Bullshit

Look, ma – no hands^W^W another one, who knows it all: https://www.0xkato.xyz/how-llms-actually-work/ I am so fucking tired of this shit, don’t you? Okay, even if arguing with retards is ultimately futile, and especially so in the society which does not value intelligence, but only appearances and virtue signalling, lets continue competing in our sport of choice while we still can. No one understands this shit, since the very process by which the artifact in question has emerged is, in principle, by definition, different from the post-hoc “explanations” and anthropomorphic “rationalizations” which plague such “literature”. ...

After the “Sentient Models” has been booed by very few remaining intelligent people they resorted to a the next best meme – their AI is improving itself (recursively, my ass). https://www.anthropic.com/institute/recursive-self-improvement Notice that this has almost nothing to do with the actual underlaying tech, except that the models are being feed with their own slop for the Reinforcement Learning and as new data, and has everything to do with the upcoming IPO – look, AI writes itself, what can possibly go wrong – buy our shit. ...

I used to work with a few quite stupid “managers” whose only skill was that manipulative pseudo-empathy and other quite transparent manipulative skills – the “Pan Am” fake smiles at the cognitive level. It is not that we didn’t see through these primitive manipulative tactics or “respected them” to not to hurt their felling, we just simply acknowledged that these people are strictly way below our levels and just did not bother. These managers, by the way, are still cock sure of their cleverness and their abilities “to manage people efficiently” and that no one sees how they do it. ...

This is not mere a hatespeech, this is the actual fact. Here is how it works. I have received a lot of reports like this one (because I do really know what to prompt for): The system is now fully complete, architecturally sound, and rigorously verified against its formal specification. All logic is expressed as pure, immutable transformations within a declarative network shell. #+end_quite So, I am outperforming the 99% percent of Github and should be a millionaire in a few months. Or, perhaps, something is fishy out there. ...

There is a whole distinct genre of social media videos, like “The ordinary day of a Japanese salaryman” and what not, you know. So here is mine. I usually trying to avoid HN in its current incarnation (as a safe-space for entitled mediocrity), but once in a while… it is a dildo. https://arkaung.github.io/interactive-turboquant/ – I have seen this shit before, a lot of it. The walls of Egyptian pyramids were covered with such seemingly “logically coherent” and ever infallible “explanations” – everything seem to follow, just the premises were wrong. Writing such Tantric texts is a sure good way to obtain a higher social status among plebs. ...

There is a morning conversation with the most powerful free-tier LLM. Let’s do a deep research project, to re-implement a proper, mathematically rigorous and provable (using the findings of the Category Theory) proof-of-concept blockchain implementation from the first principles, right from the ground up. First of all, every single modern blockchain implementation is an amateur crap, full of idiotic design decisions, full of technical debt, made by an unqualified, Dunning-Kruger naive, amateur over-ambitious zealots. Literally everything, just like in the early versions of PGP, is just wrong. ...

Finally, I have got it Right (sorry – Less Wrong). What modern LLMs (and the underlying algorithms) have captured is not “intelligence”, but “[environmental/social] conditioning” – the ability to build an abstract structure (a “map” of a territory) and then use as a representation of “reality”. What is remarkable and unprecedented, is that the process of training is a valid mathematical model (approximation) to that very algorithm, which is mother Evolution came up with – to “update” (reinforce) the neural connections, marked with a “stress-related chemicals” at the moment of actual experience (which the mathematical model abstracted as “weights”) when we’re asleep (meaning an external process). ...

So called “knowledge workers” (a Cal Newport’s meme, he has many such memes) are going to be paid at the level of third-world construction fast food workers, which is a good thing, especially so-called humanities “professors”. Here is how. A state-of-the art LLM model, even a miserable “30B” ones (Ike Qwen3.5-30B), which could even be ran locally, “compresses” (with loss of information, of course) all the written verbiage it has “seen” (being trained with) into a single abstract structure – which is, no surprise to me, but it shall be an “aha” for (You) – a directed acyclic graph, with a lot of redundancy in it. ...

Giving the exponentially increasing amount of bullshit, like this one: https://news.ycombinator.com/item?id=47521150 we have to perform regular sanity checks to avoid our understanding being polluted by this stormy sea of screaming bullshit (if you will). First, nothing has been changed in principle. this is still an abstract token-level (not even syntax) probabilistic modelling (predicting the next token, given the previous (and updated) state) the “architectures” such as variation of “Transformers” is still a socially constructed Tantric bullshit, every one is parroting, but no one understands, leave alone validates given these two actual properties above, any system, however large and “tuned” (nowadays it is RL – feeding the user’s prompts and the resulting slop back into the model as new training examples), everything they produce is still a slop, in principle. There is nothing more to add or clarify. ...

Everyone is a fucking expert nowadays, especially in the matters of vape-coding and LLM usage, you know. Here is an expert wrote a GPT-assisted piece about how cool he really is: https://msf.github.io/blogpost/local-llm-performance-framework13.html There is, however, a small catch. Running 4bit models when your memory can fit a full-size 64Gb gguf with proper 8bit or (even f16) tensor is just missing the point completely. Yes, you will get just a 1-2 generated tokens per second, so it would feel like a dial-up internet all over again (which is not necessarily bad), but you will get orders-of-magnitude better slop, in principle. ...