Showing 1–6 of 6 · #methodology

Nov 06BENCHMARKSBuild an LLM eval set in an afternoon6 min ↗Nov 05BENCHMARKSBenchmark contamination: how to check whether a model has seen the test6 min ↗Oct 24BENCHMARKSModel graders are useful, if you calibrate them first6 min ↗Oct 01BENCHMARKSYour 2-point improvement might be noise: confidence intervals for LLM evals6 min ↗Aug 26MODELSHow to read a model card critically6 min ↗Aug 05BENCHMARKSWhen a benchmark saturates, what should replace it?6 min ↗