Virtual assistant

I love this mural; it roughly translates to “To. be born, Portugal, to die, the world”. It’s on Rua Sao Bento, in Lisbon – a fact which I found impossible to determine by using Google, Google Maps, Bing, or ChatGPT.

How do I know it’s on Rua Sao Bento then? I used a sort of virtual assistant, an “agent”, called BabyAGI. You can try it here: https://babyagi-ui.vercel.app/ with an API to an LLM like OpenAI’s.

Basically BabyAGI (questionable name) automates the combined and focused searching of the web in an iterative process, against a goal that you set for it.

In my case, the goal was, “find the street in Lisbon, Portugal in which the mural «Para nascer, Portugal: para morrer, o mundo.» is located”.

I also gave it a first step: find references to this mural online in English and in Portuguese. In concrete terms, that meant it would perform a Google search, click on result, read results, and pass the results back to the language model for analysis,

As it does all this, it tries to fulfill the goal. If it does, it stops.

If it doesn’t fulfill the goal, and this is the key – it creates its own next step, using AI. And it keeps doing this, over and over again, to infinity (or for many steps as you limit it too).

There’s a hurricane of AI information this year but I feel that this is one of the most important pieces of it. It takes a little practice to use an agent like this, as with Google or ChatGPT. It also takes a little more planning.

But if you can figure out how to delegate parts of your work to it, then maybe you’ll figure out how your products and services solutions can do the same.