Same Signal, Opposite Meaning: Direction-Informed Adaptive Learning for LLM Agents

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Li, Ziming, Huang, Jiatan, Guo, Xiaoguang, Wang, Guilin, Zhang, Chuxu
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911660895109120
author Li, Ziming
Huang, Jiatan
Guo, Xiaoguang
Wang, Guilin
Zhang, Chuxu
author_facet Li, Ziming
Huang, Jiatan
Guo, Xiaoguang
Wang, Guilin
Zhang, Chuxu
contents Adaptive test-time compute for LLM agents aims to invoke extra computation only when it improves performance. Existing methods typically use confidence-, uncertainty-, or difficulty-based gates, assuming a fixed direction from the gating signal through compute need to the value of computation. This makes gating a utility-calibration problem: gating signals should align with whether extra computation improves the final outcome over the base policy. We show that this alignment is unstable: the same signal predicts rollout benefit in one setting and rollout harm in another, with reversals across environments and backbones even when the task is fixed. Wrong-direction gates can therefore worsen performance by precisely selecting harmful states. This reversal reflects a deeper distinction between compute need and compute suitability: a high uncertainty signal may indicate decision-difficult states where rollouts help compare alternatives, or intervention-unsuitable states where the current context does not support useful rollout-based improvement. Under this two-source model, fixed-direction gates are unreliable across heterogeneous settings. To address this, we propose DIAL (Direction-Informed Adaptive Learning), a sparse gate trained from signal-agnostic counterfactual exploration to learn the utility direction of state features per (environment, backbone). Across six environments and three backbones, DIAL yields a stronger overall success-cost trade-off than fixed-direction baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2605_06908
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Same Signal, Opposite Meaning: Direction-Informed Adaptive Learning for LLM Agents
Li, Ziming
Huang, Jiatan
Guo, Xiaoguang
Wang, Guilin
Zhang, Chuxu
Machine Learning
Artificial Intelligence
Adaptive test-time compute for LLM agents aims to invoke extra computation only when it improves performance. Existing methods typically use confidence-, uncertainty-, or difficulty-based gates, assuming a fixed direction from the gating signal through compute need to the value of computation. This makes gating a utility-calibration problem: gating signals should align with whether extra computation improves the final outcome over the base policy. We show that this alignment is unstable: the same signal predicts rollout benefit in one setting and rollout harm in another, with reversals across environments and backbones even when the task is fixed. Wrong-direction gates can therefore worsen performance by precisely selecting harmful states. This reversal reflects a deeper distinction between compute need and compute suitability: a high uncertainty signal may indicate decision-difficult states where rollouts help compare alternatives, or intervention-unsuitable states where the current context does not support useful rollout-based improvement. Under this two-source model, fixed-direction gates are unreliable across heterogeneous settings. To address this, we propose DIAL (Direction-Informed Adaptive Learning), a sparse gate trained from signal-agnostic counterfactual exploration to learn the utility direction of state features per (environment, backbone). Across six environments and three backbones, DIAL yields a stronger overall success-cost trade-off than fixed-direction baselines.
title Same Signal, Opposite Meaning: Direction-Informed Adaptive Learning for LLM Agents
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2605.06908