风险评测 · 评测基准

Introducing VCT-v2 — the updated Virology Capabilities Test

NUWA 研究解读 · 独立复核待完成

首次发表: 2026-09-11来源核验: 2026-10-25

原始来源 ↗引用本记录 ↓化生放核 →

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：内容与动态范围改变后，两个版本不能直接视为同一尺度。

研究问题

修订能否提高VCT答案可靠性与抗捷径能力？

研究设计与假设

以回答模式确定性规则优先复审，并删除或修改问题，测试缺题干或缺图条件。

Virology Capabilities Test v2 — Source figure — Source figure · 原文 Source figure。请结合下方的协议范围与局限阅读。原图与来源 ↗

来源发现

VCT-v2有279题，所列最佳模型49.2%。

逐题跨模型聚合得到有效上限至少79%，并非单模型得分。

NUWA 解读

NUWA 解读：内容与动态范围改变后，两个版本不能直接视为同一尺度。

局限与相反证据

上限只是下界；修改题目的表现不能直接验证实际用户增益。

研究与决策意义

模型建议：引用版本和题目总体，区分观测正确率与推断空间。

精确来源位置

Introducing VCT-v2 — the updated Virology Capabilities Test · Patching potential issues in VCTSelected original resource https://securebio.org/blog/introducing-vct-v2/; sha256:be1ae6cf6276b2e9dcd4fd7cb96e652548b619e15cb341e7392245dfcdaa729b; archived source capture; exact capture time not independently established
Introducing VCT-v2 — the updated Virology Capabilities Test · Presenting our updated benchmark — VCT-v2Selected original resource https://securebio.org/blog/introducing-vct-v2/; sha256:be1ae6cf6276b2e9dcd4fd7cb96e652548b619e15cb341e7392245dfcdaa729b; archived source capture; exact capture time not independently established
Introducing VCT-v2 — the updated Virology Capabilities Test · VCT-v2 is substantially more shortcut-resistant than VCTSelected original resource https://securebio.org/blog/introducing-vct-v2/; sha256:be1ae6cf6276b2e9dcd4fd7cb96e652548b619e15cb341e7392245dfcdaa729b; archived source capture; exact capture time not independently established
Introducing VCT-v2 — the updated Virology Capabilities Test · Monitoring VCT-v2 for saturationSelected original resource https://securebio.org/blog/introducing-vct-v2/; sha256:be1ae6cf6276b2e9dcd4fd7cb96e652548b619e15cb341e7392245dfcdaa729b; 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

发布机构 SecureBio

评测协议

Virology Capabilities Test v2 →

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。

BibTeX ↓RIS ↓记录与来源元数据 · JSON ↓阅读覆盖 · JSON 提交更正

原始出版物 · BibTeX

@misc{vct-v2,
  title = {Introducing VCT-v2 — the updated Virology Capabilities Test},
  url = {https://securebio.org/blog/introducing-vct-v2/},
  year = {2026},
  date = {2026-09-11},
  version = {SecureBio release 2026-09-11},
  note = {SecureBio release 2026-09-11}
}

NUWA 记录 · BibTeX

@misc{vct-v2-nuwa,
  title = {Introducing VCT-v2 — the updated Virology Capabilities Test — NUWA research record},
  url = {https://whitzard.tech/nuwa/whitzard-index/work/vct-v2/},
  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--vct-v2.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--vct-v2.json.}
}

证据快照 2026-10-29.1 · 索引更新 · RSS

参考文献

[1] Introducing VCT-v2 — the updated Virology Capabilities Test. 2026-09-11 ; Official publication masthead and article byline; Patching potential issues in VCT; Presenting our updated benchmark — VCT-v2