a core team of researchers from MIT etc. have built an approach where they let agents move around a dataset and find chunks on their own without using embeddings or vector such, results in 56% higher retrieval quality. Worth looking at I think, esp for high-stakes domains in legal/healthcare/finance etc.
a core team of researchers from MIT etc. have built an approach where they let agents move around a dataset and find chunks on their own without using embeddings or vector such, results in 56% higher retrieval quality. Worth looking at I think, esp for high-stakes domains in legal/healthcare/finance etc.
this doesn't link to any research or results or even claims though?