#machine learning
Jul 2026
Final Checkpoints Are Not Enough: Analyzing Latent Reasoning Faithfulness Along Training Trajectories
These results show that evaluating only a final checkpoint can obscure both when counterfactual responsiveness changes and what the latent states contribute, and show that high task accuracy can coexist with low counterfactual responsiveness.
Heng-Yu Jin, Shu Yang, Di Wang
· arXiv.org · 1 citation