Rethinking LLM-as-a-Judge: Representation-as-a-Judge with Small Language Models via Semantic Capacity Asymmetry

TL;DR

This paper proposes a new evaluation paradigm using small language models' internal representations, which are more efficient, reliable, and interpretable than traditional prompting-based methods with large models.

Contribution

It introduces the Semantic Capacity Asymmetry Hypothesis and the INSPECTOR framework, shifting evaluation from output generation to internal representation probing.

Findings

INSPECTOR outperforms prompting-based small LMs on reasoning benchmarks.

Small models encode rich evaluative signals in hidden states despite weak generative ability.

Representation-as-a-Judge offers a scalable, interpretable alternative to large LLMs for evaluation.

Abstract

Large language models (LLMs) are widely used as reference-free evaluators via prompting, but this "LLM-as-a-Judge" paradigm is costly, opaque, and sensitive to prompt design. In this work, we investigate whether smaller models can serve as efficient evaluators by leveraging internal representations instead of surface generation. We uncover a consistent empirical pattern: small LMs, despite with weak generative ability, encode rich evaluative signals in their hidden states. This motivates us to propose the Semantic Capacity Asymmetry Hypothesis: evaluation requires significantly less semantic capacity than generation and can be grounded in intermediate representations, suggesting that evaluation does not necessarily need to rely on large-scale generative models but can instead leverage latent features from smaller ones. Our findings motivate a paradigm shift from LLM-as-a-Judge to Representation-as-a-Judge, a decoding-free evaluation strategy that probes internal model structure rather than relying on prompted output. We instantiate this paradigm through INSPECTOR, a probing-based framework that predicts aspect-level evaluation scores from small model representations. Experiments on reasoning benchmarks (GSM8K, MATH, GPQA) show that INSPECTOR substantially outperforms prompting-based small LMs and closely approximates full LLM judges, while offering a more efficient, reliable, and interpretable alternative for scalable evaluation.