What a language model is and how it works
Tools such as Claude, ChatGPT or Gemini are built on so-called large language models (LLMs). What a model actually does, why it can write text as well as code and where its natural limits are.
Welcome to Garth.cz. If you are a programmer or you are interested in programming, programming languages and related topics, you will probably find some useful information here.
I work on the site in my spare time and add content gradually, so please be patient with the categories that fill up very slowly. If you would like to share your own findings, tips or experience, feel free to write to me.
Broadly speaking, we talk about all IT topics you can come across in the real or virtual world. If it is worth a click, it is worth a conversation.
Tools such as Claude, ChatGPT or Gemini are built on so-called large language models (LLMs). What a model actually does, why it can write text as well as code and where its natural limits are.
Language models do not work with characters or words but with tokens. The number of tokens determines the price as well as how much text the model can hold at once. What that means in practice and why answers can get worse in a long conversation.
AI sometimes answers confidently and to the point – and yet wrongly. It makes up a nonexistent function, parameter or source. Why this happens and how to catch such errors in time.
AI can write a function, find a bug or explain unfamiliar code within moments. For the result to really save time, it pays to have a workflow – from the prompt through review to tests.
Code from AI often works at first glance. That does not mean it is secure, correct in all cases and a good fit for the project. A checklist of what to go through in generated PHP code before you use it.
Everything you paste into an AI tool goes to the server of the provider. What does not belong in a prompt, how to remove sensitive data from code and what to watch out for with agents that have access to your computer.