Comment by lucianbr

What's the difference between using GPT to write the prompt to GPT, and "thinking"? The LLM uses the first tokens to predict more tokens, and then uses those tokens to predict even more tokens.