Exploring System 1 and 2 communication for latent reasoning in LLMs
This work addresses the problem of efficient reasoning in LLMs for AI researchers, showing current dual-model approaches are mostly incremental and add computational cost without qualitative improvements.
The study investigated whether dual-architecture latent reasoning (with separate Base and Coprocessor models) improves LLM reasoning compared to a unified model, finding that joint finetuning (H2) performed best but a unified baseline nearly matched it, while increasing channel capacity (H1) or scaling latent-token budgets yielded minimal gains.
Should LLM reasoning live in a separate module, or within a single model's forward pass and representational space? We study dual-architecture latent reasoning, where a fluent Base exchanges latent messages with a Coprocessor, and test two hypotheses aimed at improving latent communication over Liu et al. (2024): (H1) increase channel capacity; (H2) learn communication via joint finetuning. Under matched latent-token budgets on GPT-2 and Qwen-3, H2 is consistently strongest while H1 yields modest gains. A unified soft-embedding baseline, a single model with the same forward pass and shared representations, using the same latent-token budget, nearly matches H2 and surpasses H1, suggesting current dual designs mostly add compute rather than qualitatively improving reasoning. Across GSM8K, ProsQA, and a Countdown stress test with increasing branching factor, scaling the latent-token budget beyond small values fails to improve robustness. Latent analyses show overlapping subspaces with limited specialization, consistent with weak reasoning gains. We conclude dual-model latent reasoning remains promising in principle, but likely requires objectives and communication mechanisms that explicitly shape latent spaces for algorithmic planning.