I'd rather have the additional information they present, and to see what goes into their assumptions in analyzing the results, than for them to just come out and declare one set of numbers as being "correct". At least this way we can see the possible range of outcomes. The fact is no one really knows how best to define likely voter this year. Why should a pollster try to pretend that know any better?
My problem is when they start dipping into their samples to find sub-samples like Harris did with this one. A poll of 250 people is meaningless. And the jist of the article is "these sub-samples are statistically meaningless, but here they are anyway, and here are some conclusions you can draw from them. It's misleading. The only worthwhile number is the main one. 1,016 adults, Bush ahead 48-46. But that's not news, so they have to find sub-samples to try and "sex it up."