LLMs generate one token at a time, so anytime they happen to generate some garbage for any reason, you get "garbage in" for further tokens for as long as it sta

LLMs generate one token at a time, so anytime they happen to generate some garbage for any reason, you get "garbage in" for further tokens for as long as it stays in the context window.