People need to realize the power they are wielding when it comes to LeoAI.
It is one thing to add to the Hive blockchain. When it comes to LeoAI, however, things go exponential. Here is where individual have an outsized impact.
At the core of this is the knowledge graph that is being constructed. Doing a deep dive into this area reveals some interesting tidbits about how these operate.
Looking at the king of data, Google, this is what we see:
According to Google, its knowledge graph contains over 500 billion facts about 5 billion entities and 3.5 billion relation connections.
In other words, facts are generated about the different entities in the database, forming relationships. Sometimes these entities are called nodes.
So what is a entity?
It is anything that is extracted from the data. This can be a business, person, medical condition, location, or organization.
The Exponential Nature of the LeoAI Database
This is a separate database from Hive. We are referring to a vector database that is being developed to power LeoAI. It incorporates all the data on the Hive blockchain along with whatever else is fed into it.
With the LeoAI database, the most important factor is the relationship connections. Whenever we post anything onto InLeo whether it is thread, article, or comment, the entities are pulled. From here, connections are formed which are then utilized for queries.
Before going any further, we have to mentioned the difference between LeoAI and the LLM. This model is using Llama3 from Meta as the basis. Obviously, Meta set up a vector database along with RAG (Retrieval-Augmented Generation).
The difference comes in the weighting. LEO will take the base model and change the weighting, adjusting the relevance that Meta provides. This is compensated by weighing the data that is provided by Hive.
So how does this all come together? We will use this graph as the basis for the example.
Source
Let us use the entity "Bitcoin". This is something that is commonly mentioned on InLeo. Of course, we cannot overlook all the times over the years that it was discussed since Hive started.
Just off the top of my head, we start with the idea of "crypto" (or cryptocurrency). This will be one set of relationships. Therefore, anything that is tied to crypto will be related with Bitcoin.
Another idea is "blockchain". Here again, anything that is discussed about blockchain will have a relationship with Bitcoin.
Then we have "network".
"Wallet" is another one.
"Coin"
"Money"
"Decentralization"
"ETF"
"Freedom"
"Microstrategy"
"Blackrock"
"Grayscale"
"Satoshi"
These are a few of the entities that are pulled from the Hive Database. If we impose some of those over the graph above, we can see how relationships are formed.
This happens for every entity. The exponential nature forms when we how it expands.
For example, we know Tesla holds Bitcoin. This is an entity that has different relationships. With Tesla we have EV, energy, Elon Musk, Wall Street, AI, and Rivian.
Obviously, there is little overlap.
What happens if Rivian starts to accept Bitcoin? If this is written to the Hive blockchain, we can see how it will easily form relationship. Rivian has relationships with many of the entities of Tesla but would then start to form with some relating to Bitcoin.
Extraction
All of this is vital for the ultimate output of LeoAI. The extraction takes place based upon the relationships. This is based upon not only language but also weighting. Here is where the establishment of the Leo system will differ from what Meta set up.
The relation is based upon contextual co-occurrence and semantic patterns. This is why the system improves as more is fed in.
A model like Leo, due to the different weighting, well have greater accuracy with what it is trained on. This is much smaller than what Meta (and others do) with the LLMs. Hence, feeding LeoAI is crucial for the success.
It also enables a great deal of personalization.
For this, we look at another entity: the Leo user. Each person adding to the database is an entity. This is what allows for the personal recommendation along with the extraction of data based upon one's activity.
Therefore, by simply interacting, we are providing facts which are used to establish relation connections.
We not only are relating other entities, we are an entity.
Whatever is written to the Hive blockchain is entered into the vector database. Here is where pulling data based upon votes, comments, or whatever criteria is desired emerges.
All of this is exponential in nature. As we will see in the future, this is imperative as the Internet moves away from one based upon the webpage to one of AI agents.
The digital platform becomes vital.