New on my site: sledding photos and movies,
Several months ago, I saw a TV documentary that showed the excavation of seahenge, a mostly-submerged wooden circle with a huge oak tree buried head-first in the center. (It wasn't originally installed underwater, but that was 4050 years ago.) To understand it better, the researchers also tried to build their own, only using tools that would have been available at the time. It was hard work, and they gained a lot of respect for the people who built the original.
The plans were to preserve the unearthed pieces in a museum, but doing it well would be expensive, and so it looks like they're going to rebury everything, since that worked quite well for 4050 years.
John over at Genehack mused about training a rule-based system to recognize porn, and Lyn quoted him in Medley (but without a stable URL, alas). I already replied by email, but I thought I should post my reply here for people who read their sites, in case they wander over.
Eventually, after the thing had run for awhile, the results (this picture is porn, this is art, this is crap) would be fed into some sort of expert system that could then generalize rules for discerning whether a particular image was pornographic or not.
Unfortunately, it's not that simple. If you want a rule-based system that has rules at nice abstract levels, you have to feed it nice abstract data or give it a way to extract the high level features itself. Computer vision research is progressing, but it's not there yet. Without this, you'd be asking it to form rules like "If there are some pinkish pixels in this area and some pinkish pixels over there, this is porn.".
Humans have systems for extracting complex high level information from images. We can look at a picture and know whether it contains people or people-like things and can easily recognize their poses, expressions, and situations. While it's possible for a system to learn everything from the pixels up (we are examples of this), it's very hard and is not really suited to a rule-based approach (nor to our relatively wimpy computers).
David Forsyth, an associate professor at Berkeley, has published several papers on his research into automatically detecting naked people in images. There's also been lots of work on spotting and recognizing facial expressions and body poses, but I haven't seen anything reliable.
The best thing about having a digital camera is that I take a lot more pictures than I did when I used film. The worst thing is that I'm way behind in organizing my photos. I have about 1100 pictures that I haven't entered into my big list yet, and that means I probably won't be able to find them when I want to. I'd also wanted to put many of them online, and that plan is fading towards pipe dream.
I'm on the ball today though, coming through with same-day photos of our neighborhood transformer being replaced.
I also realized today that digital cameras are less bad for the environment than film cameras are. They use more electricity, but avoid all the nasty chemicals involved in making, developing, and printing film.
Thanks, Beth, for telling me that the archives page was missing the most recent months. This has happened before, since I had been making that page by hand. Now it's automated and will never happen again. Yay.
I like the portability of my laptop, but it's hard to write big documents on a small screen. I want to spread everything out so I can see it at once, but instead I have a tiny window onto my work. It's as awkward and frustrating as working on a small desk.
Even the NY Times is annoying me with its coverage of the California power crisis. They, like many others, keep referring to California's deregulated power market, but California doesn't have a deregulated power market. Free market electricity is getting a bad rap without justification. The NYT does explain, in many of their articles, how the market is not actually deregulated, but they continue to call it that.
Under California's oddly crafted deregulation formula, the utilities must buy most of their power on a state-run wholesale market and pay whatever the market will bear. However, the prices they can charge their customers are capped until March of next year.
This half-assed measure is what's causing problems, and anyone who has taken a single economics class would have been able to predict our current crisis. Adam Costello clearly explains with text and diagrams exactly how this moronic scheme causes power shortages.
I just thought of a really geeky joke.
Why will vi never win the text editor wars?
Because Emacs users have a universal argument!
Ha ha ha!
Sorry. I don't really expect anyone to get it.
Control-U in Emacs is normally bound to the universal-argument function, which veteran Emacs users use all the time. It's an easy way to pass numbers to other interactive functions, which is extremely handy.
Section 4.0.4 of the XEmacs FAQ contains a single sentence:
"Obsolete question, left blank to avoid renumbering.".
That's reasonable, but you know they were giggling when they did it.
