问题
Why do we stay inside echo chambers?
把你的贴纸用在最重要的内容上。有证据支持的想法可以占用一个保留席位。
已关闭 · 返回行动阶段
正以访客身份查看。输入用户名即可加入并参与。
你的贴纸额度
每个想法最多一张，共 5 张。至少投一票。
有 5 个席位为被忽视的想法保留——附上链接提名一个，当 2 人认同来源相关时就会计入。
为什么
阶段结束时，一名被提名者通过随机抽签入选，其余按提名数入选，每个利益相关方类别最多一个。模型只检查页面是否关于同一主题——从不检查它是否证明了这个主张。那要由房间来判断；没有其他人支持的提名不能占用席位。
类别 1
We cannot publish what we remove without teaching people to evade it
Narrower than it reads; the wider version is someone else's card.
2 票讨论
An A/B test measures engagement because that is what we can measure in a week
3 票讨论
类别 2
The rules are forty pages and the decision takes twelve seconds
Observed, not measured — I would want the data too.
5 票讨论
The rules are written in one country and applied in forty languages
3 票讨论
类别 3
Moderation is a cost centre and ranking is a profit centre
Scope: here specifically, not everywhere.
2 票讨论
A moderator sees the worst of it for the wage of a call centre
5 票讨论
未归类
The metric was chosen because it was easy to log
6 票讨论
Nobody can explain why a particular post was shown
2 票讨论
An appeal is judged by the system that made the decision
Scope: here specifically, not everywhere.
1 票讨论
A change that reduces engagement is reverted before anyone asks why
0 票讨论
The founders were twenty-five and nobody senior said stop
Scope: here specifically, not everywhere.
3 票讨论
Every external audit proposal dies in procurement
Meaning the recurring case, not one incident.
0 票讨论中（1）
We built for a scale nobody had governed before
5 票讨论
Every fix ships to the users who complain loudest
Observed, not measured — I would want the data too.
5 票讨论
Ad money rewards volume of attention, not quality of it
Observed, not measured — I would want the data too.
14 票讨论
There is no reviewer who speaks the language of half the reports
Narrower than it reads; the wider version is someone else's card.
5 票讨论
The people affected worst are in markets we do not staff
3 票讨论
Safety work has no metric, so it has no headcount
To be precise: this is about the situation, not a person.
6 票讨论
Regulation asks for transparency reports nobody can verify
1 票讨论
The team that would study long-term effects does not exist
0 票讨论
Harm shows up in a quarter and the review cycle is six weeks
Narrower than it reads; the wider version is someone else's card.
1 票讨论
Recommendation systems are not inspectable from outside
8 票讨论
The model is retrained weekly and nobody diffs its behaviour
0 票讨论
Ranking optimises for time spent, not for being informed
15 票讨论
The people who raise concerns are moved to another team
To be precise: this is about the situation, not a person.
0 票讨论
The company would rather be regulated than be sued
Meaning the recurring case, not one incident.
0 票讨论
People choose the sources that already agree with them
12 票讨论中（1）
Automated removal is cheaper than being right
4 票讨论
Nobody outside can reproduce a single claim we make
13 票讨论
Nobody agrees on who should decide what is true
11 票讨论