Part 7/11:
This process underscores a fundamental issue: current models cannot process all the data at once. Despite their size, models like GPT-3 or GPT-4 still operate through "straws"—processing limited slices of information due to hardware and architecture constraints. The solution, he suggests, involves external knowledge bases—building search-optimized repositories that the AI can query dynamically.
From Raw Text to Knowledge Graphs: Structuring Legal Data
An exciting concept David explores is transforming legal documents into knowledge graphs—interconnected data structures that represent entities (cases, dates, parties) and their relationships. Using formats like JSON-LD, he experiments with converting case law into nodes and edges, enabling efficient querying and visualization.