(4.0.4: Not found)
Several months ago, back in Edinburgh, I heard that some people had tried to destroy some nearby genetically modified crops as a protest and in the interests of public safety. They failed to destroy the crops, but they had some success with the publicity, as evidenced by the fact that I heard about it, even though it had happened six months before I arrived.
The perpetrators were found guilty of vandalism, having done a whopping £1.50 (US$2.25) worth of damage, but the amount of publicity they've gotten is worth far more than the fine cost them.
I just found out that I know one of the culprits. Go James! James is the one most prominently featured in the group photo in BBC's story about the trial results.
In other news, scientists stop light. How weird is that?
I have upcoming interviews at SRI and SGI. Maybe I should follow this acronymic theme and send my resume to SAI, SBI, SCI, and all the rest. Three of the first five are software companies, and a fourth is in electronics. I guess the high tech companies grabbed the short domains before the old industries clued in.
I've been making and playing around with new and strange devil stick variants. Back in June 1997, I made my first devil's triangle. Last year I finally started playing with handsticks connected by a string, using the string as a manipulator. This turns out to be fun and interesting, and, as with the triangle, I suck at it because I haven't practiced much yet.
Yesterday, I made a three-sectioned handstick, basically three handsticks connected in series with flexible universal joints (eye screws and zip ties in this version). I've only fiddled with it for an hour so far, but already I'm blown away. I had some ideas of what it might be good for, based on my experience with the string, but actually trying it out has let me glimpse a whole world of possibilities that hadn't occurred to me. It's going to take me years to explore them and to get good at using this new tool. This is like making the first violin and thinking, "This isn't much like a guitar after all, but it's got potential.".
As an added (planned) bonus, I can use one more connector to turn it into a rigid triangle. It's lighter than my other devil's triangles, which makes it more challenging, but it's nice to have the versatility. It can also be used as a manipulator in that form, which has potential I won't be skilled enough to explore for at least a few years.
Making a new tool is always fun, but having it turn out to be far more than I first envisioned is exciting. I'll post some pictures soon.
I just saw The Imposters. It's hilarious slapstick and has some brilliant moments. Go rent it.
My list of upcoming interviews is now SGI, SRI, and NASA. I wish my resume could drop acronyms like that.
A while back, I wrote a script that could help me play Word Sprout. Actually, that would be cheating. The real point was so I could check once the game was over to see whether I could have done something better. Not a very positive thing to do, but it's fun and maybe I'll learn something. Tonight I stuck a form on the Word Sprout page so y'all can play with it.
Yesterday was a great day.
Three weeks ago, I did something wonderful. I ignored a bunch of reasons why I shouldn't, and I started acrobatics classes at the San Francisco School of Circus Arts. This is something I've always wanted to do. Classes are always lots of fun and hard work. Even just throwing myself at the ground and missing would be great fun, but it's extremely satisfying to be improving.
Yesterday, I did a cartwheel and it wasn't a complete train wreck. And then I did it again and again and it kept working. I couldn't do it at all last week. I did other fun things too, but nothing else so surprising. Then I got a tuna sandwich (my favorite) from JJ&F, who make the best tuna sandwiches in the world, and ate it while sitting in the sun, reading a book.
When I finished eating, I went to Stanford and heard a great talk by Tom Mitchell on co-learning, a nifty machine learning technique for training classifiers with mostly unlabeled data (like, say, all the web pages in the world). He even signed my copy of his book. Then I went home, picked up some food, ditched my car, and biked over to my friend Bill's place to read Macbeth with some friends. We took a brief intermission for dinner, and because I'd thought ahead and picked up what I wanted at home, I had exactly what I wanted, even though everyone else was having something different. In the scattered spare hours throughout the day, I wrote some timesheet software to help out a local high school robotics team.
I didn't do much today.
Some Netscape settings only take effect when I restart it. I restarted it today, and I'd completely forgotten that a few days ago I changed the Waiting-For-Godot cursor from a watch to Gumby.