I am mad, annoyed, and just wanna punch someone
This blog post was released on the nineteenth of September 2026.
It's always the same people
First of all, I don't buy the Hugging Face thing. Anyone with a brain understands that it's possible to fully isolate LLM instances. You just need to really want to do your job well and do it well. Mind-blowing, isn't it?
They did that to, I don't know, see what happens or, more likely, attract some attention. Same goes for "MYTHOS WILL KILL US ALL." Neither Fable nor Astra changed the world the way snake oil sellers were trying to make it out to be.
Secondly, I would like to remind you that we have yet to see people being replaced with "AI." For now, people were fired because of overhiring during the COVID era.
You think Apple doesn't use new technologies enough? Well, they don't have their own strong frontier model, but I am sure they are allowed to use it when it helps. Then why were there no layoffs at Apple? Have a look at the headcounts of Meta, Amazon, Microslop, and others before and after 2020—the numbers will tell you everything.
Why are you mad then?
One might say I am chronically online; maybe it's true. More likely, I am just the kind of person who is difficult to trick because I started developing information-hygiene skills around 12 years ago, like many other Ukrainians.
When you see something online, you need to ask the following questions: WHO shows this to you, and WHY do they show it to you? When you keep this in mind and set your emotions aside, it becomes extremely difficult to trick you.
You see, OpenAI, Anthropic, and others who joined them in the idea of "pacing the frontier" are not only whiny losers who cannot lead their companies to profit and/or lead humanity to real, actual improvements in everyday life. They are thieves, thugs, and just belong in uranium mines.
They just want money, money, and money. They don't want to change the world; they are creatures with medieval mindsets: find investors, sell them the idea, be unprofitable for decades, and expect to create a market or make the government create legislation that favors your cartel. Profit.
As I said, it is backwards thinking. Sometimes people like Scamio talk about a revolution and how everything will change and people will lose their jobs, etc. You think they believe that? Well, it's, of course, possible, but I think they are just selling you something. That's why I am mad.
The main danger
OK, so what's the problem? The problem is that they will try to destroy the open models. You see, they get really good, and there is so much room for improving computing with the help of LLMs. It's just an untapped area; I hope to be one of the people who offers a new tool or domain that will help people build something or just change their everyday life.
I was always saying that AI companies cannot be profitable, as LLMs make sense mainly as an on-device thing. Privacy, as well as fine-tuning and post-training, are all possible on-device. Sometimes you don't need more intelligence but a model tuned for YOUR workflow. You cannot do that with cloud models.
Cloud model providers can reroute your prompt to a cheaper model and charge you full price. They can start serving you a quantized model. You think I cannot recognize the output of a quantized model when I use Codex? I see how differently those models perform; I see that something changed, but I wasn't fucking notified.
On the other hand
OK, let's imagine the American frontier slows down. China follows suit... I am joking; they will not. So what's the point of those potential regulations — just to mess with regular people?
Fortunately, agentLIBRE works on itself, and I try to find the parts of, let's say, "average workflows" that require being deterministic. Currently, I work on integrating planning and implementation with the runtime. The runtime works with finite-state machines and is therefore deterministic at its core.
I canceled my Codex subscription. Yes, subsidized tokens, maybe I will give Claude a try, or maybe I will try something else entirely. I also don't want to give Chinese providers access to my computer, as I don't have illusions about what communism is.
Despite not liking frontier AI companies, I acknowledge that they still receive a lot of free money to provide compute to people. It's not like your subscription changes anything, because they get money from private investors and they don't care about profit at the moment. Anyway, we will see.
YouTube
I released a new video. YouTube is so fun and interesting — all this stuff about retention and video metrics. I am definitely not the most experienced person when it comes to video editing, speaking English in a way people want to listen to, or making funny jokes, but I am learning, and I really like it. Check my video here if you haven't seen it.
As I understand it, I need a certain number of videos for YT to know who my audience is. I also need to understand who my audience is and what they like. It's extremely fascinating to me.
What to do?
The most important thing is not to let fearmongers make you feel anxious. The world is changing fast, but it's up to us what direction it takes. The direction I chose is to make sure we have access to these shiny new things that might change our lives in the future or already have done so.
I haven't even mentioned the Jev model and the fact that someone implemented that idea before and their implementation is open. That's what we need; that's what will help us colonize Mars, not text "I agree" under another loser's tweet.