AI will become progressively better at refining datasets, even GPT-4 is quite good at it. From my understanding right now they use low-paid workers, often from 3rd world countries to go over data but that's not particularly efficient method and there just isn't any way to go through all the data with enough care, so there is lot of garbage in those datasets. But AI could do it, it would still require some human supervision but it would speed up the process by a lot and I expect datasets to get dramatically better over the next 5 years.