NUWA 风险评测 · 研究方法
Introducing Bloom : an open source tool for automated behavioral evaluations
NUWA 研究解读 · 独立复核待完成
NUWA Research Note
研究问题研究设计与假设来源发现NUWA 解读局限与相反证据研究与决策意义机构与角色原始资源模型辅助整理 · 选读原始全文或官方文章中的相关章节；尚未完成逐章节审计。
研究速读
NUWA 解读：种子属于测量定义，不是可省略实现细节。
研究问题
生成式评测能否可复现地测研究者指定行为？
研究设计与假设
Bloom依据行为描述和例轨生成场景，以评测种子记录配置，自动判断结果。
来源发现
发布展示16模型的四类行为，并与人工标签核对。
绝对指标随配置改变，一个自偏好研究的排序较稳定。
NUWA 解读
NUWA 解读：种子属于测量定义，不是可省略实现细节。
局限与相反证据
动态生成及筛选改变受测总体，特定行为验证不证明普遍判官准确。
研究与决策意义
模型建议：每项结果同时引用种子、生成筛选设置及验证任务。
精确来源位置
- Introducing Bloom : an open source tool for automated behavioral evaluations · How Bloom worksSelected original resource https://www.anthropic.com/research/bloom; sha256:67c6404128375c3d8baa90e80084fed90b800eb20485afb1d51468ea0a6ca1de; archived source capture; exact capture time not independently established
- Introducing Bloom : an open source tool for automated behavioral evaluations · Validation and trustSelected original resource https://www.anthropic.com/research/bloom; sha256:67c6404128375c3d8baa90e80084fed90b800eb20485afb1d51468ea0a6ca1de; archived source capture; exact capture time not independently established
- Introducing Bloom : an open source tool for automated behavioral evaluations · Case study: Self-preferential biasSelected original resource https://www.anthropic.com/research/bloom; sha256:67c6404128375c3d8baa90e80084fed90b800eb20485afb1d51468ea0a6ca1de; archived source capture; exact capture time not independently established
来源核查、数据提取、编辑批准与实验复现分别记录。 方法说明 →
机构与角色
引用与复用
引用研究发现时请引用原始出版物；引用 NUWA 的整理与解释时，单独引用本站记录及对应快照。
原始出版物 · BibTeX
@misc{bloom,
title = {Introducing Bloom : an open source tool for automated behavioral evaluations},
url = {https://www.anthropic.com/research/bloom},
year = {2025},
date = {2025-12-19},
version = {Source capture 2026-10-25},
note = {Source capture 2026-10-25}
}
NUWA 记录 · BibTeX
@misc{bloom-nuwa,
title = {Introducing Bloom : an open source tool for automated behavioral evaluations — NUWA research record},
url = {https://whitzard.tech/nuwa/whitzard-index/work/bloom/},
version = {2026-10-29.1},
organization = {NUWA Frontier AI Safety Lab},
howpublished = {Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--bloom.json},
note = {NUWA research note. Source-reading scope and independent review status are stated on the record. Snapshot 2026-10-29.1. Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--bloom.json.}
}
证据快照 2026-10-29.1 · 索引更新 · RSS
参考文献
- [1] Introducing Bloom : an open source tool for automated behavioral evaluations. 2025-12-19 ; Official publication masthead and article byline; How Bloom works; Validation and trust