What is an AI agent? How does this differ from AI models?

These are important questions regarding Leo. We are seeing more added to the platform, so it is helpful to understand the difference.

LeoAI is something that is discussed a great deal. It is something that is in the process of being developed. We are the ones who feed it the data to train on so this is a community-wide initiative.

In fact, the AI agents can help with this process.

Here is where Leo stands now. In this article, we will go through it to explain how things are shaping up.

AI Models Versus Agents

An AI model is something that is trained. Over time, it learns, incorporating the new data in.

This contrasts with agents, which are designed for a specific purpose. One way to look at it, from the software world, is as middleware.

Basically, the agent is there to perform a specific task. One example I use is to book a hotel room. If the agent is tied to your calendar, and you have to be in Dallas on the 27th-29th, the agent will book a room based upon programmed criteria.

The agent doesn't change. It is the leopard doesnt change its spots. This is what it was designed for.

An AI model is actively learning. It is tied to a vector database, incorporating all new data that is entered. It is embedded and indexed, helping to develop meaning from the data already present.

This is what ChatGPT, Grok, Llama, and other LLMs do.

LeoAI and Agents

LeoAI is a small language model. It is taken from Llama, and then integrates the data from the Hive blockchain. Each time a new post (or thread) is made, it feeds into the vector database. This is used to train the model, similar to how Big Tech does it.

We are in the process of seeing LeoAI developed. The problem, thus far, is we lack the required data. People simply are not producing enough on Hive to properly train the model. This means we could roll out something that is insufficient in its capabilities.

An agent is something suck as Rafiki. This is designed to do certain tasks. Here, it takes input (a thread) and submits that as a prompt to VeniceAI. That model provides output, which Rafiki then posts as a reply thread.

This means that Rafiki is not going to learn. It is dependent upon the output of VeniceAI. Improvement comes from the expansion of that model.

What Rafiki does is to enlarge the vector database. This is what is feeding LeoAI, being used to train the model. Hence, Rafiki is a component of LeoAI, albeit not a model.