aiqbench analysis
This paper is about LLM-as-judge evaluation, so AI is the primary subject and it passes the topic gate. It best fits AI Research: it runs controlled experiments on seven benchmarks, pre-registers 83 testable predictions, and reports that only 32 met their bar. For practitioners, the useful findings are that the information a judge receives can matter more than which judge is used, and that denying a judge the constraints a system must respect can make iterative revision worse than the starting output. The paper's candid reporting of failed predictions is a useful calibration point for how much current evaluation theory can be trusted.
The Anatomy of a Blind Spot: What an LLM Judge Must Be Shown, and How Much
arXiv:2610.15938v1 Announce Type: new Abstract: Evaluation of text generated by large language models (LLMs) is increasingly delegated to LLMs acting as judges, yet what such a judge must be shown remains unresolved. We treat a judge whose reports are acted on as a regulator, in the sense of cybernetics, and ask what it must be shown, how much, in what form, and what its reports must get right. On seven public benchmarks we varied only what the judge was shown. Across three judges whose prices differ 10- to 17-fold, what the judge was shown mattered more than which judge it was. When a system revised its answers on feedback from a judge denied the material they had to respect, it ended below its starting point on instruction following and question answering. On instruction following the loss grew with each round, and it persisted when the judge saw the user's request but not the constraints and with a second judge family. On news summarisation, a correct reference summary beside the source undid most of the source's contribution, whereas on single-answer questions a reference helped. About a tenth of the source did almost all of its work, and a small entailment graph matched an LLM judge at finding errors in grounded text. We committed 83 testable clauses before running; of the 81 that ran, 32 met their bar and 50 went in the predicted direction, and most quantitative predictions from the theory failed. We report every shortfall and derive design rules tied to measured effects.
Abstract:Evaluation of text generated by large language models (LLMs) is increasingly delegated to LLMs acting as judges, yet what such a judge must be shown remains unresolved. We treat a judge whose reports are acted on as a regulator, in the sense of cybernetics, and ask what it must be shown, how much, in what form, and what its reports must get right. On seven public benchmarks we varied only what the judge was shown. Across three judges whose prices differ 10- to 17-fold, what the judge was shown mattered more than which judge it was. When a system revised its answers on feedback from a judge denied the material they had to respect, it ended below its starting point on instruction following and question answering. On instruction following the loss grew with each round, and it persisted when the judge saw the user's request but not the constraints and with a second judge family. On news summarisation, a correct reference summary beside the source undid most of the source's contribution, whereas on single-answer questions a reference helped. About a tenth of the source did almost all of its work, and a small entailment graph matched an LLM judge at finding errors in grounded text. We committed 83 testable clauses before running; of the 81 that ran, 32 met their bar and 50 went in the predicted direction, and most quantitative predictions from the theory failed. We report every shortfall and derive design rules tied to measured effects.
Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2610.15938 [cs.AI]
(or arXiv:2610.15938v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.15938 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.
Full text under the publisher's open licence · 346 words
Related coverage
A Complete Guide to AI Red-Teaming (With Garak Tutorial)
Analytics Vidhya · Jul 25, 2026
Research Note: Filtering Subversion-Relevant Information From Pretraining Data Is Feasible
LessWrong · Oct 5, 2026
Does preservation make sense before we know how to revive?
LessWrong · Jun 15, 2026
Echoverse: Deep, evolving environments for computer-use agents
Microsoft Research · Jul 30, 2026