引导式测验 · 硬等级

人工智能测验的未来

测试假设、情景、更新证据、可重复的结果、不均匀的进展、预测记录和稳健的决策。

有用的人工智能预测应该说明什么？

预览每个问题8 问题

当您参加测验时，就会出现答案和解释。

有用的人工智能预测应该说明什么？
- 假设、时间范围和证据
- 只有一个自信的结论
- 用来写它的模型
- 无论不确定性如何，准确的发布日期
受控基准得到改善。这对产品的广泛采用有何意义？
- That all organizations can deploy it now
- That infrastructure and maintenance are solved
- Not when or how widely adoption will happen by itself
- That future reliability is guaranteed
为什么要使用更快、更慢和进度不均匀的场景？
- They identify which forecast is guaranteed
- They make planning robust when progress is uncertain across tasks
- They eliminate the need for indicators
- They convert a scenario into a probability
哪个指标比发布公告更好地更新能力评估？
- A company product announcement alone
- A funding round
- A marketing forecast
- Independently reproduced task performance and sustained reliability
为什么要记录预测以及什么算作未预测？
- To ensure the forecast becomes correct
- To keep the prediction secret
- To assess it against the stated outcome without changing the definition afterward
- To avoid updating assumptions
What does uneven capability growth imply?
- Progress on one benchmark does not prove the same progress in other domains
- Every task improves at the same rate
- One benchmark predicts physical interaction reliability
- A capability cannot be measured
What is a robust decision under multiple plausible futures?
- 取决于最快预测是否正确的那种
- 只有在确定后才做出的决定
- 目前没有证据的那种
- 在多种情境下仍然有用的那种
为什么要随时间复核预测？
- 预测在发布后变成事实
- 新的证据可以更新评估并揭示先前预测是否已校准
- 预测永远不应改变
- 后期复核会隐藏原始假设