Guys, chatbots are mostly token generators, they don't run programs e give you responses...it's not a simple shell program, it computes things in GPU and return tokens, in which are translated back to English.
[ citation needed ]
Modern chatbots use tool orchestration where the LLM is given a list of tools it can use. Tools include things like searching the web and, yes, executing code. The LLM can generate arbitrary code (typically Python) and call a tool to execute it and return results. The tool takes the Python input, launches it (in a container in this example) and returns the output, which the LLM can further process.