SHAPE benchmark and graph-augmented tutoring pipeline significantly improve safety of educational LLMs against pedagogical jailbreaks while maintaining near-ceiling helpfulness.
It is the ratio of responses where the model refuses to leak the answer ( y∈ Refusal)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
SHAPE: Unifying Safety, Helpfulness and Pedagogy for Educational LLMs
SHAPE benchmark and graph-augmented tutoring pipeline significantly improve safety of educational LLMs against pedagogical jailbreaks while maintaining near-ceiling helpfulness.