Same with all these big models. GPT-3 and codex benefit from all the text and code that has been made publicly visible without giving back. At least e.g. imagen

Same with all these big models. GPT-3 and codex benefit from all the text and code that has been made publicly visible without giving back. At least e.g. imagenet classifiers had to take the effort to label the images they scraped. IP laws need to catch up, so that ML models that use unlicensed public content have to give back in some way.