Search papers, labs, and topics across Lattice.
The study investigates the latent planning capabilities of LLMs by training and prompting them to solve graph path-finding tasks without explicit supervision on intermediate steps. The key finding is a "depth ceiling" where even large models like GPT-4o and GPT-5.4 struggle to discover latent planning strategies beyond a limited number of steps (5-7), despite generalizing slightly beyond this during testing. This suggests inherent limitations in LLMs' ability to perform complex, multi-step reasoning within their latent representations.
Even massive LLMs like GPT-5.4 hit a surprisingly low "depth ceiling" of just 5-7 steps when discovering multi-step planning strategies latently, suggesting that complex reasoning may require explicit training or externalization.
The viability of chain-of-thought (CoT) monitoring hinges on models being unable to reason effectively in their latent representations. Yet little is known about the limits of such latent reasoning in LLMs. We test these limits by studying whether models can discover multi-step planning strategies without supervision on intermediate steps and execute them latently, within a single forward pass. Using graph path-finding tasks that precisely control the number of required latent planning steps, we uncover a striking limitation unresolved by massive scaling: tiny transformers trained from scratch discover strategies requiring up to three latent steps, fine-tuned GPT-4o and Qwen3-32B reach five, and GPT-5.4 attains seven under few-shot prompting. Although the maximum latent planning depth models can learn during training is five, the discovered strategy generalizes up to eight latent steps at test-time. This reveals a dissociation between the ability to discover a latent strategy under final-answer supervision alone and the ability to execute it once discovered. If similar limits hold more broadly, strategies requiring multiple coordinated latent planning steps may need to be explicitly taught or externalized, lending credence to CoT monitoring.
