Training the machines that will take your job
A growing number of experienced journalists, writers, editors, content creators and academics are working from home, usually with better rates per hour than other employers. Who for? AI companies.
Their job? Writing. However, what they write is not meant to be consumed by humans, but by machines. Their task is to answer hypothetical questions, to what would be prompts sent by human beings like them, to AI assistants. The idea is to provide a generative assistant with an example of what a good response is, a quality response, well documented and referenced, so that it can be trained with it and learn to generate similar replies.
Many people think that what these companies are doing is building a huge library of answers to give to everyone who asks questions. Instead, the answers are used for the models to learn and take as an example, so they can generate their own responses. Even if the answer these people gave was outstanding, it will never be reproduced by the algorithm, nor will it be recognized, because it will be hidden behind layers and layers of additional processing. As the Gospel says, it is not a matter of providing a man with fish, but of teaching him to fish.
Companies do this because the data set with which to train their algorithms, while huge, is finite. Some of the repositories compiled by…
