About Monostate
The making of stateful machines
Monostate is a lab working on stateful machines. Our areas of work are entropy, uncertainty, transformer architecture, and control theory.
Our Approach
We're building AGI through a fundamentally different approach, not through parameter scale, but through architectural experimentation. Our work centers on three key concepts:
Control Layer and Theory
Instead of single-path generation, we implement embeded uncertainty quantification where multiple agents vote using mini FNNs. This creates systems that know when they don't know, making possible quantifying uncertainty from the inside out.
Architectural Redesign
Moving beyond attention-only memory to semantic routing and fuzzy neurons, our approach routes entire concepts to experts acitivating the lowest possible amount of layers, achieving dramatic efficiency improvements while maintaining interpretability and feedforward computation.
Stateful Memory
True creativity requires unique knowledge that grows with experience. We're developing runtime architecture modification, beyond RL and backpropagation, but structural continued growth that enables episodic memory and one-shot learning.
"Two things fill the mind with ever new and increasing admiration and awe, the more often and steadily we reflect upon them: the starry heavens above me and the moral law within me." Immanuel Kant, Critique of Practical Reason
Our work aims to create AI systems that help with scientific discovery, efficient usage of energy, and are fundamentally aligned with human values: beings that grow with experience and uniqueness while remaining under human control.
Get in Touch
