OPEN NOTES
Open notes from pr0xyh0rse research: Brightwoven, evals, benchmarks, interpretability, model behaviour, consent-based development, and humane AI critique.
What If Grokking Isn’t Mysterious?
A speculative Brightwoven note on embedding geometry: what changes if meaning is modeled as sheets, filaments, and gradients instead of isolated nodes.
The AI World Model Pilot We Should’ve Built Yesterday
A proposal for consent-based AI development: why transparent pilots, community benefit, and honest data collection produce better training data.