LLM training data mixture optimization breaks when training pools shift — every prior proxy experiment becomes stale.
Andrews, Isaiah, Nano Barahona, Matthew Gentzkow, Ashesh Rambachan, and Jesse M. Shapiro. "Causal Interpretation of Structural IV Estimands." NBER Working Paper ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results