Google Research recently released a paper on Patchscopes, a framework aimed at integrating various past techniques to explain the internal mechanisms of large language models. It aims to understand the behavior of LLM and its consistency with human values. This paper was written by Google researchers Asma Ghandeharioun, Avi Caciularu, Adam Pearce, Lucas Dixon, and Mor Geva. Patchscopes utilizes the inherent language skills of LLM to provide intuitive, natural language explanations for its hidden internal representations. It even offers explanations about its internal workings, addressing the limitations of previous explanation methods. Understanding the problem of illusions? Although the initial scope of Patchscopes is limited to the natural language domain and autoregressive Transformer models, its potential applications are broader. Researchers believe that Patchscopes could be used to detect and correct model illusions, explore multimodal expressions (including images and text), and study how models make predictions in complex contexts. Patchscopes can be explained in four steps: The Setup, The Target, The Patch, and The Reveal. A standard prompt is presented to the model, followed by a secondary prompt designed to extract specific hidden information. Inference is performed on the source prompt, injecting the hidden representation into the target prompt. The model processes the enhanced input, revealing its understanding of the context. This paper only scratches the surface of the opportunities created by this framework, and further research is needed to understand its applicability in different domains.