[dahart]

LLMs got their training data from somewhere. But maybe they’re good at percolating the good code to the top and filtering the bad code.