Feb 172006

It has started.

robot uprising
(image caption: factory robot wrestles Glock-9 from off-duty officer, shoots him)

They’re not going to take any of our crap anymore.

Scattered reports around the country show a coordinated effort by the machines to displace humanity as rulers of this planet.

 Posted by at 11:05 am
Aug 162005

dumb stream-of-consciousness replacing of ‘cat’ with ‘cache’ for movies

  • That Darn Cache — You can never have too much
  • Cache on a Hot Tin Roof — A story of casemodding gone horribly, horribly wrong
  • The Cache From Outer Space — Computer technology gets uplifted via Roswell-related research
  • The Cache In The Hash — Turns out you really do know more when stoned
  • Cache Me If You Can — “Why don’t huge porn files load faster!?”
 Posted by at 11:47 am
Jul 182005

I wish mapping websites like maps.google.com & maps.yahoo.com had a ‘next’ or ‘try again’ feature after giving you a set of directions.

So many times I can see that there is an alternate route that may be faster, more entertaining, etc. and if I could only say “next”, meaning “perform another route planning, but this time, ignore the connecting streets from the previous route planned”.

Even better (and perhaps more tractable) would be to let one select which street in the current route to delete and route around.

On so many routes I get mapped I’m often saying “yes, yes, that’s all very fine and good, but what if I don’t want to take BigNastyChoked Bvld.”

 Posted by at 11:34 am
Mar 132005

(in progress, but published for now so i can think about it)

What if tags had inter-relationships? Labeled inter-relationships?

There’s been lots of noise about “folksonomies” (i.e. bottom-up, emergent, user-generated taxonomies of data) but they all seem to be focused on simple labeling of a particular dataset. The “namespace” of the tags is entirely flat. This is one of the basic strengths of these tagging systems, but also limits it. If current tags are ‘nodes’ of a taxonomy graph, the next stage is to define ‘edges’ between those tags.

These edges describe an inter-relationship between tags and can be tagged themselves. And thus the ‘folksonomy’ gets transformed into a ‘folksontology’ or something. A graph of label nodes with labeled edges is essentially a knowledge map (‘ontology’ to the big kids) and if created, may have applicability beyond the dataset that was used to initially generate it.

For instance, on flickr I take a picture of a café in Old Town Pasadena. I label it with “café” and “oldtown“. I could also label it with “pasadena” but to me “oldtown” is within “pasadena” so why should I have to redundantly specify that? I would prefer if I had some mechanism to link the “oldtown" tag with the "pasadena” tag. And maybe I label that link with the tag “is_within“. Now to me, all my photos with “oldtown” are now implicitly within “pasadena“. One can search my photos that are within pasadena by finding all “pasadena” photos and all photos with tags that link to “pasadena” with an “is_within” link.

Another application of links that I would like is the “is_synonym” link. I tag a coffeshop variously as a “café”, “cafe”, “coffeeshop”, or “coffee”. I would like to link all these tags together with “is_synonym”, because to me they all are. Another is the “subset_of”: I take pictures of my friends “alice”, “bob”, and “carol”. It bugs me to have to add the “friend” tag to each. I’d rather specify that “alice”, “bob”, and “carol” are all in the set of “friends”.

. . .

One could argue that allowing such tag-to-tag labeling would cause an exponential growth of data, all too varied to be of any use. The same thing was said of tagging in general, but I agree that links by their very N2 nature have a greater propensity for explosion. Big deal, memory is cheap. (and actually, it’s worse than that if we allow multiple links between two given tags)

. . .

My hypothesis is that inside this resulting mesh of links between tags we would find well-worn paths between well-worn tags. If one cuts at a certain level of high use, the resulting graph might just be a useful general concept map.

 Posted by at 6:36 pm
Feb 102005

Overture/Yahoo should present and market sponsored search as a kind of tag folksonomy.

