[heydenberk]

I've always wanted to explore the word2vec space to find the most sparsely populated region, then (assuming you can train some sort of reverse vec2word model) coin that word that the English language was apparently sorely lacking.

[Der_Einzige]

It's too bad that even the best dimensionality reduction algorithms still really suck. Visualizing and inspecting latent spaces is super sweet and eventually I think it will become extremely useful and common...