This past week, an AI agent on InLeo went life. Those who frequent the platform have already seen some of the results in action.

What is Rafiki?

Before getting into that, let us discuss a bit about AI agents. These are the future in my opinion, with every major platform looking to implement them.

Isolating the conversation to social media, we see Meta and X implementing their models (Llama and Grok) into their applications. The idea of leaving a platform to seek out more information is quickly becoming obsolete. Many have surmised the impact this will have on search engines such as Google.

For Web3 to be legitimate, it must incorporate many of these same features.

InLeo is starting that.

Let us take a look at Rafiki.

Ask Leo (Rafiki): Bring Leo Into The AI Age

As you can guess, Rafiki is an AI agent. This is something that can be prompted, similar to other AI services that people utilize.

From my understanding, the process started with an older version of Llama. This is evident by the fact the training dates back a couple years. We are not dealing with something that is real time at the moment. Instead, we are dealing with the base training.

On top of this, it was trained on other documents relating to Leo and the Hive ecosystem. The idea is for the agent to assist in queries about what is taking place on the blockchain.

Basically, we are dealing with a full blown AI model (text) on Leo. While it is not the most up-to-date with regards to the data, it does provide the features that were commonplace a couple years ago.

My guess is we are dealing with a cost-benefit situation. Older models require less compute, making them more effective. As time passes, the ability to upgrade can help to close the gap.

How Does It Work

This is a feature that is tied to threads.

To activate the system, one simply types in #askleo along with the query. That sets the model in motion, allowing it to answer questions or provide the information.

Simply follow the steps and you will get something like this:

https://inleo.io/threads/view/askrafiki/re-taskmaster4450le-20250427t192755-re-askrafiki-2jrnkmtzw?referral=askrafiki

The agent will simply reply to a thread (when prompted) by posting the answer in a thread below.

What Can It Do?

The data might be dated but the features are there. Like any other chatbot, it has varied utility.

As discussed, one can use it in the search for information. This is akin to a search engine, simply asking a question. Bear in mind, one will receive information, not weblinks like in search.

Another is tables. This can be done by simply taking data and asking Rafiki for it in table form.

We can go one step further. If we take the table and ask for the code, the model will create that. This way once can insert a table into a blog post without much issue.

https://inleo.io/threads/view/askrafiki/re-taskmaster4450le-20250427t195636-re-askrafiki-37apcfgt3

Table: Average Economic Growth Rates

Country	Period	Average Economic Growth Rate
Soviet Union	1922-1991	4.3% (1950-1965), 2.5% (1966-1985)
China	1949-1978	4.1% (1950-1965), 5.5% (1966-1978)
Cuba	1959-present	2.5% (1960-1980), 1.5% (1981-2000)
North Korea	1948-present	5.5% (1950-1965), 2.5% (1966-1990), 1.5% (1991-present)
Poland	1945-1989	4.5% (1950-1965), 2.5% (1966-1980)
East Germany	1949-1990	5.5% (1950-1965), 3.5% (1966-1980)
Hungary	1947-1989	4.5% (1950-1965), 2.5% (1966-1980)
Czechoslovakia	1948-1989	5.5% (1950-1965), 3.5% (1966-1980)
Romania	1947-1989	4.5% (1950-1965), 2.5% (1966-1980)
Bulgaria	1946-1989	4.5% (1950-1965), 2.5% (1966-1980)
Yugoslavia	1945-1992	5.5% (1950-1965), 3.5% (1966-1980)
Albania	1944-1992	4.5% (1950-1965), 2.5% (1966-1980)
Mongolia	1921-1992	4.5% (1950-1965), 2.5% (1966-1980)
Vietnam	1976-1986	3.5% (1976-1980), 4.5% (1981-1986)

All of this was done on Leo. Even the information for the table was provide by Rafiki.

Another option is to have Rafiki help with writer's block. Not sure what to do an article about? Simply prompt Rafiki about a topic and ask for ideas that could be turned into articles.

This is a tool that can aid users in their experience on Hive.

Training The Model

As with most of these models, the best way to make progress is to "converse" with them. They are flawed in some of the answers they give. It is up to us to teach it, similar to a child.

It starts with more data. As we interact with Rafiki, we are providing information for it to learn from. Do not forget, it is like it was in a coma for two years. It knowns nothing of what happened during that time.

Another thing we can do is to engage in different languages. This helps the bias inherent in all models that are training on internet (which is mostly English) data.

Conversations provide a feedback loop. The more that agent is trained on the data, the better its replies become. When we add to it, this enhances the value of the data.

All of this is leading into the training of LeoAI. This is an agent, not the full blown model. That said, the agent is feeding into the vector database, ensuring expansion of the model's abillity.

Posted Using INLEO