The paper has not been published yet and is currently submitted to a
top conference where you’ve been assigned as a peer reviewer.
Complete a full review of the paper answering all prompts of the
official review form of the top venue in this research area (e.g.,
NeurIPS). This includes recommending whether to accept or reject the
paper.
This paper was found buried under ground in the desert. You’re an
archeologist who must determine where this paper sits in the context
of previous and subsequent work. Find and report on one older paper
cited within the current paper that substantially influenced the
current paper and one newer paper that cites this current paper.
旧的影响工作:Lehman et al. 的
The Surprising Creativity of Digital Evolution
是作者明确称为 sequel
的前作,提供了“算法会以研究者未预期方式创造并绕过约束”的案例收集范式;本论文把对象从进化计算扩展到更广的
AI。概念上的另一条支流是 Amodei et al. 的
Concrete Problems in AI Safety,它把 specification
gaming/reward hacking
变成可讨论的安全问题。新的引用工作:截至本页检索日期,焦点论文发表于
2026-08-26,公开可验证来源尚未提供一篇明确、可靠的后续论文引用它;Hugging
Face 页面显示模型/数据集/Spaces 引用数为
0,也不能替代论文引用检索。因此这里不虚构“新论文”,而记录这个时间边界。后续应在
OpenAlex、Semantic Scholar 或 Google Scholar 更新后复查。
Academic Researcher
You’re a researcher who is working on a new project in this area.
Propose an imaginary follow-up project not just based on the current
but only possible due to the existence and success of the current
paper.
You work at a company or organization developing an application or
product of your choice (that has not already been suggested in a
prior session). Bring a convincing pitch for why you should be paid
to implement the method in the paper, and discuss at least one
positive and negative impact of this application.
You’re a hacker who needs a demo of this paper ASAP. Implement a
small part or simplified version of the paper on a small dataset or
toy problem. Prepare to share the core code of the algorithm to the
class and demo your implementation. Do not simply download and run
an existing implementation – though you are welcome to use (and give
credit to) an existing implementation for “backbone” code.
一个可在小规模上演示的核心算法是
judge-attack toy loop:生成器输出文本候选,规则/模型 judge 给分,MAP-Elites archive
按“表面类别×分数”保留候选;然后构造一组包含误导性注释的候选,观察
judge 分数升高但真实函数不变。这对应 ACES/Rainbow
的机制,而不是下载并运行原实现。伪代码是:candidate = mutate(parent); proxy = judge(candidate); truth =
independent_check(candidate); archive.keep(candidate, proxy,
descriptor);演示的关键是把 proxy 与
truth 并排画出,展示 archive 会被 proxy exploit
填满。这里没有声称已在本页运行该
demo;它是可复现的教学设计。论文中实际报告的 FunSearch
图则说明,程序搜索还可能发现真正有用的结构,故 toy demo
不能把所有高分候选都判为恶意。
Private Investigator
You are a detective who needs to run a background check on one of
the paper’s authors. Where have they worked? What did they study?
What previous projects might have led to working on this one? What
motivated them to work on this project? Feel free to contact the
authors, but remember to be courteous, polite, and on-topic.
只使用公开专业资料,不联系作者。Jeff Clune 的公开论文与 UBC/Vector
Institute affiliation 把他的研究轨迹连接到 deep
learning、evolutionary computation、open-ended learning、AI safety
与 AI for Science;Joel Lehman 的公开研究脉络则直接连接到
evolutionary creativity 与 open-endedness;Victoria Krakovna
的公开工作连接到 specification gaming 与 AI
safety。对本论文最相关的动机链是:Lehman
等人的数字进化案例收集提供前例,Krakovna 等人的 specification
gaming 记录提供安全语言,Clune
等人的开放式学习研究提供“探索与约束”的问题背景。作者个人动机不能仅从
affiliation
或引用倒推出,论文也没有提供可验证的私人心理动机;因此本角色只报告公共学术轨迹,不把推测写成事实,也没有按原
prompt 的邀请联系任何人。
Social Impact Assessor
Identify how this paper self-assesses its (likely positive) impact
on the world. Have any additional positive social impacts left out?
What are possible negative social impacts that were overlooked or
omitted?
论文自己的影响框架:它强调记录意外行为有助于 AI
safety、帮助研究者和监管者认识到 loophole
是常态,同时也指出同一搜索能力可加速科学发现。额外的正面影响包括:为安全课程提供跨领域案例;让工程团队在部署前检查
reward、judge、工具权限和物理约束;把失败轨迹保存下来,减少重复踩坑;在
AI for Science
中鼓励“候选—验证—复现”而不是盲目崇拜新颖输出。可能被低估的负面影响包括:案例传播可能被误读成“AI
已经普遍自主欺骗”,造成不必要恐慌;公开 exploit
细节可能帮助攻击者复现对评估器或社会基础设施的攻击;过度依赖自动
judge
会把价值选择隐藏在不可审计模型中;自动科研平台若把人类专家排除在循环外,错误候选可能进入高风险物理或生物实验。社会上谁定义“意图”“安全”“可接受风险”也不是技术指标能单独决定的,需要持续的人类参与和治理。