Beyond Fixed Benchmarks and Worst-Case Attacks: Dynamic Boundary Evaluation for Language Models
For LLM evaluators, DBE offers a more informative evaluation method that avoids saturation and adapts to model ability, though it is an incremental improvement over existing adaptive testing approaches.
The authors argue that fixed benchmarks cause ceiling/floor effects in LLM evaluation and propose Dynamic Boundary Evaluation (DBE) to locate each model's decision boundary (pass probability ~0.5). DBE provides a calibrated item bank, a search algorithm (SGBS), and an adaptive protocol, demonstrating broader coverage without saturation across safety, capability, and truthfulness categories.
Evaluating large language models (LLMs) today rests on fixed benchmarks that apply the same set of items to any model, producing ceiling and floor effects that mask capability gaps. We argue that the most informative evaluation signal lies at the boundary, where the per-prompt pass probability is near $0.5$ under random-sampling decoding, and propose Dynamic Boundary Evaluation (DBE), which actively locates each model's boundary and places it on a globally comparable difficulty scale. DBE delivers three artifacts: (i) a calibrated item bank covering safety, capability, and truthfulness, with per-item difficulty labels validated across $9$ reference LLMs; (ii) Skill-Guided Boundary Search (SGBS), a search algorithm that finds boundary items for a given target LLM using only API-level query access; and (iii) an evaluation protocol that places a new LLM on a unified ability scale and grows the evaluation set adaptively when the target falls outside the bank's coverage. We instantiate DBE on four categories spanning safety (harmful request refusal and over-refusal), capability (constrained instruction following), and truthfulness (multi-turn sycophancy resistance). The resulting evaluation covers a broader model spectrum without saturation while remaining compatible with existing datasets.