3 ms·
This is strangely similar to generative adversarial training in neural nets. They are composed of a generator and a discriminator. The generator has the purpose
by visarga 10y ago
This is strangely similar to generative adversarial training in neural nets. They are composed of a generator and a discriminator. The generator has the purpose to generate examples that fool the discriminator and vice-versa. They improve together, iteratively. It's a smart approach but has stability problems in training, because both the generator and the discriminator have to be balanced.
Maybe they could try to apply this kind of technique to improve their auto-generated articles to the point where they can't be distinguished from those created by humans, and at the same time, fool the FB bots which filter the feeds.