TRACE: Trajectory Correction from Cross-layer Evidence for Hallucination Reduction

TL;DR

TRACE is a training-free, inference-time algorithm that improves hallucination correction in large language models by leveraging cross-layer evidence, significantly enhancing factual accuracy across multiple models and benchmarks.

Contribution

The paper introduces TRACE, a novel inference-time method that dynamically corrects hallucinations using internal model evidence without additional training or external data.

Findings

TRACE improves factual accuracy across 15 models and 3 benchmarks.

It achieves mean gains of +12.26 MC1 and +8.65 MC2 points.

Gains reach up to +47.20 MC1 and +43.38 MC2 points.

Abstract

Hallucination correction is not a one-direction problem. We show that intermediate layers are neither uniformly more truthful than final layers nor uniformly less trustworthy. Yet hallucination reduction is usually instantiated through one fixed intervention form: contrast one layer against another, steer along a truthfulness direction, or defer to external evidence. This framing is structurally incomplete. Cross-layer factual evidence does not evolve uniformly: in some failures truthful support is present internally and later suppressed, whereas in others candidate competition remains genuinely multi-directional across depth, so no single signed scalar family is generally sufficient. We introduce Trajectory Correction from Cross-layer Evidence for Hallucination Reduction (TRACE), a deterministic, training-free algorithm which corrects hallucinations at inference time by deriving both the corrective layer and the appropriate correction operator from each input's cross-layer candidate trajectory inside the LLM's own forward pass. Under one frozen hyperparameter setting, TRACE selects among scalar reversal, earlier-state recovery, and candidate-space correction using only model-internal evidence. Evaluated as a single universal algorithm across 15 models, 8 model families, and 3 factuality benchmarks, TRACE improves every evaluation cell, yielding mean gains of +12.26 MC1 points and +8.65 MC2-style points with no regressions, with gains reaching +47.20 MC1 and +43.38 MC2-style points. The method uses no labels, retrieval, pretraining, finetuning, or per-model calibration.