评分 6 · 来源：cs.LG updates on arXiv.org · 发布于 2026-04-30
评分依据：Shows input embeddings can control aligned model behavior at test time, extends beyond simple profanity reduction
Test-Time Safety Alignment
发布
采集
学术前沿 6.0 分
— Shows input embeddings can control aligned model behavior at test time, extends beyond simple profanity reduction