I found that quite interesting actually.
Implementing this kind of thing in an open source(!) browser may be the only way to deliver custom ads without violating the user's privacy.
That being said I still think it is a horrible idea...
Just as a little question, are all the ads pre-downloaded, or could someone theoretically observe which ads your browser downloaded and deduce things about your history therefrom?
The picture in the Techcrunch post is pretty clear. The browser talks with an adserver (step 5). And how could it get ads if it didn't do that, pick them from a static list compiled into the browser binary? Soon out of date and hardly interesting for advertisers.
GETs to the adserver combined with the in-browser selection algorithm leak details on the browsing history of the user (if he's got this ad he could be that kind of person and been there and there). Maybe not the worst of the spywares but about as concerning as any ad in a web page.
The picture in the Techcrunch post is pretty clear. The browser talks with an adserver (step 5). And how could it get ads if it didn't do that, pick them from a static list compiled into the browser binary? Soon out of date and hardly interesting for advertisers.
Based on what I've read in the actual implementation discussion in Bugzilla, the process actually consists of:
1. A Mozilla-controlled server hosts bundles of tiles, which are periodically updated.
2. The browser periodically downloads updated bundles of tiles from that server.
3. Using metadata in the bundles, the browser decides which ones to display.
A lot of discussion seems to have centered on ensuring the bundles and their metadata have enough overlap to ensure that even Mozilla's servers can't determine which sites in the browser history caused a particular tile to display.