analiza aiqbench
This research introduces JEV-as-a-Judge, a decision-only model that significantly reduces evaluation costs by accepting confident verdicts while escalating uncertain ones to stronger judges or human adjudication. The study demonstrates that this frozen cascade approach retains 99% of the accuracy of state-of-the-art LLM judges at a fraction of the inference cost, particularly for standard preference and factuality tasks. However, performance gaps persist in complex scenarios requiring derivation checks or resistance to elaborate incorrect answers, highlighting the need for robust escalation strategies in low-confidence regimes.
JEV-as-a-Judge: Accept When Confident, Escalate When Unsure
arXiv:2609.26550v1 Announce Type: new Abstract: LLM-as-a-judge enables evaluation across diverse tasks, but inference cost and confidence reliability become critical at scale. We study whether a decision-only judge can provide an economical first pass and identify when stronger evaluation is needed. Comparing jev-as-a-judge with sixteen generative and reward-model judges, with blinded human adjudication, we find it within three percentage points of a state-of-the-art LLM judge, our strongest comparator, on ordinary preference and evidence-grounded factuality at 0.36% of the comparator's fee. Larger gaps arise when judgments require checking a derivation or resisting an elaborately written wrong answer. On several benchmarks, JEV's gap to this comparator is concentrated in low-confidence decisions. A frozen cascade that accepts confident verdicts and escalates uncertain ones retains 99% of the comparator's accuracy at lower cost.
Abstract:LLM-as-a-judge enables evaluation across diverse tasks, but inference cost and confidence reliability become critical at scale. We study whether a decision-only judge can provide an economical first pass and identify when stronger evaluation is needed. Comparing jev-as-a-judge with sixteen generative and reward-model judges, with blinded human adjudication, we find it within three percentage points of a state-of-the-art LLM judge, our strongest comparator, on ordinary preference and evidence-grounded factuality at 0.36% of the comparator's fee. Larger gaps arise when judgments require checking a derivation or resisting an elaborately written wrong answer. On several benchmarks, JEV's gap to this comparator is concentrated in low-confidence decisions. A frozen cascade that accepts confident verdicts and escalates uncertain ones retains 99% of the comparator's accuracy at lower cost.
Artificial Intelligence (cs.AI)
arXiv:2609.26550 [cs.AI]
(or arXiv:2609.26550v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.26550 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 217 cuvinte
Subiecte conexe
JEV as a Hallucination Filter: Let the LLM Write, Let JEV Check
Analytics Vidhya · 16 oct. 2026
Article: Agentic Fitness Functions: Extending Evolutionary Architecture Beyond Deterministic Rules
InfoQ · 17 aug. 2026
Frontier AI Risk Report (February–March 2026)
METR · 19 mai 2026
Frontier Risk Report (February to March 2026)
METR · 19 mai 2026