Grok Heavy

New method uses entropy centroids for intrinsic rewards in LLM test-time scaling

Researchers have introduced a novel method called "Lowest Centroid" to improve the selection of high-quality responses from large language models during inference. This technique leverages the temporal structure of mode…