arXiv 2508.03333
CTTS: Collective Test-Time Scaling
By Zhende Song, Shengji Tang, et al.
Published 2025-08-05
Wiki summary
Explore the paper's summary, context, and related research on Papiers.
Test-time scaling (TTS) has emerged as a promising, training-free approach for enhancing large language model (LLM) performance. However, the efficacy of existing methods, such as Best-of-N and Self-Consistency, is fundamentally constrained by the dominant single test-time scaling (STTS) paradigm, which relies on a single LLM agent interacting with a single reward model (SA-SR). Inspired by recent work showing that…