#LLM Self-Improvement

Prevent Loss of Talents During Self-development of Large Language Models

The researchers developed a solution to prevent LLM from losing previous gains while improving itself with the self-distillation method.