Multiprompting

Multiprompting (Multi-task Prompting) is the technique of consolidating multiple, related instructions (e.g., analyze data, then summarize findings) into a single, structured prompt for maximum LLM efficiency.

Multiprompting is a high-efficiency technique: it consolidates a sequence of related tasks into one prompt, significantly reducing total inference time and token usage (Google research confirms this advantage). This approach demands clear delineation (Task 1: Generate Code; Task 2: Write Documentation) to prevent cognitive overload, especially with smaller models. Advanced systems (like GPT-4 and LLaMA-2-Chat-70B) can manage this complexity effectively, sometimes outperforming sequential single-task prompts. Precision is mandatory: prioritize single-task prompts when absolute accuracy is non-negotiable, but leverage multiprompting for a single-cycle, comprehensive output.

https://promptingguide.substack.com/p/single-task-vs-multi-task-prompting