OmniConfess：免训练逐token供述缓解全模态幻觉

1. 12:00
   arXiv · Computation and Language

   OmniConfess：通过引出 token 供词缓解全模态幻觉

   OmniConfess 是一种免训练的全模态幻觉缓解方法，固定候选回答并在受控的逐通道证据干预下以 token 粒度重新打分，生成逐 token 逐通道的结构化供词，揭示回答的证据依赖，从而保留有据内容、纠正由无关或矛盾证据驱动的表述。研究者还构建了 OmniHalluBench，包含 3,540 个样本，覆盖文本、图像、音频、视频六大数据集及判断与自由生成两类任务。代码与基准已公开。

Current heat 7·Comparable peak 10（Oct 5）·Comparable change over 24 hours –

The trend compares only the same participants observed continuously; its range may be smaller than the current heat count. Move or click on the chart to inspect hourly heat; use the left and right arrow keys to switch.