This is even more impressive: http://twitter.com/#!/selfrefer/status/3128391843 – A self-referential tweet that does not use a URL shortener.
It was in response to a challenge I tweeted about two years ago. There's an explanation of how it was done here: http://www.spinellis.gr/blog/20090805/
This gist of Diomidis's approach was to estimate how much tweet IDs increase in a fixed interval of time. He estimated that between two tweets sent in rapid succession (using a curl script) there were around 120 other tweets. So if the first tweet had ID 1,000,000, he could expect the ID of the next to be around 1,000,120.
Once he figured this out, he wrote a simple script that posted tweets of the form
Self referential tweet http://twitter.com/SelfRefer/status/
N was the ID assigned to the previous tweet.
If the most recently posted tweet's ID wasn't N+120 it was deleted. Once the script posted one that was self-referential, it stopped.