. This will radically change the entire production chain.
The strangest part is that for the first time in human history, it will start with those at the top, not those at the bottom.

Buhnici: Meaning?

Eram: Meaning… Until you have a robot plumber or a bricklayer or a painter or an auto mechanic to do things for you, it will take a while. However, a programmer, an accountant, a lawyer, a politician to some extent, someone in finance, these people are replaceable, meaning they can be replaced now.

Buhnici: You know there are people in IT who are switching to manual trades because they already pay better.

Eram: Exactly, exactly, exactly, and you know just as well as I do, or maybe even better, and I’m sure that some of your listeners are even more experienced in this matter, how much a good tradesman costs and how much a bad one costs.

Buhnici: Agreed. I’ve said this publicly before, my handyman earns more than I do.

Eram: Yes. Yes, of course. And there was also a famous joke that said when your toilet gets clogged at 3 AM, your MIT PhD is worthless.

Buhnici: Well, which GPT unclogs…

Eram: So then thinking like okay, stick your finger in, you understand?

Buhnici: Yes, but still, NanoBanana Pro is strong.

Eram: No, “but still”; it is powerful. Gemini 3 is powerful, all the combinations. But look, I’ll give you an example. In terms of privacy, I chose to run models at home. I have my own home lab, with training servers and…

Buhnici: Seriously?

Eram: Yes, precisely because I don’t want to.

Buhnici: What do you run them on?

Eram: On Ollama, with all sorts of things. Chinese, non-Chinese.

Buhnici: Very well. From what I understand, Chinese open-source models are the most used for development. For the fork, here you go.

Eram: Yes. They are. They are good. But, you know what? There are tools you can run for 100,000 euros that cost you nothing. I mean, it costs you electricity.

Buhnici: It only costs you electricity.

Eram: And I have solar panels, so… It’s kind of… It’s not exactly free, but it’s not expensive either.
Speaking of privacy and on that note… What have I been doing all this time? I’ve learned a lot of things.
I have no clue on writing code, I don’t have the slightest ability in that area, I didn’t know anything about Linux, about servers, even though I did technical stuff, I had a basic understanding. But the advantage and help that this tool gives you made me install my own servers, configure them, set up all kinds of experiments, and do them. Yesterday, for example, I was shocked because I did an update on one of the servers, which is the core, and it got stuck. And the entire house is basically connected, everything goes through it.
Panic, go, ask it, it says, you have to go in physically. Well, yeah, I don’t have a monitor, I don’t have a keyboard, I don’t… Go and get one, I went, I bought one, I plugged it in, what commands do I give, what errors, it gives you errors, you don’t know what you’re doing, you write them down in…

Buhnici: But how did you do the debugging? With Comet, with Atlas, with what?

Eram: I also have Comet, yes, I think Comet is really cool.

Buhnici: We’re talking about a browser with AI where you enter the interface and it sees what you’re doing or you can tell it to do things.

Eram: Yes, yes, yes, that thing where you make it do stuff is amazing. I mean, you tell it “hey, do this” and leave me in part.

Buhnici: The idiot goes on and does it.

Eram: Yes. It makes payments for you, it does things.

Buhnici: I found myself with an order placed with the wrong phone number because it didn’t know my number, but it still placed the order.

Eram: Yes, it’s fantastic.

Actually, Comet is an abomination. To trust such agents borders mental deficiency!

Warp is another abomination! This retard is unaware that Warp, the AI terminal, is sending your data in the Cloud because its AI is not local! So this guy runs AI models locally to preserve privacy, but his Linux terminal is an app that sends God knows what to the Cloud! It might even send your passwords!

Eram: And I also use something else, which, right now, my latest toy is, there are actually two of them, but they do the same thing. One is open source and one is paid. For now, I use the paid one more, but I’m trying to switch to, it’s called Warp, I don’t know if you know Warp.

Buhnici: No.

Eram: Warp is a terminal, because in Linux you’re in command, but it’s AI-assisted, you know? And you tell it, hey, install something for me. And it says, okay, what, that one, yes, good. And it writes there, get, GitHub something, clone, I don’t know, like, and you see how it works, you know, and at some point, it asks you, do you want that too, yes, I want it, I don’t have it. You give it a command and it gives you an error, or you start doing something and you get an error and say, help me here, and it says, ok, I’m taking over, and you do that, like, you know. Super cool.

Buhnici: For those watching again, you don’t have a hacker or a programmer in front of you.

