Early learning research
Start small.
Study what stays.
An experiment in learning from a small beginning: a local recurrent network, a visible console, and questions about memory.
Read the field notes ↓Python / MLX / GRU
A question, made observable.
What can a small, initially untrained network retain through experience? Trifallax connects training, recurrent state and saved checkpoints in a local console built for observation.
Begin deliberately
Training starts through an explicit user action. The observer keeps operation supervised.
Learn from experience
The current network trains on generated patterns and deliberately submitted messages.
Examine what remains
Saved training state and bounded recall checks make retention something to investigate.
Research note / Current evidence
What we know.
What remains open.
A small cued recall task has passed, but broader language ability is unproven. Trifallax is an experimental learner, not a conversational assistant or evidence of general intelligence.
More about this project
The name comes from Oxytricha trifallax. The project takes inspiration from rebuilding internal machinery while retaining continuity.
A project by Jaidon Clinton