Over the past couple months, we discussed the idea of AI agents and how people are going to be building on digital platforms. It is through this that value is increased.
Obviously, this is applicable to InLeo. This is a digital platform that is starting to move into the AI age.
A hot topic of discussion is LeoAI. It is something that is crucial for any platform going forward. I am of the view that, without AI integrated throughout, platforms are going to get left behind.
Of course, mention AI and one of the first things that crops up is data. companies are requiring an ever growing amount of data to meet the needs. Here is where social media platforms could have an advantage. These has the benefit of a continuous stream of new data being added daily. The leaders are Meta, Google, and X.
AI-Summaries: A Data Generating Machine
Since data is the core component of the training of AI models, LeoAI would be nothing without it. Fortunately, more is being added to the database each day.
Before getting into this, we have to embrace the fact that the democratization of data is crucial. Whether we are talking X, OpenAi, or Meta, the common factor is they are company controlled. The data there is not willing shared with anyone else.
Hive is a permissionless blockchain. That means data added to Hive is open to anyone. There is one problem: Hive is text only. Any video or images is not resident on the network. For this reason, we have to increase the amount of text data on Hive.
AI-Summaries achieves this.
Here we have an agent that takes a YouTube video and summarizes it.
Let us look at some of the screenshots:
The first is the thread itself with the video. Notice the comments has 9.
Next we have the summary. When we open up the original thread, we see there are 9 AI generated comments. This is summarizing the video in text form.
As we can see, the agent gets fairly detailed with its summary. All this is being written to the blockchain, expanding the amount of democratized data that Hive offers.
Synthetic Data
Is there value in AI generated data (also known as synthetic data)?
This is a topic that is hotly debated. The short answer is definitely. Where things come under dispute is whether training a model all on synthetic data is a wise move. Some argue that degradation will take place, arising in a number of ways.
Here is how Venice.ai described the dangers:
- Lack of realism: Synthetic data may not accurately represent real-world scenarios, which can lead to models that perform well in controlled environments but struggle in real-world situations.
- Limited diversity: Synthetic data may be generated using the same algorithms or techniques, which can result in a limited diversity of data points and experiences for the model to learn from.
- Insufficient generalization: Models trained on synthetic data may not generalize well to new, unseen data, which can limit their ability to perform well in real-world applications.
- Overfitting to the generator: Models trained on synthetic data can sometimes overfit to the generator, which can result in poor performance on real-world data.
- Difficulty in evaluating model performance: It can be challenging to evaluate the performance of a model trained on synthetic data, as it may not be able to generalize well to real-world scenarios.
- Lack of transparency: Synthetic data can be difficult to understand and interpret, which can make it challenging to diagnose and debug issues with the model.
- Dependence on the quality of the generator: The quality of the synthetic data depends on the quality of the generator, which can be a complex and time-consuming process to develop.
- Limited ability to handle out-of-distribution data: Models trained on synthetic data may struggle to handle data that is significantly different from the training data, which can be a problem in real-world applications.
- Difficulty in transfer learning: Synthetic data may not be easily transferable to other tasks or domains, which can limit the model's ability to generalize.
- Ethical concerns: Training AI models on synthetic data can raise ethical concerns, particularly if the data is generated to manipulate or deceive humans.
In spite of this, OpenAI claims that Strawberry was trained on nothing but synthetic data.
The difference with Hive is we are not talking about a database solely made up of synthetic data. Each day, we have blog post and and short form content that are adding to it. All of that text is captured on Hive.
Millions Of AI Agents
We are going to see millions of AI agents appearing over the next 12 months. Many platforms are already building them in. Here is where we could potentially see exponentials forming.
Agents are going to generate more data. This is what Meta, Microsoft, and Coinbase, 3 companies already in the game, will use to feed their databases. In addition to providing utility, they will transact based upon what they are designed for.
As stated above, this is feeding Big Tech yet does little for smaller companies. Start ups are going to have a problem due to the fact that companies with data are charging a great deal for it. Reddit cut a deal with Google for $60 million to allow Google to use Reddit posts to train its models.
The positive for Reddit is this is a rinse and repeat model. It is likely they cut deals with other companies, such as OpenAI, to provide similar access.
These companies are hungry for data.
Of course, the challenge is start ups most likely cannot shell out this money. This is where a database like Hive can come in. By filling the database, anyone can access the data. It is why I call it the democratization of data.
This particular agent was built by 
We have to focus upon two aspects:
- data generation
- utility
AI agents are going to have to provide both. The latter, utility, is obvious. An agent has to perform some function that people utilize. However, as that happens, more data is generated.
We can bet the ranch this is what ever major platform is doing.