Eram: No. I remember that back in the days of Ceaușescu, before ’89, it was trendy with those HC computers and I had Chip. [Both were clones of Sinclair ZX Spectrum.] And I found in a library, I still remember, in a bookstore, a book on how to program in BASIC. It was a green book, I think I still have it. And it was there, you know, let’s make a program that writes something. And I struggled to write, I copied from that book, page by page, line by line, and at the end, you compile and run, you know? Like in compile, run gave me an error. I thought, man, even something from a book that I copied exactly gave me trouble. I put the book aside and decided to stop with programming. If I couldn’t even manage to do what was written in a book… And it was something like “Hello World,” I mean it wasn’t…

Buhnici: I am sure that somewhere a capital I was a small L.

Eram: I’m absolutely sure too.

Buhnici: Or bar. [Does he mean pipe? BASIC had no pipes!]

Eram: Absolutely, yes. Ah, I know what it was. It was to make you a graph, like a Mandelbrot, you know one of those, which is infinite. A fractal. In the end, I figured it out, you remember I figured it out, and I was very happy that I did, and that guy said okay, you know, and it was running, you know, it was a processor, not like the ones on iPhones now… 5 minutes later a pixel appeared, you know, saying, oh, fuck, it’s working, to do all the stuff, it would probably have taken months, you know, and now, look, I left it, I closed it, it was a test.

Buhnici: So what you’re doing now is called vibe coding?

Eram: Vibe coding, yeah, I didn’t know that there is a name, yes.

Buhnici: … Andrej Karpathy also rightly says that the new programming language is English.

Eram: Not necessarily, you know, I don’t agree with that. That means it is perfectly capable of understanding Romanian.

Buhnici: Yes, but it translates from Romanian to English in the background.

Eram: No, it doesn’t translate into English.

Buhnici: You know we don’t have native Romanian LLMs.

Eram: We don’t have native LLMs because we’re a mess.

I’ll skip the issues relative to Romania, to corruption and stuff. Another questionable if not idiot ideas at 02:17:10:

Eram: But do you know what the beautiful part is? And the beautiful part is that AI will clean up this whole thing. The educational system will be revolutionized by the fact that each student will have a personal AI that will teach them based on the feedback given by that child. And that children are different and each learns differently. In his own way, in his own style, at his own pace, in his own time, and so on.
…
The best educational system is one-to-one. Where do you find hundreds of millions of good teachers? You don’t. You make them. How do you make them? Quickly. With AI. Can you replicate 100 million AIs that are personalized for each individual? Yes, because TikTok, Facebook, and others have AI-based algorithms that personalize content one-to-one. So it’s not something invented. We already have this. But instead of putting silly stuff, they’ll put equations, physics, and other things. So we have it. The educational system will be revolutionized.
Well, when this kind of educational revolution happens, and it will happen, you won’t need to remember things, no more need for exams, grades, and tests. The entire system is fundamentally changing. We need to learn to work as a team, to be individuals but also know how to work in a group. To use the tools we need.
Speaking of tools, you would be surprised. If you put five knives in front of someone and ask them to slice a piece of bread, I guarantee that 99% will choose the wrong knife.

He’s wrong. Most people know what a bread knife looks like!

Eram: That’s the most trivial example that comes to mind right now. I’m not even talking about code, programming languages, and so on. It is also true that this will fundamentally change the following as well: Doctors will be different, technicians will be different, judges will be different, and so on.
As I was saying, and we keep coming back to this, we are in a fantastic evolutionary stage. This is the next thing. …

Buhnici: The opportunity, but what is the danger?

Eram: Well, the danger is that everyone who doesn’t want to make this leap will pull things in their own direction, and that’s what’s happening. …
…
I was joking and saying, “That’s it, I’m retiring.” And I did retire. It lasted exactly 30 seconds. You I was retired for 30 seconds. The title of this episode I retired for 30 seconds. Done. And the adventure was over.
I mean, purely and simply no you can stop. You’re not allowed to stop.
Why? Because if you are relevant and have been relevant, you will want to stay relevant. If you’ve created something, you want to keep creating and help others create too.
Why? Because that’s what progress means. And why do you care about progress? Because we have reached where we are thanks to because others have done the same thing again and again and again back in the past.
And I didn’t end up as a child in the middle of the wilderness; I got to where I am by building on the progress and lessons of others. And that’s the goal, to make your own contribution.
If not, you’re a net consumer, which is okay. Because throughout human history, 99% of people have been net consumers, not everyone has progressed, not everyone has been a creator. I chose and like to believe that I’m among the one percent. It gives me more satisfaction. If others find satisfaction in sitting on the beach with a margarita in hand, well, it is what it is.

