Coherence Maximization Improves Pluralistic Alignment 文章

ArXiv CS.CL2026-06-03NEWSen作者: Taslim Mahbub, Yiding Pei, Shi Feng

详细信息

来源站点
ArXiv CS.CL
作者
Taslim Mahbub, Yiding Pei, Shi Feng
文章类型
NEWS
语言
en
发布日期
2026-06-03

摘要

arXiv:2606.03110v1 Announce Type: new Abstract: Aligning AI systems with diverse human values requires value specifications grounded in concrete examples, but generating such examples without extensive human supervision remains an open challenge. We investigate what makes these examples effective, using Internal Coherence Maximization (ICM) -- which infers labels by maximizing their mutual predictability -- to generate persona-specific examples that steer a model toward a target group's values, without human supervision. Across four benchmarks spanning classification, preference, and open-ended generation, ICM-inferred in-context examples match the performance of gold labels. Crucially, coherence matters beyond individual label accuracy: with accuracy held constant, more coherent examples generalize substantially better than incoherent ones.

相关事件

暂无数据

相关公司

暂无数据

相关人物

暂无数据

相关产品

暂无数据