Even with the gold solution in hand, large language models locate the first error step in student math solutions poorly; providing a generated corrected student solution improves accuracy.
Carpenter, Elizabeth Fennema, Maria L
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LLMs cannot spot math errors, even when allowed to peek into the solution
Even with the gold solution in hand, large language models locate the first error step in student math solutions poorly; providing a generated corrected student solution improves accuracy.