Loading

PeerCurrent.com|Stay current with the people and research shaping what comes next.

PcPeerCurrent

Your Feed AI Summary Saved Board Expert Discussion This Week in Your Field

Follows Journals Trials Guidelines Markets

Set your cadence Alerts Settings Help

Back to follows

Researcher

Haohui Yang

1 development in the corpus

Following this puts new work involving it at the top of your briefing, with a note saying why it is there. Links are taken from the source record, never inferred.

For You Latest Breakthroughs Trials Reviews Guidelines

Latest

Everything in the corpus, newest first.

AUG 14, 2026 · PREPRINT

More Correct Mass, Worse Answers: Why Power Sampling Can Fail and How to Fix It

arXiv

This is an unreviewed preprint describing a methodological problem in language model sampling and proposing a fix, demonstrated empirically across benchmarks but without peer review or clinical/real-world validation.

PreliminaryJiaxing Sun Xiujun Ma Haohui Yang

Reported

Accuracy drop from Power Samplingup to 18.5 percentage points

PeerCurrent.com|Stay current with the people and research shaping what comes next.

About Privacy Terms Support