Re:NPR suspends veteran editor as it grapples with his publi...
by LamaOfRuin on Wednesday April 17, 2024 @11:39AM
(#40066222)
None of these data sets are based on "the Internet". They are a specific subset of the internet, and the training and reinforcements are in no way neutral (because that's not a thing).
Also, the models embeds the biases of all the people who create content on the Internet. The designers biases come in to l look at with the prompt hacks put in place later to prevent the AI from expressing bad thoughts.