I think they've iterated on the design a bit over the years, but my primary criticism is that it can be hard to distinguish that an input is, in fact, an input. I think Google has since changed their design to have a filled background or border, which definitely helps - but this particular implementation doesn't seem to do that (except when the input is disabled, which sort of defeats the point).

There's also a lot of problems around how the placeholders and labels work - particularly in distinguishing them from each other.