Scedastic AI

Billions of dollars are spent every year developing and training Large Language Models which do amazing things but cannot interact with its environment reliably nor continually learn from it as even the most simplistic animals can. At ScedasticAI, we are developing a system of intelligent agents, based on evolutionary findings, that can learn and gain experience in order to explore, adapt, and shape its environment.

US Patent 12,387,141 Systems and Methods for Probabilistic Representation-Based Machine Learning covers the use of Deep Structured Probabilistic Models to capture non-linear hierarchical relationships and create reusable representations that update incrementally when learning new tasks. This overcomes significant deficiencies in LLMs which fail to adequately detect Out of Distribution signals, suffer from Catastrophic Forgetting, and can’t synthesize solutions in real-time.

 

Scedastic Learning

Continual learning requires the application and reuse of learning achieved from experiences; thus, agents need to interact with their environments to create fundamental inductive representations that they can build upon rather than simply training on mountains of data. Just as a baby giraffe learns to stand and run within an hour of being born, our goal is to create a new approach to building intelligence that can scale and perform tasks by applying learned experiences which enable them to evolve over time with plasticity.

 

Exploration within an environment will generate new data signals, some of which will be unlike anything the agent has ever experienced, but these data anomalies are not discarded like traditional systems, rather they are evaluated to identify traits, recurring patterns, noise, and tease apart new signals which can be categorized as reusable representations.

 

We are leapfrogging over current approaches for Artificial Intelligence and focusing our efforts on cognitive research, experience-based training, and data efficient learning. If you are interested in helping us pursue a new paradigm in General Intelligence, please reach out for more information.