Short Answer

When Human-Versus-Model Comparisons Stop Helping

Question: In the human-better subset idea, at what point do comparisons against human performance stop being useful for improving a machine learning system?

Sample answer: They stop only when the dev set no longer contains any cases where a person is correct and the model is wrong. Surpassing the average human score overall is not enough by itself.

Key points:

  • The method still works as long as such cases remain.
  • What matters is the absence of any human-correct, model-wrong examples.
  • An above-average overall score does not eliminate those subsets.

Rubric: Give credit only if the answer says the technique ends when there are no remaining examples in which humans outperform the model.

0

1

Updated 2026-08-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI