RAG or Learning? Understanding the Limits of LLM Adaptation under Continuous Knowledge Drift in the Real World

Build a benchmark framework that simulates knowledge drift over time and evaluates the performance of RAG systems against continual fine-tuning. Developers can create a standardized dataset that forces models to update their internal world states or retrieval sources.

Suggested repo: driftBench

"Measure how fast your RAG dies when the world changes."

Estimated effort: 40h