SAGE: Mitigating Long-Horizon Reasoning Biases via Topological Guidance
SAGE uses structural priors to keep long reasoning paths from drifting into plausible but unstable branches.
The paper frames long-horizon failures as two biases: over-exploring bad branches and compounding small errors across depth. Its SAGE framework combines algebraic sparsification with hyperbolic guidance to narrow search and add denser signals. The authors report gains across 12 benchmarks and 7 model families, including up to an 8-fold improvement on the Andrews-Curtis problem. Code is listed as available. ArXiv · AI/CL/LG's note
The paper frames long-horizon failures as two biases: over-exploring bad branches and compounding small errors across depth. Its SAGE framework combines algebraic sparsification with hyperbolic guidance to narrow search and add denser signals. The authors report gains across 12 benchmarks and 7 model families, including up to an 8-fold improvement on the Andrews-Curtis problem. Code is listed as available. ArXiv · AI/CL/LG's note
score 5