Hacker News Reader

That output doesn't look good, especially compared with the Obama char-rnn output: https://medium.com/@samim/obama-rnn-machine-generated-politi...

Perhaps the Trump corpus could be fed into char-rnn for comparison and figure out why the output is so nonsensical?

View on HN

I wouldn't say it's that bad. Obviously, training set is very small, much smaller than Obama-RNN's one. Hence all the repetitiveness, it just doesn't have enough examples of what more could be said. But I would say that it has even more sentences that actually make sense than Obama-RNN's output, probably because of average sentence being shorter.

A Markov Chain is only as good as its input. It doesn't have much to work with unfortunately.

It's not a Markov Chain.

Well that killed my joke pretty effectively. Drat.

Reply on news.ycombinator.com