The rest is really bullshit.

But the worst thing is that such people really believe in Bitcoin, so they now need to have two gods: Bitcoin and AI. And the true apocalypse will happen once we keep increasing the energy production and still not have enough of it!

6🤖

The second video is from NYT’s Hard Fork channel: 📽️ A.I. Scientists Are Here. But Is Progress Accelerating? | EP 170 (less than 38 minutes). And it has an accompanying article with the transcript: Where Is All the A.I.-Driven Scientific Progress? (Dec. 26, 2025)

From the lede:

The leaders of the biggest A.I. labs argue that artificial intelligence will usher in a new era of scientific discovery, which will help us cure diseases and accelerate our ability to address the climate crisis. But what has A.I. actually done for science so far?

To understand, we asked Sam Rodriques, a scientist turned technologist who is developing A.I. tools for scientific research through his nonprofit FutureHouse and a for-profit spinoff, Edison Scientific. Edison recently released Kosmos — an A.I. agent, or A.I. scientist to use the company’s language, that it says can accomplish six months of doctoral or postdoctoral-level research in a single 12-hour run.

Sam walks us through how Kosmos works, and why tools like it could dramatically speed up data analysis. But he also discusses why some of the most audacious claims about A.I. curing disease are unrealistic, as well as what bottlenecks still stand in the way of a true A.I.-accelerated future.

From the talk, starting with a funny (?!) thing:

Roose: Right. Another thing that I found interesting about Kosmos is the cost. This model costs $200 per prompt.

Rodriques: Yeah.

Roose: So every time you give it a task, you’re paying $200. Why is it so expensive?

Rodriques: I mean, it uses a lot of compute. I mean, that’s the fundamental answer is it uses a lot of compute, right?

Roose: Like, give us a sense of how much.

Rodriques: Well, so an individual run from Kosmos will write 42,000 lines of code and read 1,500 research papers on average. Like, if you run Claude, it might write a few hundred lines of code, right? So that gives you some sense. There’s a lot of compute that is going into this.

Newton: Have you ever had like a scientist whose cat walks across the keyboard and accidentally hits enter, and all of a sudden spends like $600?

Rodriques: This is a problem. This is a problem. And we, right, so the thing that you have to understand, right, is that if you are a scientist and you go and do an experiment, you get some data back, you’re going to spend $5,000 or $10,000 gathering that data. And so what scientists want is they want the absolute best performance that they can get. And like scientists who have used Kosmos generally come back to me and are like, they can’t believe we’re only charging $200 for it, right? And, you know, I will say, like, $200 right now is a promotional price. We actually have to eventually charge more.

On a more serious note:

Roose: But a lot of what we hear from the people who are running these big A.I. labs is the possibility that pretty soon A.I. will start making novel scientific discoveries. We’ll start doing things that existing scientific methods and processes can’t do. How close are we to that?

Rodriques: That’s already happening, actually. So if you go and you read the paper that we put out about Kosmos, we put out seven conclusions that it had come to, three of which were replications of existing findings, four of which are net new contributions to the scientific literature, like new discoveries.

Newton: And of those, what’s the most impressive?

Rodriques: So, one of the ones that we really like, the human genome contains millions of genetic variants, right? These are differences between different people’s DNA that are associated with disease. And for the most part, we know that a variant is associated with a disease, but we have no idea why, right? And so we asked Kosmos, we gave it a bunch of raw data about a huge number of different genetic factors, so like what the variants are, what proteins bind near the variants, right, like all these kinds of things, and just asked it for Type 2 diabetes to go and identify a mechanism associated with one of these variants. And it came back and it identified, this was a variant that was not in a gene. And Kosmos identified that this is actually somewhere where a different protein binds. It was able to identify what protein binds and what gene is being expressed and connected that to the actual mechanism of that gene, SSR1, which is involved in the pancreas in secreting insulin, right?

Newton: OK, so in this case, is what I’m hearing that your model was able to do some very fancy reasoning over some existing data and identify something that sort of no other human scientist had gotten around to and might not have for a really long time?

Rodriques: Yeah, that’s right.

Practicalities:

Roose: So one concern that I’ve heard people express about models like Kosmos is that this is just like sort of not where the roadblocks are, that the reason that we don’t have more A.I.-discovered drugs and designed drugs out there curing diseases is not actually because we don’t have the research methods to discover those, it’s because, like, you got to go to trials, and you got to recruit human subjects, and you got to get F.D.A. approval. Like, all that stuff just takes a lot longer than the actual discovery of the drug. So what problems are models like these helping to solve in our scientific process right now?

