Self-Evolving Large Language Models: Learning Like a Human
Large language models (LLMs) are revolutionizing AI, but what if they could learn and improve on their own? Self-evolving LLMs are a new concept inspired by human learning. A survey about it is reported in this paper. However, this is just an introduction; more details in the following sections:
This stage focuses on gathering new information and encountering fresh challenges. LLMs can acquire experiences in various ways:
Here, the LLM doesn’t simply store its experiences but actively analyzes them to extract valuable insights. This stage involves:
Internal Adjustments: The LLM directly modifies its internal parameters, similar to how our brains update based on new learnings. This can involve techniques like:
External Memory: The LLM utilizes an external storage system to remember past experiences and insights. This memory can store:
This stage assesses the LLM’s progress to identify how effectively it’s learning and evolving. Evaluation can involve:
