Hmmm, "offshoring" brings up Scoble's site as #5, even though he says "I'm no expert in offshoring." In that respect, I hope this "neural net" thing, although intentionally obscure in its description during the video, is able to parse the language of the page itself similar to how a human would, and understand to a degree what concepts are being presented from the site. Then rank links accordingly. It's one thing to do as Google does and simply rank search terms higher based on how many people reference them. That's easy, just counting references. And then you're suceptible to people exploiting it, as with the "miserable failure" bug. But to actually parse the thought on the page, that's hard, and that's where it will become amazing. Just as these guys are saying, that will weed out the splogs, and offer much better specificity.
Another powerful thing with a true AI type of neural net: when it sees the title "random thoughts", it should toss that out as a "noise concept" (for lack of a better term), and look deeper into the page for more of the meat.
I'll close with a little bug report, an easy thing to fix: with IE6 when you bring up the Settings screen, then show the Search Builder DIV, you've got the bug with an underlying SELECT (named setlang in this case) showing up over the popup DIV:
I had this problem on one of my pages a couple years ago. In the file common.js if you put this in the function toggleQB(), it should take care of the issue:
sl=document.getElementById('setlang');
if(sl)sl.style.visibility=show?'hidden':'visible';
Best of luck with this groovy AI project, guys!
-Lorin Thwaits