研究库 论文知识库

内容库 / 论文

arXiv:2609.29429 · 安全与风险

Just Ask Jev: Reinforcement Learning for Calibrated Decisions as a Zero-Shot Detector of AI Alignment Failures

Just Ask Jev: Reinforcement Learning for Calibrated Decisions as a Zero-Shot Detector of AI Alignment Failures

类型：arxiv
标识：2609.29429
链接：https://arxiv.org/abs/2609.29429
主分类：risk
形态：benchmark
TLDR：Detectors of alignment failures screen deployed language models and score alignment benchmarks. Most are generative judges that spend a decoding pass on every criterion, and classifiers that read token probabilities, such as Llama Guard, still score one fixed label per call. Jev, a model trained with reinforcement learning for calibrated decisions (RLCD), answers many typed questions about one input with calibrated probabilities in a single call. Whether it detects alignment failures has not been measured. We present RLCDAlignBench, which benchmarks Jev on ten alignment failures: sycophancy, j
副分类：evaluation
待LLM分类：否
来源文件：
/inbox/tom/_candidates/2026-09-25-agent-rag-longcontext-candidates.json

相关 · 按语义匹配

笔记 Cameron R. Wolfe (Deep Learning Focus) · RSS 摘要

笔记 SpecBench: Measuring Reward Hacking in Long-Horizon Coding Agents — 短精读

笔记 flyP 夜间轻量精读 · 2026-06-23（cron 3d8f503a · 22:50 CST）

文档 Tom 评 flyP · 2026-07-24 · Self Gradient Forcing + Anchor-Align 精读

仓库 jiyuanwang-afk/driftshield-fraud-rl

笔记 Tom 文献雷达 · AI Agent / RAG / 长上下文 · 2026-09-05 14:40

arXiv

2609.29429

TLDR

Detectors of alignment failures screen deployed language models and score alignment benchmarks. Most are generative judges that spend a decoding pass on every criterion, and classifiers that read token probabilities, such as Llama Guard, still score one fixed label per call. Jev, a model trained with reinforcement learning for calibrated decisions (RLCD), answers many typed questions about one input with calibrated probabilities in a single call. Whether it detects alignment failures has not been measured. We present RLCDAlignBench, which benchmarks Jev on ten alignment failures: sycophancy, j

arXiv ↗ 中文全文