Stop Chunking Like It's 2022 — Yuval Belfer, AI21 Labs

Introduced a multiscale indexing strategy for RAG systems where the corpus is indexed at multiple chunk sizes. Implemented reciprocal rank fusion to merge results from different scales at query time, enabling dynamic selection of the optimal chunk size per query. This approach significantly improves recall (20-40% over fixed chunk sizes) at the cost of increased memory footprint (2-5x) but with negligible latency impact.