> We then used the error spotting tool on the Deepomatic platform to detect errors and to correct them. I'm wondering if those errors are selected on how much

> We then used the error spotting tool on the Deepomatic platform to detect errors and to correct them. I'm wondering if those errors are selected on how much they impact the performance? Anyway, this is probably a much better way of gaining accuracy on the cheap than launching 100+ models for hyperparameter tuning.