REVIEW 6 cited by
Who Validates the Validators? Aligning LLM-Assisted Evaluation of LLM Outputs with Human Preferences
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Due to the cumbersome nature of human evaluation and limitations of code-based evaluation, Large Language Models (LLMs) are increasingly being used to assist humans in evaluating LLM outputs. Yet LLM-generated evaluators simply inherit all the problems of the LLMs they evaluate, requiring further human validation. We present a mixed-initiative approach to ``validate the validators'' -- aligning LLM-generated evaluation functions (be it prompts or code) with human requirements. Our interface, EvalGen, provides automated assistance to users in generating evaluation criteria and implementing assertions. While generating candidate implementations (Python functions, LLM grader prompts), EvalGen asks humans to grade a subset of LLM outputs; this feedback is used to select implementations that better align with user grades. A qualitative study finds overall support for EvalGen but underscores the subjectivity and iterative process of alignment. In particular, we identify a phenomenon we dub \emph{criteria drift}: users need criteria to grade outputs, but grading outputs helps users define criteria. What is more, some criteria appears \emph{dependent} on the specific LLM outputs observed (rather than independent criteria that can be defined \emph{a priori}), raising serious questions for approaches that assume the independence of evaluation from observation of model outputs. We present our interface and implementation details, a comparison of our algorithm with a baseline approach, and implications for the design of future LLM evaluation assistants.
Forward citations
Cited by 6 Pith papers
-
EvalLoop: A Methodology for Evaluation-Driven Iterative Improvement of Business AI Systems
EvalLoop improves business LLM systems by grouping metrics into dimensions, classifying failure modes, and iterating one system variable at a time, raising a sales briefing model from 82.6% to 94.6%.
-
Rethinking Query Optimization for Multi-Agent Systems [Vision]
Agentic data pipelines are built by hand today; this paper sets a research agenda for automatically optimizing their structure, model choices, and execution engines jointly as a new query-optimization problem.
-
Gensors: Authoring Personalized Visual Sensors with Multimodal Foundation Models and Reasoning
Gensors lets everyday users define personalized visual sensors by decomposing their sensing goal into testable criteria, and a user study shows improved perceived control and understanding over prompt-only authoring.
-
Adaptive Testing for LLM-Based Applications: A Diversity-based Approach
A farthest-first diversity-based selection method for prompt templates finds LLM failures faster than random selection, with compression distance giving the strongest average gains.
-
Personalized Constitutionally-Aligned Agentic Superego: Secure AI Behavior Aligned to Diverse Human Values
An external 'superego' module that filters agentic AI plans against user-selected 'constitutions' plus a universal safety floor is reported to cut harmful outputs by up to 98% on safety benchmarks.
-
Addressing Bias in Generative AI: Challenges and Research Opportunities in Information Management
A position paper synthesizes LLM bias research and proposes a stakeholder-based research agenda for information management, without presenting new empirical results.
Discussion (0). Continue with ORCID to comment.