This will make the ratings more accurate...

Ratings are a measure of playing strength relative to players in the same rating pool. Your chess.com rating measures your playing strength relative to other chess.com members. If you want to know your true playing strength relative to other USCF players, you need to play in USCF rated events.

Glicko is superior to Elo.

My USCF is higher than my live rating here, but comparisons of this sort are absurd. Each rating is valid within the pool of players in which it was earned, not outside.

 I couldn't agree more.

Evidently you fail to comprehend that you are advocating a return to Elo ratings after a site has adopted the much superior Glicko system. I understand your proposal, and disagree that it will solve the alleged problem. Rather, it will create other problems that you have failed to imagine (see "The Glicko System" for an explanation).

Also see "Math People Only" on this site.

Your proposal is incorrect on one point as well: Arpad Elo himself designed the system with an assumption that a 400 point difference would predict that the higher rated player would score 92%. 100% with a mere 300 point difference is not a reasonable assumption, nor is it borne in practice. I once played a four game match where the rating difference was a tad over 500. After three games, the match concluded 2 1/2 - 1/2--a mere 83% score for the superior player.

Ratings as a system cant be inaccurate. Start everyone at 4000 and a 4500 player will beat a 4000 player almost every time. 

The ratings can't be compared to any other system, but they aren't inaccurate, or inflated (inflated from what?).

People can individually have inaccurate ratings however. This happens often when a player makes a new account, a 1400 live player may win three games against 1200 players and receive a 1700 rating, obviously this is inaccurate, and within a few more games the rating should adjust itself accordingly.

Least of all you it seems.

 Try looking at the percentile in the "today's rank" box on your stats page.  If you're at, say, the 80% percentile, then you can go to the other site or organization and see what the players at that percentage ranking are rated.  Then you can know that, for example, 1500 on Chess.com = 1350 USCF/FIDE or whatever (Those probably aren't the real numbers - I'm just saying that the percentage won't lie to you, 'cause it's always based on 100% of the players).

 FIDE/USCF also uses rating floors, which I'm sure some people would enjoy.  Most people hate to see their rating go down, so a nice rewards-based system like what is used in prize tournaments would help out in that regard.  You'd just have to live with the fact that when you beat everyone at your floor number, you're not eligible for a prize on that floor anymore.

http://en.wikipedia.org/wiki/Elo_rating_system

Obvious troll/stubborn git/math fail is obvious.

Though I wonder (though I'm sure it's been suggested before), would it be possible to pin rated players' ratings on chess.com to their real-life rating? After all, chess.com already allows one to display the master titles once one's identity is verified. If this could be done for players' ratings as well, then it would provide a "benchmark" against which the rest of chess.com would play. This should (I think) end up making the entire pool of chess.com ratings closer to real-life ratings; because once you have a few reference points between the two pools everything else should fall into place.

Downside of this though, is that rated players' games on chess.com will no longer affect their rating...basically they're playing unrated games all the time. That could cause a drop in the effort they put into their games...which in turn would cause inaccuracies again.