Thanks Dzmitry, I trust your judgement and intuition on the effectiveness of the GAN more than my…
Yoav Goldberg

Just to substantiate your point on “spikiness”, the method described (based on the description in your post) wouldn’t work for something like dialog generation, where one context will potentially have a large number of acceptable responses (perhaps even at each step of token generation).

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.