As someone who has lied to people about my progress on developing games before, the comments from Sean sound exactly what I'd say if I was asked about "x" thing. , but it makes me suspicious when combined with the fact that no one has played this game yet.
First off, the demo he showed was probably in-engine (although that engine is barely formed), but looked like it could have been totally automated (the "player" not being controlled by a human). I'm almost certain those animals scurrying around were put there by hand; not by algorithm. In fact, I'm pretty sure none of what was shown was actually auto-generated, but rather a portrayal of what the game "could be."
I think these guys got an idea for auto-generating their universe, but haven't figured out how to do so at all in any playable game-shaped way yet. All they've given people is the same "it's easy to create this world when it's all generated for you" spiel, talked about laying a perlin map to create random geometry, and assumingly have used random seeds to generate forms of everything else in the game. These are words any novice game dev can spit out, regardless of how true they are. It's easy to do these things on a small scale, but when you're talking about endless planets with their own flora and fauna there is a lot more work that needs to be done.
As I think Jeff pointed out, you need to keep things varied while still keeping them sensible. Notably, making sure Dinosaur 12321 doesn't look too similar to Dinosaur 24532, while still not being an incomprehensible monstrosity. If you look at something like Starbound, for instance, you'll see their variations in randomly-generated monsters are extremely un-varied and extremely boring. Designing and implementing features that get around this is extremely fucking hard, and requires much more man-hours than they'd have you believe. These guys can't realistically auto-generate walking animations for 2, 3, 4, 5, 6, etc.-legged animals of all different sizes and proportions, each with different bodies, arms, heads, (personalities?), and overall sizes and shapes. That shit is hard, and that's just one piece. Now take that and apply it to the world at large, and you've got a thousand new problems that need solving... by hand, not by algorithm.
Basically, I'm extremely, extremely skeptical of this project. I'm hoping it turns out to be the real deal, but right now I'm really not left with much choice but to think of this guy as a really toned-down Peter Molyneux.
edit: bolded, italicized, and underlined the most important part of the first paragraph for clarification.