EDM

LLM Reasoning Predicts When Models Are Right: Evidence from Coding Classroom Discourse

Tue Jun 30, 10:45 AM–11:10 AM · North 209

Generative AI & Large Language Models NLP & Discourse Analysis Programming & CS Education Automated Assessment & Scoring

★ Notable speakers

René Kizilcec ★★ — Learning at scale; algorithmic fairness and equity in digital education

Kirk Vanacore ★ — Learning analytics; educational data mining; causal inference in tutoring; AI annotation of learning discourse at scale

Shows that LLM reasoning processes can predict when a model will answer correctly, using evidence from automated coding of classroom discourse in programming courses.

Authors

Bakhtawar Ahtisham, Kirk Vanacore, Zhuqian Zhou, Jinsook Lee, Rene Kizilcec