That's exactly what I mean and it is inevitable without halting all non-human verified updates to the dataset going forward. There are two worlds now, pre-GPT and post-GPT. In the pre-GPT world you have human error and in the post-GPT world you have human error and machine error compounding ad infinitum.