they are used for RLHF, its training the AI. Machine learning in general requires a shitton of high quality labelled data. Labelling the data is usually very tedious, and often done via mechanical turk etc.
African workers arent able to write a unique 500 word story about whatever scenario you dream up in about a second in whatever language you choose.
African workers arent able to write a unique 500 word story about whatever scenario you dream up in about a second in whatever language you choose.