Cassi, named after Cassandra, the oracle of Greek myth, is a research and engineering project seeking to understand the physics of consciousness, such that it can be emulated inside a physics simulation.
The grand vision is to create a reality simulator that is capable of accurately simulating all physical scales through the same rules, creating a physically-accurate simulated world that an emulated living intelligence can reside within, operating on the same physics.
The physical theory does not suggest that success would create a real conscious being, because a CPU or GPU isn't going to create and control the necessary structure in the physical field; however, there would still be the philosophical question of whether or not the distinction between real and simulated truly matters, should sufficient complexity and a degree of self-awareness be achieved.
The project is a prototype and still rapidly evolving. It is disorganized and difficult to understand, due to it being written by generative AI, but LLMs are getting good at the important parts, math and code!
CassiTheory develops the organizational laws that provide the foundation for everything else.
CassiCosmos applies the theory to a realtime, 3D, field-mediated N-body particle gravity simulation built with Godot and Vulkan shaders. It is capable of smoothly simulating globally-effective gravity between millions of particles, all on consumer GPUs. It is developed and tested on a 7900XTX.
CassiFI is where the theory is applied to machine learning. It utilizes the theory's hypothesized mechanics of living intelligence to create a continual learning system stored in a single tensor.
CassiPi is an OhMyPi plugin using the FI to manage memory and context, turning the context window into its own attention system with exact recall. CassiQwen is experiments with using an existing LLM with llama.cpp to act as a selective teacher for the FI by fusing it with the inference process. CassiCraft will use the CassiCosmos engine to simulate the entire Minecraft universe, making Minecraft the ultimate physics sandbox where you can truly do anything. I don't have the resources to develop it further, but the ambition is there 😊
CassiCore contains legacy experiments in classical AI agent memory and orchestration. CassiAI is an archive of early field-based machine learning architectures.
Public release has been greatly accelerated due to serious health issues affecting a member of my family. If you would like to support the project, you can donate to our cat, Jade. She is very grateful to everyone that is helping her recover from kidney failure! ❤️ https://gofund.me/808dbbd66