Blog

Notes on deciding at volume.

What we learn training decision models and running them on real queues.

October 11, 20263 min read

Your work is not a benchmark

Models are tuned to raise general scores. The only score that predicts how a model will do on your work is a score on your work.