What is an AI Agent
Discover more from 🕵️♂️ ABC's for AI Agents
Agents are a way to use a large language model and have it interact with the outside world through tools.
Source: https://lilianweng.github.io/posts/2023-06-23-agent/
Tools can be anything — a calculator, access to a weather API, access to the internet
The agent has access to these tools and its part of their toolbox. The agent can decide when and how to use these tools given the task that it is given from the user. Tools will have a description, which acts as a way for the agent to understand what the tool does and how to use it.
A simple example
What is the stock price of Microsoft?
The Agent will need to create a reasoning plan on how they’re going to achieve this.
Step 1: Get what the stock ticker of Microsoft is.
So how will it do that? Because this is knowledge that isn’t current or hasn’t changed, it actually already knows this. So it returns back that the symbol is MSFT.
Step 2: Find MSFT stock price
Cool, Now that it has this information, the agent will need to find the current stock price. How will it do that? Well, the model does not have that data on hand and it knows that it doesn’t, so it will look in it’s box of tools, which we’ve given a stock API, the ability to search stock prices given its ticker!
So it will return that data and find out that the stock price is $335.05 (as of June 29, 2023) and return that to a user.
Cool, but why is this different than a google search?
This seems pretty simple as most search engines can do that. But let’s put it in the context of a chat bot, which is what makes building agents unique.
Chat bots traditionally have a “hard coded flow” - filled with a bunch of conditional trees, and you need to “predict” how a user will go through a flow.
Now if you use the LLM as the brain or the core controller, it can decide for you, gather information from the tools, and scale a lot faster.