Tag
A framework-free prototype learner called Jayce enables local LLMs to learn and correct facts instantly via Adaptive Prototype Memory, achieving 1.6x to 4x faster training updates than standard backpropagation in benchmarks.