Research
- We Accidentally Created a Digital Brain That Never Forgets We taught a spiking network to say “four”. It remembered far better than we expected, and the reason had nothing to do with anything we designed.
- How a Cosine Learning Rate Smooths Out Uneven Accuracy as a Model Trains A short note on something we noticed while training Pathfinder, and why the jumpy accuracy that looked like trouble turned out to be the model behaving exactly the way it should.
- More research soon