The field of natural language processing is witnessing significant developments in the area of contextual understanding and attribution in large language models. Researchers are exploring novel methods to quantify and analyze the importance of specific tokens, sentences, and context in generating accurate responses. One notable trend is the integration of token-level attribution methods, such as Shapley value-based approaches, to provide fine-grained insights into the decision-making processes of large language models. Additionally, there is a growing interest in understanding how these models prioritize and integrate contextual and parametric knowledge, with studies highlighting the challenges of contextual grounding and the need for more effective prompt-based methods. Noteworthy papers include: TokenShapley, which proposes a novel token-level attribution method that outperforms state-of-the-art baselines, and Lost-in-the-Later, which introduces a framework for quantifying contextual grounding in large language models and reveals a strong positional bias that affects contextual grounding.