{"slug": "when-more-sampling-hurts-the-modal-ceiling-and-correlation-ceiling-of-test-time", "title": "When More Sampling Hurts: The Modal Ceiling and Correlation Ceiling of Test-Time Scaling", "summary": "A new study reveals that test-time scaling in language models, where models sample multiple answers to a question, hits a ceiling beyond which additional samples do not improve selection accuracy and can even degrade performance. The authors identify a \"modal ceiling\" for voting and a \"correlation ceiling\" for benchmarking, proposing an \"effective number of samples\" metric to determine the optimal cutoff. The findings challenge the assumption that more sampling always leads to better outcomes, highlighting the identifiability gap between coverage and selection.", "body_md": "arXiv:2606.28661v1 Announce Type: new\nAbstract: People overthink; language models over-sample, and the extra effort can talk both into a worse answer. Reasoning systems answer a hard question by sampling it many times (test-time scaling), and the more they draw, the more often a correct answer turns up somewhere, so coverage, the fraction of problems with at least one correct try, climbs and appears to be progress. But a deployed system must return one answer, and choosing it, not knowing which try is right, is selection; selection is capped, and past a point extra samples only make the model surer of a confident mistake, even as every draw adds cost. The gap between climbing coverage and stalled selection, the identifiability gap, is the answer a model can produce but not pick. So the real question is not whether to sample but how far, and the answer is: not far. For picking an answer, the vote has already settled within a few dozen draws, the modal ceiling; for scoring a benchmark, sooner still, the correlation ceiling. Beyond that, extra draws cost compute and add nothing, and can even make the answer worse. This paper turns the cutoff into a single number, the effective number of samples, that any sampling run already reveals. The bottleneck is recognizing a right answer, not generating one.", "url": "https://wpnews.pro/news/when-more-sampling-hurts-the-modal-ceiling-and-correlation-ceiling-of-test-time", "canonical_source": "https://arxiv.org/abs/2606.28661", "published_at": "2026-06-30 04:00:00+00:00", "updated_at": "2026-06-30 04:30:29.677646+00:00", "lang": "en", "topics": ["large-language-models", "ai-research", "ai-safety"], "entities": [], "alternates": {"html": "https://wpnews.pro/news/when-more-sampling-hurts-the-modal-ceiling-and-correlation-ceiling-of-test-time", "markdown": "https://wpnews.pro/news/when-more-sampling-hurts-the-modal-ceiling-and-correlation-ceiling-of-test-time.md", "text": "https://wpnews.pro/news/when-more-sampling-hurts-the-modal-ceiling-and-correlation-ceiling-of-test-time.txt", "jsonld": "https://wpnews.pro/news/when-more-sampling-hurts-the-modal-ceiling-and-correlation-ceiling-of-test-time.jsonld"}}