A deep dive into the methods and takeaways I realized alongside the way in which

Final month, I had the unimaginable honor of profitable Singapore’s first ever GPT-4 Immediate Engineering competitors, which introduced collectively over 400 prompt-ly good members, organised by the Authorities Know-how Company of Singapore (GovTech).

Immediate engineering is a self-discipline that blends each artwork and science — it’s as a lot technical understanding as it’s of creativity and strategic considering. This text is a compilation of the immediate engineering methods and insights that I realized alongside the way in which, that push any LLM to do precisely what you want and extra!

This text covers the next, with 🟢 referring to beginner-friendly prompting strategies whereas 🟠 refers to superior methods:

1. [🟢] Structuring prompts utilizing the CO-STAR framework

2. [🟢] Sectioning prompts utilizing delimiters

3. [🟠] Utilizing system prompts with LLM guardrails

4. [🟠] Analyzing datasets utilizing solely LLMs, with out plugins or code

Efficient immediate structuring is essential for eliciting optimum responses from an LLM. The CO-STAR framework, a brainchild of GovTech Singapore’s Information Science & AI staff, is a useful template for structuring prompts. It considers all the important thing features that affect the effectiveness and relevance of an LLM’s response, resulting in extra optimum responses.

Right here’s the way it works:

(C) Context: Present background info on the duty.

This helps the LLM perceive the precise situation being mentioned, guaranteeing its response is related and aligned along with your expectations.

(O) Goal: Outline what the duty is that you really want the LLM to carry out.