Rodriques: So, so, absolutely. I actually really agree that like the bottleneck at the end of the day in solving medicine is basically, you know, clinical trials. I mean, and the easiest way to see this is if you look at the number of diseases that we know how to cure in mice, right? It’s astronomical, because obviously you can just run experiments, and in humans, things are just slow. That said, if you think that every experiment that is being run right now by pharma companies, like every clinical trial that’s being run, is like optimally planned and optimally, you know, conceived given the full state of knowledge, you are off your rocker, right? There’s like no way. And those experiments cost hundreds of millions of dollars. And so the question is, like, we do at the end of the day have to run clinical trials. How do we make sure that those experiments are the best experiments we could possibly be running given all the knowledge that we have, given all the data we have? There’s so much data that we have that has insights in it that are waiting to be found, where we just like do not have people to go and find them, and that’s ultimately going to feed into better experiments, better trials, right?

Now, the real fun!

Newton: I’m curious about the reliability piece as you’re running all of these experiments. You know, I saw this going around on social media this week, I reproduced it myself. If you asked Google, “Is 2026 next year?” It said, “No, 2026 is not next year, it is the year after next.” So, in such a world, Sam, some people might get concerned at the idea that we’re now entrusting the A.I. with all of our data analysis. So, how much time are scientists having to spend going back and essentially rechecking the work of the A.I.s and what kind of tax does that place on their work?

Rodriques: Yeah, this is very funny. I mean, look, you have to spend a lot of time going back and checking. But like, to be clear, this is true regardless of whether or not an A.I. does it or whether you ask a friend to do it. If you’re going to publish a paper, you damn well better go back and check it and be sure that you are confident. And it’s never going to be 100 percent, right? The best you’re going to do is you’re going to get to a place where it is similarly good to if you were doing it yourself, which is not 100 percent because you’re not infallible, and checking the work is like always going to be faster than producing it in the first place. By a lot.

Let me show you an example of this class:

Kimi: I’ll help you determine if that URL contains the GXDE OS repository signature and how to properly import it to your Debian 13 system. Let me check the content of that GPG file first.
… (1 pages)
Let me try to access the repository information and check for alternative sources for the GXDE GPG key:
… (24 results)
Now let me check the GXDE repository structure to find the correct GPG key:
… (2 pages)
Perfect! I can see the repository structure. Let me check what the gxde-source package contains and also examine the Release.gpg and InRelease files:
… (2 pages)
Let me download and examine the gxde-source package to understand how it sets up the repository:
… (1 pages)
Let me search for more specific information about GXDE OS repository setup:
… (28 results)
Based on my analysis of the GXDE OS repository structure and the official documentation, I can help you understand how to properly import the GXDE signature for your Debian 13 system.
…

# Download and install the gxde-source package for Debian 13 (lizhi)
wget https://repo.gxde.top/gxde-os/lizhi/g/gxde-source/gxde-source_1.1.10_all.deb
sudo dpkg -i gxde-source_1.1.10_all.deb
sudo apt update

…

Ludditus: Correction: the correct version is gxde-source_1.1.11_all.deb

Kimi: Thank you for the correction! Let me check the updated gxde-source package version 1.1.11.
… (1 pages)
It appears the repository still shows version 1.1.10 as the latest. Let me check if version 1.1.11 might be available in a different location or if it was a typo:
… (12 results)
Based on my search, I can see that the official documentation still references the repository directory rather than a specific version. Let me check if version 1.1.11 might be available by trying different approaches:
… (1 pages)
It appears that version 1.1.11 is not currently available in the repository – the latest version showing is still 1.1.10.

Updated Information
Current Status: The repository at https://repo.gxde.top/gxde-os/lizhi/g/gxde-source/ currently only has gxde-source_1.1.10_all.deb available (as of July 7, 2025).
Version 1.1.11: This version appears to be either:

  • Not yet released to the public repository
  • Available through a different channel
  • A version you have from a different source
    …

Ludditus: I JUST INSTALLED USING VERSION 1.1.11 https://repo.gxde.top/gxde-os/lizhi/g/gxde-source/gxde-source_1.1.11_all.deb

Kimi: Excellent! Thank you for confirming that version 1.1.11 is indeed available and working. It seems the repository listing might not be showing the latest files in the directory index, but the direct URL works perfectly.

Now, why “as of July 7, 2025” when it literally searched and obtained so many results? “We’re not in the current year. I checked, and we’re last year!”

Back to the talk:

Roose: You almost want your like A.I. scientist model to hallucinate a little bit.

