Meta FAIR Self-Improving Memory Hits 56.0 on Continual Learning Bench
Researchers at Meta FAIR have developed a method for the Kimi-K2.6 model to autonomously generate and refine its own memory management instructions. By distilling these self-authored prompts directly into the model weights, the system achieved a score of 56.0 on the Continual Learning Bench, surpassing the 44.6 score set by traditional hand-written techniques.
Covered by 1 source
- MMarkTechPost↗Michal Sutter2d ago