Prompt engineering: a deeper dive

What is prompt engineering?

• Prompt are your messages to Toqan
• They instruct Toqan on what to do and what to create
• Prompt engineering is writing your request to produce the best results

Why prompt engineering matters

• Large language models cannot read minds, if you want something, you have to tell them
• If you don’t tell the models, they will generate the most appropriate response they can, without
  having information. This is where gen AI becomes generic AI, instead of generative AI.

How do Large Language Models (LLMs) work?

They learn how to predict the next word. They split entences into individual words, hide some of them, then predict the hidden words. After doing this billions of times, the model has “read” enough Shakespeare to predict how this phrase usually ends.

The anatomy of a prompt

Prompts are made up of two parts: an instruction and context. Here are a few examples below.

Two core principles of prompt engineering

1. Be clear and specific about what response you expect.

For example, instead of saying “Review this email”, you should say, “Review this email. Check it for consistency and clarity. Highlight any spelling errors. Provide a list of suggestions for how to improve the readability.”

2. Ensure Toqan has all the needed information to provide a good answer.

Instead of saying, “Draft a framework to evaluate our user journey”, say, “Draft a framework to evaluate our user journey. Find attached our product description and OKRs.”

A few more examples

Basic: Summarise this PDF.
Better: Summarise this PDF into 3 key takeaways. Focus on actionable learnings for data analysts.

Basic: Generate code.
Better: Generate code that follows our coding principles (attached).

Basic: Help me set up personal development goals
Better: Help me set up personal development goals in the this format. These are the things I want to work on. (Here you’d attach the relevant documents for Toqan to use).

Watch the webinar