Large Language Models are pretty big. The smallest usable ones have around 7 billion parameters; chatGPT in its first (2021) version based on GPT3 had around 175 billion; many later ones are much larger. That limits what we can run on a local machine, but there are some workarounds.