What is Prompt Engineering for LLMs?

Prompt engineering is the practice of carefully designing the prompts that are inputted into large language models (LLMs) to produce better quality and more controlled outputs. As LLMs become more powerful, prompt engineering is a key technique for utilizing them effectively.

How do prompts work with LLMs?

• Prompts provide context: LLMs generate text statistically based on the prompt, so the prompt sets the context.
• Prompts constrain outputs: Well-designed prompts can constrain the LLM's output to stay on topic and on voice.
• Prompts elicit capabilities: Different prompts can elicit different capabilities from the same LLM by framing the task differently.
• Prompts require iteration: The best prompts are found through trial and error as well as learning prompt engineering techniques.

What techniques are used in prompt engineering?

• Priming: Providing examples and sample outputs primes the LLM to continue in that fashion.
• Plain language: Using natural phrasing helps the LLM understand the prompt.
• Specificity: More details and constraints in the prompt produces more tailored responses.
• Multiple inputs: Different prompts can be combined to guide the LLM.
• Template prompts: Starting with templates and customizing them saves time over writing from scratch.
• Tools: Software is being developed to help design, test, and optimize prompts.

How is prompt engineering advancing LLMs?

• Unlocking potential: Better prompting allows accessing more of an LLM's knowledge and capabilities.
• Control: Prompts enable safer, more reliable output by constraining the LLM.
• Efficiency: Carefully engineered prompts require less compute, fine-tuning, and supervision.
• Automation potential: Advances in prompt engineering may enable end-to-end automation of certain LLM-powered workflows.
• Productization: Prompt engineering is key to developing consumer and enterprise products powered by LLMs.
• Skill development: Prompt engineering is becoming an important AI skill and profession.