Recent developments in the field of AI have highlighted ongoing issues with large language models (LLMs), specifically concerning their tendency to 'hallucinate' or generate factually incorrect responses. Researchers are actively working on methods like conformal abstention to mitigate these errors by improving the factuality of LLMs. This technique is particularly crucial as it helps in avoiding the generation of plausible but incorrect answers, a problem notably present in models like GPT-4, especially with less common queries. Moreover, monitoring strategies such as tracking perplexity and semantic coherence are being employed to detect and reduce such hallucinations, emphasizing the need for rigorous observability, testing, and human oversight in AI systems.