Dataset size unclear

It is often unclear how large a dataset needs to be because the answer depends on the task, the quality of the labels, and how much variability exists in the real world. More data is helpful, but only if it is representative and usable.

Small high-quality datasets can outperform large noisy ones, especially when the use case is narrow. On the other hand, broad and messy problems usually need more coverage to work reliably.

The safest approach is to measure learning curve behavior instead of guessing. When performance begins to flatten, the team can judge whether more data will still produce meaningful gains.