It's maybe way premature to know if it can actually work, but at least it is trying: Findory Blogory is a 'personalized weblog reader' that tracks what you click and uses that to decide what to show you next time you use it.

Among the perks:

• no login -- yes, absolutely, and I really don't know why more sites can't do this; FB drops a cookie and that is your handle, you just go there, it knows it's you, and that's all it needs to know. The only reason to login is if you want to hit the site from more than one computer and share the same profile, but hey, it's me so if it truly finds my interests, it would be me no matter where I go, right?
• categories are always really hokey, and FB is no exception, but you can enter keywords and when you use the search results, those will be the links it uses to hone your preferences.
• simple simple simple, it does what it does and that's all, no bogus ASN and no trying to lure you away to sponsored news, just the facts, just what you came in for, just text. It is two columns, which is going to frustrate blind users and makes it awkward on small-display devices but at least it's fast.

On the downside ...

• am I really so monotonic and shallow that a robot tracking a single handle will anneal to a nuances of a profile that makes sense? True, given a neural network with interior nodes, the engine could auto-deduce composite nodes that would become abstractions for unnameable 'categories' (the Red Riding Hood Example for those who know Backprop) but when I log in, I still get the same old same old taxonomy that just doesn't speak my language. That leaves me with only that first column of the first box as likely to be interesting, but hey, it's early.
• the search is primative; if it has a syntax, the rules are not disclosed -- as Dave said last night, it's good to have a simple control panel, but it should lead me to more effective use of the device -- with this one, it seems WYSIAYG and not even so much as stem* words or OR; the default is a full-page scoped AND.
  
  I can relate to needing to keep it simple, but if someone knows regex, why are we so hesitant to let them use it (the /pattern/ syntax is a pretty dead give away). Perhaps if search results count above a certain threshold, it might suggest ways to reduce ambiguity.
• ambiguity ... will it know when I look for kernels that I'm really after agricultural news?

Nonetheless, I'm not complaining, I'm just thinking to myself of the scale of their challenge and I know I'm going to give it a run to see if just maybe this might not be a good way to find that Dark Matter in the blogosphere, like having a Blogdex that learns my priorities. It might end up like Bayesian spamfilters and eventually tip itself into ludicrous chaos, but then again, maybe it won't.

Which ever it does, it's comforting to see people out there who are working on the problem. And for the bottom line, Does it work? I can only say that some search keys that should return hits have such sparse results it can only mean the collection of blogs is relatively tiny (blog software should offer to list people in these sorts of indexes as part of the install, or at minimum, FB should troll weblogs.com) and while I didn't find anything truly outside the box and worth pumping through my own aggregator, I did find ...

• Ripley -- talk about being "outside the box" ... ok, the litter box ...
• Victoria's true secret
• the quasi-useful and sadly no-Oggs webpage-playlist generator WebJay (and the associated bookmarklet from hublog); you give it a page of links, it generates a RealPlayer stream. This one found, quite ironically second hand via Stephen Downes ... who got it from Seb.
• MP3 Blogs List unsorted, uncategorized and unabashed but nonetheless a starting point for further siftings and good fodder for somebody's robot.

No great revelations, but hey, it's only after a half-hour's training the thing.