aiqbench analysis
This study empirically demonstrates that even flagship-tier LLMs suffer from severe calibration errors in entity extraction, where confidence scores fail to verify truth and are easily manipulated by minor text perturbations. The research reveals that the primary failure mode is often background noise or scope gaps rather than hallucinations, suggesting that confidence tiering should only be used to prioritize human review rather than as a proxy for accuracy. For practitioners deploying LLMs in high-stakes domains like cyber threat intelligence, these findings underscore the necessity of robust post-processing pipelines over reliance on model-provided confidence metrics.
From Unreliable to Usable: Decomposing Failure Modes and Confidence Signals in LLM Entity Extraction - Evidence from Cyber Threat Intelligence
arXiv:2610.20417v1 Announce Type: new Abstract: Audits of large language model (LLM) extraction from full-length documents report severe unreliability, leaving two questions open: what the failures actually consist of, and whether confidence signals can turn unreliable output into usable output. We run a preregistered, fully frozen evaluation of two new-generation low-cost LLMs (DeepSeek-V4-Flash, GLM-5.3-Flash) with a flagship-tier control (GPT-6.1-Sol) on a public corpus of 354 incident campaigns (315 unique reports, 630 model-report evaluations), scoring extraction quality against gold, confidence calibration, and robustness to single-sentence perturbations. Three findings emerge. (1) Unreliability persists across tiers: micro precision is 0.03-0.34 and verbalized calibration error (ECE) 0.53-0.94 across both splits; the gap does not close at the flagship tier. (2) The dominant failure is not hallucination: a full mention census plus a three-tier annotation of 360 sampled candidates (two independent LLM annotators from vendors outside the evaluated set, Cohen's kappa = 0.79 on the preregistered acceptance line, fully disclosed) back-projects to 65.6% background mentions and 19.9% gold-scope gaps, with only 1.2% unsupported candidates. (3) Confidence ranks but does not verify: zero-cost self-consistency-frequency tiering passes a preregistered top-30% precision-gain bar in all six test-split model-type cells (4/6 on train), the zero-cost text-frequency x SC frequency combination wins 3, ties 2, loses 1 across cells, yet relation-level counterfactuals on all 45 targets show that a single misattribution sentence flips high-confidence judgments in 95.6-100% of cases, and even a background sentence explicitly marked as unrelated still binds a fake vulnerability in up to 26.7% of them. We conclude with a deployment boundary: confidence tiering may set review priorities, never truth.
Abstract:Audits of large language model (LLM) extraction from full-length documents report severe unreliability, leaving two questions open: what the failures actually consist of, and whether confidence signals can turn unreliable output into usable output. We run a preregistered, fully frozen evaluation of two new-generation low-cost LLMs (DeepSeek-V4-Flash, GLM-5.3-Flash) with a flagship-tier control (GPT-6.1-Sol) on a public corpus of 354 incident campaigns (315 unique reports, 630 model-report evaluations), scoring extraction quality against gold, confidence calibration, and robustness to single-sentence perturbations. Three findings emerge. (1) Unreliability persists across tiers: micro precision is 0.03-0.34 and verbalized calibration error (ECE) 0.53-0.94 across both splits; the gap does not close at the flagship tier. (2) The dominant failure is not hallucination: a full mention census plus a three-tier annotation of 360 sampled candidates (two independent LLM annotators from vendors outside the evaluated set, Cohen's kappa = 0.79 on the preregistered acceptance line, fully disclosed) back-projects to 65.6% background mentions and 19.9% gold-scope gaps, with only 1.2% unsupported candidates. (3) Confidence ranks but does not verify: zero-cost self-consistency-frequency tiering passes a preregistered top-30% precision-gain bar in all six test-split model-type cells (4/6 on train), the zero-cost text-frequency x SC frequency combination wins 3, ties 2, loses 1 across cells, yet relation-level counterfactuals on all 45 targets show that a single misattribution sentence flips high-confidence judgments in 95.6-100% of cases, and even a background sentence explicitly marked as unrelated still binds a fake vulnerability in up to 26.7% of them. We conclude with a deployment boundary: confidence tiering may set review priorities, never truth.
13 pages: 7-page main text within the 8-page limit, plus Limitations, references, and appendices A-J. Code, protocol, and frozen evaluation data will be made public
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2610.20417 [cs.CL]
(or arXiv:2610.20417v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.20417 arXiv-issued DOI via DataCite (pending registration)
Full text under the publisher's open licence · 304 words
Related coverage
Quantifying Consistency in LLM Logical Reasoning via Structural Uncertainty
arXiv · Jun 17, 2026
Value Generalisation 2: The Missing Hole in AIs’ abilities
AI Alignment Forum · Jul 29, 2026
DI Series: Serving Kimi-K3 MXFP4 Wide-EP Disaggregated on AMD Instinct MI300X and MI325X
AMD · Oct 8, 2026
[AINews] High Taste Testers Ep. 1
Latent Space · Oct 17, 2026