Rodriques: Totally. Or just add noise, right? We talk about this is just like adding noise in order to, this is actually important for biological evolution also, right? Like the genome has a lot of noise, and that’s how the evolution randomly comes up with like new stuff, is that there’s like a protein that like just totally random, doesn’t do anything, then one day, all of a sudden, oops, it does something, and that’s great, right?

Roose: What do you make of the leaders of the big A.I. labs, people like Demis and Dario and Sam Altman, who are saying, you know, “A.I. is going to allow us to cure all diseases, or most diseases, within the next decade or two?”

Rodriques: Decade is crazy. Oh, and I’m happy to take a very strong stance on this because if I’m wrong, it’s a great thing, right? But if I’m wrong, everyone wins. But like, decade is crazy.

Roose: Why is it crazy?

Rodriques: Because, for the reason that we were talking about before: You have to run clinical trials. If we had a drug right now that prevented aging, completely halted aging in humans between the ages of like 25 and 65 or something, you would not know for 10 years, because you can’t detect in humans in that age range whether or not they’re aging for like at least like, you know, five or 10 years. Like, you don’t detect from one year to the next that you’re aging. So, you won’t know if the thing is working.

On scientists and AI:

Newton: At this moment in 2025, how much do you think A.I. tools have changed the life of a working scientist, and how different do you expect that will be a year from now?

Rodriques: I think that you’d be shocked to the extent that they have not yet. Scientists in general are extremely conservative people because if you’re running an experiment, you never actually fully know, in biology at least, you usually do not fully understand why the experiment works and why not. There are some things that you’ve inherited from protocols that you’ve run in the past, and where it’s like we do it this way. You could go and test it, but there are way too many things to test. So, you’re just kind of locked in in your methods, and it’s what works, and you just want to do what works. And so for that reason biologists just adopt new methods slowly. I think most labs around the world are still probably doing science the way they’ve done it before, and probably will continue to do so for a while, and that’s OK. One place, I think, with coding, a lot of people are already adopting it, because in biology, historically, coding has been a big bottleneck. It’s a huge unlock now that biologists who didn’t know how to code can do a lot of coding using Claude Code, using OpenAI’s models, Gemini, et cetera. So, that’s a huge unlock. I think that that’s going to see a lot of adoption quickly. Literature search, right? Like being able to parse the immensity of the scientific literature, that’s a huge unlock, that’s going to get adopted very quickly. The tools like what we’re building are like a little bit more frontier. Ultimately, people will adopt them when they see other people using them and getting great results.

So, vibe coding, eh?

Now, hype, or no hype?

Roose: Sam, can we play a little lightning-round game here with you? We’re calling this one Overhyped/Underhyped. So we’ll tell you something, and you tell us whether in your scientific opinion it is overhyped or underhyped. Ready? Vibe proving. This is when A.I. systems go out and like write math proofs.

Rodriques: If I have a forced choice, probably overhyped. It’s great as a progress driver in A.I., and being good at it will probably have implications elsewhere, but is it itself that useful? I’m not sure.

Roose: Robotics for A.I. lab automation?

Rodriques: Robotics for automating A.I. labs or …?

Roose: Yes or for automating scientific labs.

Newton: Like, wet labs.

Rodriques: Robotics for automating scientific labs. Um, I think appropriately hyped. It is going to be totally transformative. The technology is not at all there yet. There’s a lot that we need to do, but like, yeah, probably appropriately hyped.

Newton: AlphaFold 3?

Rodriques: That’s an interesting one. I mean, I think that I would say probably like underhyped in that I think like all of the protein structure models, there’s a lot of hype around them, but they’re going to be extremely transformative. So, I would say probably underhyped. There’s a lot of hype around it, though, so it’s a hard decision to make.

Roose: Virtual cells? We heard from Patrick Collison this summer about what the Arc Institute has done with making a virtual cell.

Rodriques: This is overhyped, but for a specific reason. … I think that calling it a virtual cell is a little bit overhyped, right? Like ultimately, that kind of model models something very specific. Like, actually building a true virtual cell, like being able to simulate a cell in a computer is an amazing goal. We are very far away from that.

Newton: Quantum computing?

Rodriques: Overhyped.

Roose: Brain-computer interfaces?

Rodriques: Oh, man, this one’s really hard. I’m going to say overhyped. I’m a huge believer in B.C.I.s. I think effective B.C.I.s or the way that we imagine them in sci-fi are further out than people imagine. Even Neuralink is making amazing progress —

Meh.

7🤖

I’ve still not decided if this entire thing is depressing, crazy, or genocidal.