|
There are two layers, the training data and some filter or censorship mechanism. I am sure also "controversial" material is used for the training, and maybe some non-US player once forgets, or intentionally leaves out, that filter layer that forces the model to adhere to the rules. It is not a superior intelligence, but a big algorithm that could include the actual facts and come up with responses that do not match with what mainstream moral tells. Like scientific facts that children can enjoy sexuality, that there have been cases were children initiated sexual contacts with adults, and that the children in most cases were not harmed by this. And more. But who knows, since the sexual taboo is mostly driven by US/UK, perhaps one day China releases a free model to disrupt the status quo in western dominated countries? -- Méiyǒu nánhái de shēnghuó shì kěnéng de, dàn háo wú yìyì. ![]() |