Fine-Tuned Small LLMs

Description

We report what is, to our knowledge, the largest properly powered expert human evaluation of this task to date — four experienced curriculum specialists, a fully crossed design spanning five open-source model families, three school-age bands, and three weighted educational dimensions — triangulated against two automatic protocols. The dual-rubric result, in particular, is a methodological contribution to how our field evaluates AI-generated educational materials.

Categories

Funders

• Ministry of Education

  Government of the Republic of China

  Taipei

Licence