REPEATING PROMPTS

Summary from the TLDR AI newsletter

When not using reasoning, repeating the input prompt improves performance for popular models without increasing the number of generated tokens or latency. It is interesting that tricks like this are still possible despite the amount of work being put into improving large language models. The discovery proves how much room for improvement there still is for current models.