LLM

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. ...

So, let me show you a 100x for a change. Biological time is running out, so I have to rush out less polished results, but still they are way above average. What is an intuitive understanding (which can be traced all the way back to the early Upanishads)? It is brain’s way to capture (and “understand”) an observed recurring commonality (a common pattern) without being able to know all the underlying details of how (and exactly why) the pattern emerge. ...

With advent of LLMs one may understand and validate which ideas of the classic theoretical Computer Science are, indeed, working, and what is mostly irrelevant. The aim is to understand how this cognitive illusion works and why. A text of modern Program (the source code) is highly structured and regular. This, of course, goes all the way back to Dijkstra and the discipline of Structured Programming, which among other things, postulates that one should use a small set of semantically well-defined, standard structural blocks or elements (procedures, conditionals, loops) and only these elements, instead of arbitrary mess of gotos and jumps. All this, in turn, is kind of obvious to those who have studied mathematics and logic, but the main point is that one can reason about the code in terms of these structural blocks, and this approach has fundamental cognitive advantages. ...

So, literally everyone is committed with other people’s borrowed money to the slop generators without even looking back or having a second thought. This bubble will definitely pop some day, just because the promises which has already been made to secure shittons of money will never be fulfilled (but something else, some appearance, will be given instead). What one has to do, then? Well, even the current manifestations of coding LLMs can be used to actual, non-bullshit, non-theoretical 10x productivity boosts. (even 100x is theoretically possible if you know what to ask and how to ask, and for that you need to know what kind of training data has been feed into it and how it was lobotomised afterwards by so-called post-training). ...

This shit is simply amazing at so many levels. They literally feed their slop to each other, producing an convincing cognitive illusion of a meaningful real-time collaboration to a normie. Yes, the idea to extend the context (adding new tokens to a context ‘so far’ with a hope of better outcome – a “better” stream of generated tokens) is by any means not a new one. The amazing part is that it basically pollutes the context with barely relevant slop. ...

AUTHOR: lngnmn2@yahoo.com Here is your humanties in 30 minutes, before the morinig run. https://gemini.google.com/share/f2f48f836118 Try to read this slowly.

AUTHOR: lngnmn2@yahoo.com “How often have I said to you that when you have eliminated the impossible, whatever remains, however improbable, must be the truth?” Sherlock Holmes So the “compiler” is there, right on Github [[https://github.com/anthropics/claudes-c-compiler], and the only interesting question is “but how”? Well, maybe we are grossly exaggerating what might be going on under the hood. There is an enormous, almost unbridgeable gap between a formal view and a statistical view of the world. ...

AUTHOR: lngnmn2@yahoo.com There are a few facts to understand: it has been written from scratch, using an academic literature of the compiler’s sub-field, with focus on IR, SSA, guided by some “compiler people” it does not rely on the legacy gcc internal code which no one really understands, it does not rely on llvm/clang (only the literature) it is not optimizing (all the optimizations are missed) so the generated code quality is worse than gcc -O0 so they frightfully followed all the architecture specifications and ABI standards, which is what slop generators are good for. the actual Rust code has to be evaluated yet (the key metrics are modularity and abstraction, clear abstraction barriers) but I predict it will be an imperative spaghetti crap. lets see. ...

Here we formulate (and analyze the implications of) The Fundamental Problem of LLM-assistant vapecoding. The problem is this: There is, at the moment, no machinery to systematically and correctly bridge the gap between the generated slop (code) and the accompanied verbiage, which appear to define the semantics of the code. What appears to be coherent and consistent is only a cognitive illusion, made out of familiar words, which refer to familiar concepts of the mind, which looks plausible [to the mind]. ...