No organism perceives reality directly, every mind works from a filtered subset or perceptual window. A bat navigates by echo timing, a tick responds to butyric acid and warmth and we see a narrow band of light. Wildly different senses, wildly different experiences and yet we all navigate the same causal structure. The cat and I both know that pushing a glass off the table ends with it on the floor.
What you perceive of the world forms your logic. The way your experience over time happens for you dictates the logic by which you navigate the world. “Push causes fall” exists whether you see it or hear it, for you and the cat. The relationship between the objects and gravity forms a logic by which you can predict the world around you. Meaning is this relational structure and if you want to build a mind that generalises, it has to operate here. Not in pixel space, not in token space. In meaning space.
A dual-predictor architecture for grounded memory and concept formation from lived experience.
A model trained on 10,000 novels discovers what text does rather than what it’s about. Explore the results interactively.
Three papers validating temporal co-occurrence as a learning signal across text and genomics.
Investigating photorefractive crystals as inference engines, with a mathematical framework for continual learning.
Eridos AI is a research company founded by Jason Dury in Perth, Australia.