Are Latent Reasoning Models Easily Interpretable?
A study by researchers testing the Coconut and CODI models found that these latent reasoning models barely use their hidden reasoning steps for logical tasks like PrOntoQA and ProsQA, with performance stemming from train…