On the Limits of Test-Time Compute: Sequential Reward Filtering for Better Inference
This work addresses a foundational problem in AI by improving inference efficiency for large language models, though it is incremental as it builds on existing test-time compute paradigms.
The paper tackled the problem of unclear fundamental limits in test-time compute methods for large language models, proving that standard best-of-n sampling is suboptimal and showing that reward-filtered sequential inference yields stronger theoretical guarantees and consistent empirical improvements across diverse benchmarks.
Test-time compute (TTC) has become an increasingly prominent paradigm for enhancing large language models (LLMs). Despite the empirical success of methods such as best-of-$n$ (BoN) sampling and sequential revision, their fundamental limits remain unclear. We address this gap by analyzing a mixture-of-reference policy model and proving that standard BoN is inherently suboptimal. To move closer to the optimal frontier, we study reward-filtered sequential inference, a simple procedure that selectively incorporates only high-reward generations into the context. This mechanism concentrates computation on superior policy candidates and suppresses inferior ones. On the theoretical side, we show that reward-filtered sequential inference yields strictly stronger guarantees than standard TTC paradigms. On the empirical side, we evaluate such an inference strategy across diverse benchmarks and observe consistent improvements over widely used approaches, demonstrating the practical effectiveness of our framework.