Hacker News new | past | comments | ask | show | jobs | submit
whichever LLM GP is using, there's undoubtedly mutliple JS engines' source code in the training data, making it not really "from scratch".