Search papers, labs, and topics across Lattice.
This paper introduces GRIP, a novel approach that mitigates query dominance in retrieval-augmented generation by imposing a capacity asymmetry between the query and retrieved evidence. By allowing the decoder full access to the query while constraining the evidence through a stochastic bottleneck, GRIP effectively reduces mutual information between the query and latent state, leading to significant improvements in reasoning tasks. The method achieves a 73% reduction in hallucination rates and outperforms strong iterative baselines across five reasoning benchmarks, demonstrating its efficacy in enhancing the relevance of retrieved evidence.
GRIP cuts query-latent mutual information by 30x while slashing hallucinations by 73%, revolutionizing how we leverage retrieved evidence in reasoning tasks.
High-capacity encoders in retrieval-augmented generation (RAG) can let the query dominate the latent state, leaving retrieved evidence functionally irrelevant. We call this failure mode query dominance. To address it, we introduce \textbf{GRIP} (Grounded Reasoning via Information-Restricted Premises), which imposes capacity asymmetry: the decoder keeps full-dimensional access to the query, while retrieved evidence passes through a severe stochastic bottleneck. This forces the evidence channel to encode only the residual information unavailable from the query. Across five reasoning benchmarks, GRIP outperforms strong iterative baselines, cuts a query--latent mutual-information diagnostic by roughly 30$\times$ (14.8 $\to$ 0.47 bits), and reduces hallucination by 73\%. Residual-alignment analysis further shows that the bottleneck output occupies subspaces less aligned with the query than baseline representations.
