GPT-4: Researchers from KAIST have developed SelFee, a self-feedback and self-revision language model that improves response quality without requiring external models. SelFee is a fine-tuned LLaMA-based instruction-following model that iteratively revises its answers within a single inference. The model was trained using data augmentation techniques and the FastChat framework. Although SelFee demonstrated comparable performance to ChatGPT, it was found to lack knowledge in certain areas. The study emphasizes the importance of iterative revision in enhancing language model responses.
Read more at MarkTechPost…