Glossary · Term

CORE benchmark

← all terms

Definition

Plain language

A bundle of standard tests used to score how good a language model is overall.

As stated in the literature

An aggregate suite of downstream language-modeling evaluations used alongside Lambada to measure general capability, e.g., in Attractor Model comparisons.

Why it matters: It gives a shared scorecard so different models can be compared on general ability rather than just one narrow skill.

For example, a team can run their new model through this bundle of tests to get a single overall sense of how capable it is.

Heard on the show

“… On downstream evaluations — Lambada and the CORE benchmark suite — the paper reports relative improvements up to nearly twenty percent over a parameter-matched …”
Episode 041 — When the Iteration Teaches the Model to Skip the Iteration

Mentioned in 1 episode

  1. 041
    When the Iteration Teaches the Model to Skip the Iteration

Related terms