All this advancement in text generation - but it you want to use it to represent sentences or documents in a semantic space you still have to use horrifically b

All this advancement in text generation - but it you want to use it to represent sentences or documents in a semantic space you still have to use horrifically bad techniques like average or max pooling. When is someone going to advance pooling techniques? We desperately need improvement!