arXiv:2607.06648 · v3

Final Checkpoints Are Not Enough: Analyzing Latent Reasoning Faithfulness Along Training Trajectories

首次提交 2026-07-08 00:09

版本更新 2026-11-04 16:13

雷达观测 2026-11-05 11:10

摘要

Latent reasoning performs multi-step inference in continuous hidden states, promising more compact and efficient reasoning. However, these opaque states raise a question of faithfulness: whether the latent reasoning steps drive the final answer. Prior work reports unfaithful behaviors at selected checkpoints, leaving their development during training unexamined. We track behavioral and activation-based evidence across training using verified counterfactual edits and interventions on the latent reasoning states. We find that accuracy can improve while models become less responsive to edits that change the required reasoning path and the correct answer. This counterfactual responsiveness follows distinct trajectories across latent reasoning approaches. On the graph reachability task ProsQA, output sensitivity to noise replacement declines alongside counterfactual responsiveness. Across separately trained binary-choice and open-ended versions of GSM8K math problems, interventions on all latent states show opposite sensitivity trajectories. Guided by these trajectories, we intervene on the training procedure and improve counterfactual responsiveness. These results show that evaluating only a final checkpoint can obscure both when counterfactual responsiveness changes and what the latent states contribute.