This survey critically analyzes the current state of factuality in large language models (LLMs), particularly in the context of chat applications. It highlights the challenges and limitations of LLMs in providing factually accurate responses, which can lead to misinformation. The survey identifies key issues such as the distinction between hallucination and factuality, the evaluation of open-ended text generation, and the improvement of LLM factuality through various methods. It discusses the use of datasets, evaluation metrics, and strategies for enhancing LLM factuality, including pre-training, fine-tuning, and post-processing techniques. The survey also explores the challenges in automatic fact-checking and the potential for multimodal LLMs. Finally, it outlines future research directions, emphasizing the need for more robust and reliable factuality evaluation and improvement methods.This survey critically analyzes the current state of factuality in large language models (LLMs), particularly in the context of chat applications. It highlights the challenges and limitations of LLMs in providing factually accurate responses, which can lead to misinformation. The survey identifies key issues such as the distinction between hallucination and factuality, the evaluation of open-ended text generation, and the improvement of LLM factuality through various methods. It discusses the use of datasets, evaluation metrics, and strategies for enhancing LLM factuality, including pre-training, fine-tuning, and post-processing techniques. The survey also explores the challenges in automatic fact-checking and the potential for multimodal LLMs. Finally, it outlines future research directions, emphasizing the need for more robust and reliable factuality evaluation and improvement methods.