Sample, Scrutinize and Scale: Effective Inference-Time Search by Scaling Verification

TL;DR

This paper demonstrates that scaling up simple sampling-based search with self-verification significantly improves inference and reasoning capabilities of language models, revealing implicit scaling phenomena and principles for enhancing verification.

Contribution

It shows that minimalist sampling and self-verification, when scaled, serve as a practical inference method and introduces principles for improving test-time verification in language models.

Findings

Scaling sampling-based search improves model reasoning performance.

Implicit scaling enhances self-verification accuracy.

Benchmark reveals weak out-of-the-box verification in frontier models.

Abstract

Sampling-based search, a simple paradigm for utilizing test-time compute, involves generating multiple candidate responses and selecting the best one -- typically by having models self-verify each response for correctness. In this paper, we study the scaling trends governing sampling-based search. Among our findings is that simply scaling up a minimalist implementation of sampling-based search, using only random sampling and direct self-verification, provides a practical inference method that, for example, elevates the reasoning capabilities of Gemini v1.5 Pro above that of o1-Preview on popular benchmarks. We partially attribute the scalability of sampling-based search to a phenomenon of implicit scaling, where sampling a larger pool of responses in turn improves self-verification accuracy. We further identify two useful principles for improving self-verification capabilities with test-time compute: (1) comparing across responses provides helpful signals about the locations of errors and hallucinations, and (2) different model output styles are useful for different contexts -- chains of thought are useful for reasoning but harder to verify. We also find that, though accurate verification can be elicited, frontier models demonstrate remarkably weak out-of-box verification capabilities and introduce a benchmark to measure progress on these deficiencies.