Welcome to Advanced Techniques in Prompt Engineering. This is a more advanced course on the concept of Prompt Engineering, aiming to give you skills that can enhance your strong foundation in this emerging skill set. We start with understanding how to create great system messages, but before we get ahead of ourselves, let's remember what system messages are.

System prompts are special types of instructions typically hidden from the end-users of LLM systems, especially when the LLM system is structured as a chat-like interface like ChatGPT. While system messages of well-known LLM chat interfaces like ChatGPT are being constantly iterated on, an early version of ChatGPT's system message started as follows:

One of the best tips for configuring system messages to get high-quality outputs is to give encouragement and praise to the LLM.

Here is an example:

It's also been observed that LLMs tend to produce better outputs when you tell them they have a high IQ, as follows:

Implementing system prompts effectively allows you to dictate the tone and style of the AI consistently across all exchanges. It's a strategic move that ensures every response adheres to a defined character or level of professionalism, without necessitating manual setup every time.

Consider how style and tone requirements can be set globally:

In the practices on our platform, you can setup the system prompt in the following way:

Open the right sightbar.

Enter your system prompt in the "Custom Instructions" input field.

That's it! You don't have to click any "save" button, the system prompt is already here and will be applied to your next prompt.

With these insights into system prompts, it's time for you to experiment and practice crafting your own. Remember, the effectiveness of an AI-driven response can often be traced back to the quality of the initial prompt. By applying these tips and tricks to your system prompts, you'll be well-equipped to shape the AI's output to match your desired tone, style, and level of expertise, providing a smoother and more predictable interaction experience.