🤖 AI Summary
This work addresses the challenge of uncertainty estimation for large language models in black-box API settings, where existing methods are either computationally expensive or rely on unreliable single-pass signals. The authors propose SELFDOUBT, a novel framework that extracts the “Hesitation-to-Verification Ratio” (HVR) from a single natural language reasoning trajectory as a behavioral signal of uncertainty. This approach enables high-precision confidence gating at zero additional computational cost, without requiring multiple generations or internal model access. Evaluated across seven models and three benchmarks, SELFDOUBT demonstrates strong performance: trajectories without hesitation achieve 96% accuracy, while the full SELFDOUBT score outperforms semantic entropy at one-tenth the inference cost. In cascaded deployment, it achieves 90% accuracy with 71% coverage, showcasing its practical efficacy for real-world black-box APIs.
📝 Abstract
Uncertainty estimation for reasoning language models remains difficult to deploy in practice: sampling-based methods are computationally expensive, while common single-pass proxies such as verbalized confidence or trace length are often inconsistent across models. This problem is compounded for proprietary reasoning APIs that expose neither logits nor intermediate token probabilities, leaving practitioners with no reliable uncertainty signal at inference time. We propose SELFDOUBT, a single-pass uncertainty framework that resolves this impasse by extracting behavioral signals directly from the reasoning trace itself. Our key signal, the Hedge-to-Verify Ratio (HVR), detects whether a reasoning trace contains uncertainty markers and, if so, whether they are offset by explicit selfchecking behavior. Unlike methods that require multiple sampled traces or model internals, SELFDOUBT operates on a single observed reasoning trajectory, making it suitable for latency- and cost-constrained deployment over any proprietary API. We evaluate SELFDOUBT across seven models and three multi-step reasoning benchmarks (BBH, GPQA-Diamond, and MMLU-Pro). Most notably, traces containing no hedging markers are correct 96% of the time, revealing an emergent high-precision confidence gate at zero additional cost. For the remaining cases, the full SELFDOUBT score significantly outperforms sampling-based semantic entropy at 10x lower inference cost. A deployment cascade combining both stages attains 90% accuracy at 71% coverage without any task-specific labels. These results establish SELFDOUBT as a scalable, production-ready foundation for uncertainty estimation over proprietary reasoning models.