Do LLM modelers maintain a list of manual corrections fed by humans?

Like the how many r's in strawberry. It took off as an Internet meme and was fixed, but how did that fix happen?