Search papers, labs, and topics across Lattice.
Affiliation:
4
0
9
F2DR evaluates DeepSearch workflows across three dimensions: Content, Trajectory, and Answer, enabling comprehensive process-level assessment, and constructs DeepSearch RM-Bench, a dedicated benchmark for evaluating RMs in DeepSearch scenarios.
LLMs can compress context better than dedicated compression modules, simply by prompting them to "think" about the task.
Forget O(N^5) scaling: this new tensor decomposition approach slashes the computational cost of second-order perturbation theory down to O(N^3) without sacrificing accuracy.
By reflecting on its own reasoning, ReflectRM achieves a +10.2 improvement in mitigating positional bias compared to leading generative reward models, making it a far more stable evaluator.
