3 ms·
There's an interesting variant of the Turing test here: develop an AI sufficiently intelligent to distinguish human content from AI generated content. I might
by fny 5y ago
There's an interesting variant of the Turing test here: develop an AI sufficiently intelligent to distinguish human content from AI generated content.
I might be wrong, but I think this might be a more difficult task than generating convincing dialogue: its very easy to generate text that statistically resembles human writing, and generators trained on certain topics (i.e. some science niche) might be impossible to flag.
- zmmmmm 5y agoyou approximately just described adversarial networks (GANs). One of their training metrics is that it is how difficult it is for the "other" network (the adversary) to train itself to distinguish the real from generated.