StressEval converts observed LLM failures into dynamic, controllable test instances for knowledge-intensive reasoning, producing benchmarks like Dynamic OneEval that expose larger performance gaps than static originals.
SAME" - new_gold_answer: MUST equal the original gold answer exactly - clue: MUST be
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
-
StressEval: Failure-Driven Dynamic Benchmarking for Knowledge-Intensive Reasoning in Large Language Models
StressEval converts observed LLM failures into dynamic, controllable test instances for knowledge-intensive reasoning, producing benchmarks like Dynamic OneEval that expose larger performance gaps than static originals.