Trade-offs in Context Representation
A language model is tasked with summarizing a long legal document. It processes the document paragraph by paragraph, updating a single, cumulative memory state after reading each one. To generate the final summary, the model could rely solely on the final memory state after the last paragraph has been processed. Alternatively, it could be designed to access and re-process all the original paragraphs. Analyze the primary trade-off between these two approaches in terms of computational efficiency and potential for information loss.
0
1
Tags
Ch.4 Alignment - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Analysis in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
A language model is designed to understand a multi-chapter novel. Due to processing limitations, it reads the novel one chapter at a time. After reading each chapter, it updates a single, internal 'memory state' that combines the new chapter's information with the memory state from the previous chapters. After the model has processed the final chapter of the novel, which statement best describes the nature of this final memory state?
A language model is designed to maintain a coherent conversation with a user. It processes the conversation turn-by-turn (as segments), updating a single memory state after each turn. After 10 turns of conversation, the model needs to generate a response that considers the entire history. Which of the following is the most direct and effective use of its memory architecture to achieve this?
Trade-offs in Context Representation