LLMs very probably don't work the same way humans do, but they do appear to both plan ahead and translate internal concepts into speech: https://www.anthropic.com/news/tracing-thoughts-language-mod...