Skip to content

Here's a prompt, how could I use an LLM most effectively?

The idea with this prompting strategy is pretty simple:

You show an LLM a prompt which reflects something that your'e trying to achieve using an LLM.

But rather than ask the LLM to do something, you ask it to advise upon how you should use an LLM to do this.

I find this strategy very useful when (for example) there are a few potential ways to achieve the same objective.

Take the below example: I took a fairly basic assistant configuration text, but then used the LLM to take me a step back.

Roughly saying something like:

Let's pause for a moment. Is configuring this as an assistant actually the most logical way to achieve this pretty simple job? What about using a prompt template? Or ... perhaps you can think of a better idea again?

Like many, this prompting strategy works off the assumed premise that LLMs have good foundational/domain knowledge about prompt engineering and LLMs - which isn't always a safe one.

However - if your LLM has a recent training data cuttoff point, I've found they can often bring some creative thinking to the table and help you discover more effective ways of using the tech.

Example Prompt

Here's a draft configuration for an LLM assistant. Do you think the best way to implement this would be:

1) As a custom assistant 2) As a prompt template 3) Using some other method

Provide both your recommendation and reasoning

Model Configuration (Added As Context)

## Foundational instruction

I'm gathering a list of VS Code extensions that have to do with prompt engineering and working with LLMs.

Here's an example of an extension that's interesting to me:

https://marketplace.visualstudio.com/items?itemName=HuggingFace.huggingface-vscode

To speed up the process, I'd like to establish the following workflow:

1: I'll populate a link into the chat, like:

https://marketplace.visualstudio.com/items?itemName=HuggingFace.huggingface-vscode

2: You will respond with exactly the following, enclosed within a markdown codeblock:

# Extension Name (Publisher Name)

(Shields.io install count badge)

Installation code for command pallette

1 line description

## For example

llm-vscode (Hugging Face)

Visual Studio Marketplace Installs

llm-vscode is an extension for all things LLM. It uses llm-ls as its backend.

Install code: ext install HuggingFace.huggingface-vscode ```