AI4AI at Test-Time: Strong-to-Weak Capability Transfer via Harnesses

2026-08-12

AI Summary

This work addresses the challenge of effectively transferring capabilities from large language models to smaller ones without updating their parameters. The authors propose a training-free, test-time method for strong-to-weak capability transfer by constructing a “harness” architecture during inference. In this framework, a strong model generates deterministic code, performs task routing, and enforces answer-format constraints, thereby transforming inherently unstable reasoning processes into deterministic logical operations. This approach achieves pure inference-stage capability transfer—without any parameter updates—and substantially boosts performance on four theory-of-mind benchmarks, raising the average accuracy of target models from 0.49 to 0.91, nearly doubling their original performance, with the most pronounced gains observed in weaker models.

Abstract

Recent work on distillation transfers the capabilities of large models to smaller ones often by updating the latter's parameters, through teacher forcing, on-policy distillation, and related training-time methods. In this paper, we ask whether such transfer can instead occur at test time. We study strong-to-weak scaffolding: whether a stronger builder model can construct inference-time harnesses that help a weaker target model solve tasks more reliably without any parameter updates. Using four representative Theory-of-Mind benchmarks, each builder model uses 5% of the data as a validation set to iteratively refine its harness over multiple rounds, after which the finalized harness is evaluated on the full test set. Empirically, this form of test-time capability transfer is highly effective, nearly doubling average target-model performance from 0.49 to 0.91. Our analysis shows that the gains come primarily from offloading unstable model reasoning into deterministic code, benchmark-specific routing, and strict answer-format enforcement, rather than from encouraging the target model to reason more extensively or sample more broadly. We further find that builder-model reasoning effort improves harness quality monotonically, platform effects are modest relative to the builder model's own capability, and weaker target models receive the largest gains. These results suggest that inference-time harness design is an important complement to conventional training-time distillation, enabling strong models to transfer cognitive structure to weaker models without retraining.

Problem

• test-time transfer
• strong-to-weak scaffolding
• inference-time harness
• capability transfer
• model distillation

Innovation

• test-time distillation
• strong-to-weak transfer
• inference-time harness
• scaffolding
• capability transfer