Model Editor

Model editing focuses on correcting inaccuracies or outdated information in large language models (LLMs) and vision-language models (VLLMs) without the need for complete retraining. Current research explores various methods, including fine-tuning with modifications to optimize conditional likelihood, and novel approaches like using discrete key-value adaptors or memory-based systems that store edits in an explicit memory. These advancements aim to improve the accuracy and reliability of large models, addressing issues like hallucinations and knowledge decay, and are crucial for maintaining the performance and trustworthiness of these increasingly prevalent AI systems.