Did I claim anywhere that his early work wad bad?
Nope. The contrary as a matter of fact. But the facts are:
1) nothing worth talking about since the LSTMs
2) most of the reasons why he's been visible in the last 20 years is because of shit like this (JS harassing Ian Goodfellow about attribution in the middle of a technical presentation. Watch the face of Ian, pretty symptomatic of when AI folks have to interact with JS these days).