Introduces an audit-constrained protocol for targeted LLM reasoning tests using finite component grammar and compares score-based CAPS sampling against uniform sampling under matched budgets, finding no audited-yield improvement.
Evaluating models' local decision boundaries via contrast sets
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Targeted Tests for LLM Reasoning: An Audit-Constrained Protocol
Introduces an audit-constrained protocol for targeted LLM reasoning tests using finite component grammar and compares score-based CAPS sampling against uniform sampling under matched budgets, finding no audited-yield improvement.