TanAI

In the current state of #AI, we are working on an AR+Spread+Memory-Prediction LLM core, anticipating the potential for further development of LLM models. This model will operate with its own Tokenizer and Attention Mechanism architecture, eliminating context forgetting and hallucination-based information generated within limited context windows, and minimizing GPU usage solely for Attention Mechanisms and Inference. It will be able to semantically map millions of token connections (embedding and semantic) and capture vectorial connections without requiring a GPU (even on a CPU). One of its key features is that users can use a more decentralized and secure AI Model by keeping their own data on their own nodes without sharing it with the model (except for inputs needed for inference). We will be able to see the full architecture and discussion topics when we publish the Whitepaper in the near future.