Meta has introduced a new paradigm in language modeling with its Large Concept Model (LCM), which differs significantly from traditional Large Language Models (LLMs). Unlike LLMs that predict text token by token, the LCM predicts entire concepts within a shared embedding space, allowing for more efficient processing of long-context inputs. This approach aims to improve both speed and accuracy in language tasks. Additionally, recent advancements in LLM technology include methods for prompt compression, which enhance the performance of models when dealing with lengthy inputs. Research indicates that these innovations not only reduce computational power requirements but also improve retrieval performance in long-context scenarios. The new models reflect a shift towards more intuitive processing methods, moving away from fixed tokenization and towards a more human-like understanding of language.