A developer has released Living Weights, an engineering approach that enables open-source models to learn and update their own weights during use rather than remaining frozen. The technique allows models to retain learned information across sessions and machines without requiring cached context or external files, potentially enabling recursive self-improvement for local AI systems.