Test-Time Scaling in the Wild: Why Exploitation, Not Exploration, Is the Bottleneck

The paper examines test‑time scaling (TTS) methods that use extra inference compute to improve language model outputs. Across five open‑ended benchmarks—medicine, law, finance, general chat, and creative writing—the study finds that increasing exploration (generating more candidates) consistently yields better top candidates, but exploitation (selecting the final output) remains weak due to poor reward‑model correlation. Only the Fusion approach, which synthesizes candidates, reliably improves results, yet it recovers only about 40% of the potential quality, indicating that the bottleneck lies in choosing from the candidate pool rather than generating it.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

• llms
• benchmarks