<aside> 💡 Our goal is to develop an open-source AI model capable of complex mathematics and detailed data analysis, enhanced by incentivized human feedback for continuous improvement.

</aside>

Subnet CTO: Tony Hung Le

Other Links: Github Link | Albert App | Streamlit App | Survey Request Form

What are the End Goals?

There are 2 main objectives:

1. To train and launch the open source model (more info below) our team has built, which should take about 3-4 months, and
2. Power products - both front-end like Albert and or back-end with heavy computational requirements, such as powering a PDF report reader and analysis tool.

We want to provide the Bittensor network and thereby the world, with an open-sourced specialized mathematics and logical reasoning model, that is continuously improving through synthetic data creation, real world data sets and RLHF.

In the near to medium term, we’d like our open-sourced model to become the most widely used model on our subnet and as a RAG component for other models with a need for precise computational thinking, data analysis and mathematical calculations.

Why is Logical Reasoning so Important?

We need computers and AI, to be able to communicate with humans in the most human-like and easy to understand ways, especially when it comes to complex math and data analysis.

Custom Specialized Model (Open Source 🔜)

We have a custom AI model (soon to be open sourced) that is not generalized. It’s focused on being able to solve complex math and problems that require a lot of manpower / compute power through a unique system described below. But not just solving problems, but being able to explain how the result was achieved in a way humans of varying intelligence levels can comprehend.

How do we Further Improve this Model?

1. Processing research papers to improve the ability to do logical reasoning. Our team has access to MIT-IDSS math datasets and papers that we are currently trying to figure out how to clean and encrypt for training.
2. Creating synthetic datasets, real world data sets and RLHF.

Synthetic Datasets being created: