DataCATT tests how document averaging can hide the training evidence that matters
A controlled attribution benchmark varies redundancy and distractors while retaining known synthetic ground truth.
Official website: Massachusetts Institute of Technology
Model-building research and analysis covering Massachusetts Institute of Technology.

DataA controlled attribution benchmark varies redundancy and distractors while retaining known synthetic ground truth.