Let's do everything

Automating tasks with LLMs

Author

Nejc Susec
March 19, 2024

Hello Friend

My calendar has been filling up with tasks recently. To the point where I was pushing the same tasks from one week to the next, to the next.

I’m not one to leave things undone, so it got me thinking:

Why don’t I automate some of these?

I have just the thing for it after all. 😁

As I will be creating a lot of very personalized AI assistants, I chose to do it locally.

This will allow me to control the parameters, access them at any time and experiment as much as I want.

I went with LM studio. An interface where we can download and use open-source Large Language Models.

The interface is very simple and guides us through everything. Has explanations and instructions on every step.

In the end I chose Mistral as my LLM. The clear limitation of running LLMs locally is the available models - no ChatGTP or Claude.

With my model chosen, I started creating core system prompts.

Here are a two examples:

I ended up setting up around 8 different assistants, testing out different prompts and sometimes keeping two versions for the same assistant.

Big benefit of running this locally, I can adjust the parameters for each individual assistant.

It might seem unnecessary, but when it comes to automation of specific tasks, the more I can limit and guide the AI to do what I want it to do, the way I want it to do, the smoother it will work.

One person leveraging the power of AI can turn into a high-performing business, a production house, a multichannel marketing company…

The key point here is understanding the areas we are prompting our assistants for. The better we do, the higher quality output they will have.

I’ve started a group where I share prompts like these daily. My goal is to provide as much valuable content as possible to those who want to move forward quickly and stay ahead.

The group is free and I would be most delighted if you join the conversations. 😁

Keep creating and always remember to have fun. ☀️