so i asked a researcher, melanie mitchell for her best guess. yeah, that's hard to say, you know, hundreds to thousands of. there's human involvement at all different levels. there's the human involvement in of figuring out what data to give the system to learn. you know what, what text and so on. there's human involvement in training, it to be kind of more conversational. and to answer questions in a more human like way. and then there's humans who try to prevent it from spewing out toxic language or other kinds of things that you know, the companies might not want it to, to say, since the release of chat t p t on an investigation by time magazine, find that open a i used workers in kenya to classify disturbing and legal content, including descriptions of sexual violence. employees reported receiving in adequate mental health support and earning less than $2.00 an hour. i think it's pretty obviously, monthly, problematic to outsource the to, to countries. i mean, i have no problem with as are some of the countries that, that i'm for making ever

i asked a researcher at melanie mitchell. i think it is a game changer. you know, it's really enabled uh, a i systems to generate language to generate images and soon to generate videos in a way that it's never been able to do before. that really can be quite human, like the world of big tech took a similar view. all of a sudden, people were wondering if a child g p t powered version of microsoft being could be about to topple google's domini, ends in search. but this is about a whole lot more than a few tech companies trying to do each other. the way generative a i is integrated into us search engines will have a profound influence on how we access information. and by extension on our understanding of the world, is this big implications for who gets to decide? what is the future of a i, one of the things that we've seen with large language models i've tried to be, is there a lot of problems they can generate text, that's really impressive. in many ways it can be very convincing. it's very good quality text. but because it's trained on the data on the int