This is complete rubbish. Zero evaluation compared to any sane baseline, zero thinking about representation (predicting raw price?!). I'll just say it, ML was frankly better when fewer people knew it and there was less money in claiming to be an expert.
The DL4J folks are a joke. They make a ton of outrageous claims then and they tell you you're getting it all wrong when you try to call them on it. I know they need to justify their nonsense so SkyMind can sucker someone into an acquisition, but they're not fooling anyone. At best they can claim it's as fast as what one person built in their spare time....