The field of large language models (LLMs) is moving towards a deeper understanding of hallucinations, which are incorrect or nonsensical responses generated by these models. Researchers are developing new methods to detect and quantify hallucinations, including principled detection approaches and spectral-graph frameworks. These innovations are crucial for advancing the field and making LLMs more trustworthy, particularly in high-stakes domains such as medicine and finance. Noteworthy papers in this area include LLMs Learn Constructions That Humans Do Not Know, which highlights the issue of confirmation bias in construction probing methods, and Grounding the Ungrounded, which introduces a rigorous information geometric framework for quantifying hallucinations in multimodal LLMs. Principled Detection of Hallucinations in Large Language Models via Multiple Testing also presents a robust approach to detecting hallucinations, and An Investigation on Group Query Hallucination Attacks demonstrates the potential risks of accumulated context in LLMs.