Skip to content

Process Overview

Understanding the core fundamentals of MAGIST is crucial for designing a custom process pipeline using the powerful features the MAGIST API exposes.

Current System Diagram

This is the exact process running the current MAGIST process pipeline. There are hundreds of features left to implement.

Note: We do NOT claim that this current process is MAGIST's final form OR that this current process is Generally Intelligent. With proper language processors, transformers, MDCs, etc., it has the potential of being the World's first AGI(Artificial General Intelligence)!

Drawing

The part below the divider is going to change in later versions of MAGIST.

Upcoming Features

These features will be implemented after the divider shown above. *

  • ElasticSearch Support for NeuralDB
  • LSTM Gated RNN for Unique Response Synthesis
  • Live Object-Detection and Transfer Learning
  • Other data-types: LiDAR, Motors, etc. *
  • Configurable modules/plug-ins. *

How can this be an AGI?

Human brains, or any animal brain, process information in logical, discrete steps. It takes the data, runs some pre-processing, and then computes an environment. This is our perception of reality. In fact, if humans believe hard enough, it is possible to influence this computation and see a different reality.

With MAGIST, we have taken a similar approach. The data is processed, in turn, slowly extracting meaning from it. This contradicts most modern powerful AI models that use a single, massive network. Some examples include GPT-3, BERT, Imagen, etc. This limits the functionality of the AI, in terms of AGIs, since human-procured data has to be fed to it. This data is sorted and assessed with a fine-toothed comb and removes the stochasticity the environment contains. This also reduces its ability to teach itself new tasks. Despite being multi-purpose, it is functionally limited.

That is why MAGIST uses a multi-agent, self-supervised approach. This will allow it to teach itself new tasks simply by following the outlined pipeline. Although MAGIST is currently limited in the types of data it can process and predict, it could evolve to be incredibly intelligent given enough time, data, and processing power.