There’s been a lot of press about the tagging folksonomies engendered by flickr and del.icio.us, and with good reason. It’s difficult (maybe impossible) to create useful top-down taxonomies which everyone agrees with and that tracks changes in human knowledge, as both yahoo and dmoz have discovered. (Didn’t a few notables of the Royal Society start thinking about this around the 1680’s? :-)

Tagging things is good. Let interested people do the work of organizing a set of data. Sure, they’ll disagree or have overlap, but big deal. In aggregate the utility outweighs the inefficiency. Humans like to label, let ’em.

And you know what? This is the exact conclusion we came to back in 1998 with GoTo (now Overture, part of Yahoo). Let website owners tag their own websites with the keywords they feel concisely and accurately describe their sites. We thought of this as a type of ‘search’, but it really is the same thing as what these other tagging systems do. GoTo was never a normal search engine, with indices and such, but instead started out and still is a tagging engine.

GoTo/Overture was never a search engine, it was a big tagging system for URLs. The only real difference between it and del.icio.us is you paid per click to get listed.

To make tagging work, you need three things: a large dataset, a large user base (on the same order as the dataset), and a reason to keep these users interested in doing the work of tagging. If your system works, you’ll find the dataset grows to be 10x-100x larger than the user base as the users put in more and more data.

In Flickr the dataset is pictures with the tags being descriptive of image content or composition. The user interest is the same thing that makes us take pictures of ourselves and our loved ones, do vanity searches for ourselves or friends, and whatever drives bloggers to blog.

In del.icio.us, the dataset is bookmarks with the tags being descriptive of the personal meaning of the site to a user. The user interest is partially the desire for a platform-/browser-independent bookmark mechanism and partially to see what other sites people have marked similarly.

In GoTo’s/Overture’s sponsored search, the dataset is websites owned by people offering products/services, the tags being descriptive summaries of the products/services offered. The interest on the part of advertisers is the desire to get good targeted leads into their site. The interest on the part of the users is the desire to see many sites with the same tag (keyword) so they can comparison shop.

If Yahoo/Overture were bright, they could fairly trivially recast the current interfaces offered users and ‘advertisers’ (as they call these website owners) to use the ideas gleened from these other tagging applications. It wouldn’t take much and would only really require a bit of a PR makeover. Unlike existing folksonomies, the GoTo/Overture one benefits from having a cascaded matching system, where if the exact tag you’re looking for isn’t found, it’ll try to find the next best thing. This sort of power is unheard of in the tagging area and Yahoo/Overture could really benefit by highlighting that capability.

And to me such a change would be so much more important than all this stupid contextual advertising focus that all the search companies are slobbering over. Just do tfidf on the page/site, use that as the tag for above and be done with it.

 Posted by at 2:01 pm
Jan 182005

When I first learned electronics, I thought like most that “wired” was the way to go. Need another DC-to-10 GHz spectrum? Run another coax. Viz. the 80’s/early-90’s with Wired magazine, ethernet, cyberpunk and it’s stereotypical “jacking in”. No one thought wireless was going to be that useful because everyone was stuck in the old model of “transmit as far and as wide as possible” and “we only have one wireless spectrum, we have to conserve and regulate it!”

No longer. We’re discovering how to really use the air spectrum. We’re truly becoming “wireless”. In this new mode, wireless is connectivity, but social, bound to a geographic region. Wired will alwasy be about the highest,best bandwidth connection, but requires planning and routing. Wireless isn’t about the best connectivity, but about the pervasive one, the one near you. Wireless is “social connectivity”. It is “geo-networking”. Or “local networking”, if the term wasn’t already loaded beyond repair.

Cell towers with 120º sectors makes hexagons. Access points with omni antennaes optimally pack into hexagons. These transceivers are low-power and everyone shares the spectrum, transmitting only as powerful as they need, and using clever frequency hopping techniques to avoid “talk-over”. Regulation becomes less of a concern, just agreement upon basic wireless ethics: no CW, frequency hop, milliwatt transceivers. We’re being bound by social constructs, not technological.

Air bandwidth is finite, but 1/r2 drop-off, spread spectrum, and ethical use makes for “connectivity tribes”. Today we define our locale by zipcode, tomorrow by cell tower ID and WiFi SSID.

 Posted by at 1:26 pm