The Core Flaws of Modern AI based on Large Language Models (longpost)

There are so much noise about AI going on, and I would really like to clear out some confusion created by it. My original idea was to make a series of articles describing the mechanisms of Large Language Models (LLM) and particulary Transformer architecture to create a foundation of understanding of LLM-s for people without PhD in machine learning, because so many LLM-s are being packaged as black boxes while hiding their defficiencies, and barely anyone is taking their time to actually explain the areas where LLM fails, common mistakes it makes, why it’s absolutely not possible to fix them with the current tech, and why for some jobs you absolutely have to have an expert at least doing the final press of “accept solution” button.

However, I realized the amount of details I want to describe is just unacceptable, and most people don’t really care about it anyway. Good news is I can limit the description to mostly properties of the building parts of LLM/transformers rather than show how these parts work in details (the article is still huge though).

The key points of the article:

  1. Transformer is a good translator;
  2. Transformer has no model of world;
  3. As a result, transformer is absolutely terrible at solving truly novel tasks;
  4. Transformer is good at parroting i.e. reproducing similar known solutions;
  5. But even basic math has too many solutions to memorize, so Transformer is unreliable at math;
  6. Transformer is fundamentally unreliable, small fluctiations of input can lead to unpredictable results;
  7. Transformer is good at pretending to be an expert without being an expert;
  8. Chain of thoughts alleviate the problems, but eventually fail the same way.

On IT hiring crysis and AI

My linkedin feed is full of posts like “we stopped hiring junior frontend devs”, “AI replaces juns”, “use AI smart”. I really want to answer, not from position of hirer or a worker, but as a user.

My smartphone app glitches, a website for online ticket sale gives multiple error messages in three (3) different language for the same eror (true story), small store website takes so many resources I almost feel like there is some hidden Figma aoo running inside.
And all Linkedin is arguing about is “We don’t want more junior devs — Yeah, but whom are you gonna hire in ten years”.…

Web Components are bad for you

I used to look at Web Components like “what is this? Is it some modern thing I don’t understand, am I supposed to employ it now?”. As people move away from bloated JS SPA, some of them turn their attention to Web Components.

To not keep the suspense: Web Components are useless in their current design. The original idea of progressive enhancement is hopelessly lost, “you don’t need the JS framework” turned into Lit, Polymer, Stencil effectively creating frameworks on top of Web Components.…

State of decay in self-hosted commenting (Remark42, Artalk, Comentario code review)

I was looking for some convenient solution for simple self-hosted commenting, but instead was welcomed with a mix of personal playgrounds and vibecoding stands which are called “full-stack development” nowadays. I spent several days playing with those 3 commenting systems, studying and modifying their code, so this is not going to be a set of one-paragraph LLM-generated reviews, however, I’m still not deeply familiar with their codebase, so you might call it a “superficial code review”. In order of the encounter, here it goes…

What are microservices? (seriously)

I used to particupate in many discussions about microservices, but over time I began to notice one strange phenomena: when 10 people are talking about microservices every one of them has slightly different meaning of the word. It goes like this:
— Of course we use microservices. Our authorization lives in a separate microservice, game logic and stats live in the second microservices, and our PostgreSQL DB is a third microservice.…

Rust is a disappointment

I used to call myself a Rust hater, but really I was doing it just to compensate for the perceivable fanboyism e.g. Rust tops on stackoverflow surveys as most-loved language. There are many reasons to hate C++, and I hate C++ too. Lots of people were waiting for a better programming language, but got Rust instead.

There are several core problems with Rust:

  1. Its compilation is slow. I mean SLOW. Slower than C++.…

Your teeth were never supposed to rot

Pretty much every person today by the age of 25 year has several teeth with fillings. I actually knew few people who already lost a tooth by that time (i.e. depulped tooth, refused to put a crown, tooth cracked).
It’s astonishing that galley slaves and other malnutritioned people actually had no problem with teeth. Yeah, they did not live long, but what I want to emphasize is that there is nothing forcing you to have cavities in teeth. It is also well known that dental caries epidemic started approximately the same time as sedentary agriculture.