> Because once trained, those models provide values to millions of users. Do they? Those of us in tech often take the positive value of technological progress

> Because once trained, those models provide values to millions of users. Do they? Those of us in tech often take the positive value of technological progress as a given, but when looking at, e.g., the example in OpenAI's paper of GPT4 tricking a human into solving a CAPTCHA, I think it's quite clear there's possible negative value as well, and claims about "value to millions of users" probably need to be substantiated.