terra_t

Wow, this is the first site I've seen to use frames in about 10 years!

Its generally true, but I've developed systems that go sideways and fork into multiple versions, so the data structures get more complicated.

I said it in fewer words and didn't add any adsense junk.

I used to have this problem. Then I realized that if I want to really copy a file, I type

$cp file_from file_to

and that

$ln -s link_from link_to

has a very similar effect to the cp command above. I haven't messed this up ever since.

Unless you're targeting a specialized market where you're really in demand, I'd have nothing to do with "web freelancing".

A few years ago I worked at a job shop that did web sites for small businesses, and I was talking with my accountant about getting my taxes done and he asked me how much we'd charge for a web site. I told him it would be around $2000... That would include a CMS install, original template, and some SEO. A pretty fair price for the time of the talented people it takes, plus the sales overhead. He was shocked. Although just getting another 10-15 clients a year would have paid for his site quickly, he was hoping he could get one for more like $50.

4 out of 5 small biz clients will let you make a tiny profit, but 1 out of 5 is a client from hell who'll balloon a $10k fixed price project to something that costs you $30k and wipes out the product you made from the other 4.

Not for me. I make web sites for my own account.

I've got a rule that "If it's not impossible, it's not worth doing." If most people think that what you're doing is impossible, you've got less competition.

If it makes $30k of revenue a year and you don't think that revenue stream is about to stop, this is a good investment. Do it.

Domain age counts for something, as does the rich mat of links that a long-existing site has.

On the other hand, the price looks high for a site that gets that much traffic. I can pretty consistently create SEO-oriented sites that get that much traffic & revenue in 1-2 yrs with an investment of my time that's more like $10-15k. (Working pretty intensely for a month, and mostly waiting for the link network to mature)

(Of course, I've gotten tired of that and now I'm trying to break into making sites that are 10-100x bigger than that)

Changing a successful site is always risky. There are two risks. (1) is that you might go from something that works to something that doesn't work, and (2) in theory the ideal way to drive up your traffic would be to make small incremental improvements, watch your results, then make more changes. Google's very happy for you to do that if you're using AdWords, but they don't like you doing that in SEO, so there are things built into the system that can (sometimes) zap your ratings if you try to revamp an old site.

Personally I can think of better things to do if I had $100k sitting around, but if you like this site you should think about trying to negotiate the price down. I think a multiple of 8 times earnings is fair, so something like $25k more like it.

Of course, his operating costs are low, so like your average domainer, he can probably sit around a long time to find the guy who'll pay too much.

I've often been unhappy with off-the-shelf monitoring systems, because every major site I've worked on has had some unique kind of problem.

For instance, there was one site where user abuse was the real problem... Bad enough that I built something that detected possible abusive behavior and would beep my pager.

For another project we had about 20 geographically distributed mirror sites, and we had to monitor network connections to all of them and make sure they were all alive and staying synchronized.

Right now I've got a site where the caching system screws up periodically and then I start getting 500 errors. Sooner or later I'm going to really fix the problem, in the short term what I really need is something that gets in my face whenever the 500 error rate spikes.

This definitely sweetens the deal. AMZN's monitoring facilities have been expensive and not really all that good.

Halos around your text are quite important if you're trying to draw titles on television, particularly the old NTSC television. You see them all the time in video games too.

Yes, it is good, but not good enough for many applications. You're also left with the issue that one kind of "apple" is more common than the other kind of "apple" so the baseline accuracy of something that always assumes it's one kind of apple might be surprisingly good.

That said, text-to-speech is a system where it's important to do disambiguation of a particular set of words. For instead,

"I read the news today, oh boy", "read" sounds like "red"

"I read the news every day", "read" sounds like "reed"

You need to be able to disambiguate the word sense to be able to correctly read the world "read". There are maybe 20 or so very common words that are like this, so a modest amount of work in this area would be part of a good TTS system.

Sure, but the fact is that nobody actually needs a "word-sense disambiguator", they need a search system with better accuracy, or a classifier with better accuracy or an information extraction system that turns text into facts.

Many areas in NLP are like this. You can get 92% accuracy in a few hours of work, and then you can get 93% after a week or work, and then you can write a whole PhD thesis about how you got 94% accuracy.

To a certain extent, there are approaches, such as the Support Vector Machine that are "unreasonably effective" but once you get past that, you often have to confront issues that everybody wants to sweep under the rug to make a real breakthrough.

For instance, there was that NELL paper that came out a few months ago; NELL extracted facts from text but it had no idea that "Barack Obama is the President of the United States" was true in 2010, and that "Richard Nixon is the President of the United States" was true in 1972. If you can't handle the fact that different people believe different things and that statements have expiration dates, no wonder you can only get 70% accuracy in IX

92% accuracy, unfortunately, isn't good enough.

Bag-of-words models perform pretty well at classification and search, and the main thing you need to improve search is to boost scores when words are close together.

You might think you could improve performance by using semantically better defined features, but even 92% accuracy adds enough noise to foil your plans.

It's a big problem in A.I. systems that have multiple stages. You might have 5 steps in a chain which are each 90% accurate, but put them together and you've got a system that sucks. Ultimately there's a need for a holistic approach that can use higher-level information to fix mistakes and ambiguities at the lower levels.

Heck, I'd like to see Facebook credits be more flexible than they are now. In particular, I'd like to be able to use FB credits to pay people for work... I've got a suspicion that this would undercut mechanical Turk.

I guess the main reason they won't let webmasters pay out FB credits is that the first app people would build with it would be a gambling app... Too bad.

I dunno. I took second semester Chemistry when I was a senior in college... I never showed up at class except to take the weekly quiz. The professor was not at all offended that I got the highest score in the class.

I've long believed that standardized tests are a better measure of what people know than grades.

I think we're going to see a lot of papers like this. In a quantum gravity universe, we know there's no singularity at the big bang. Therefore, scientists are going to be looking through ways that we can see through to the other side of the big bang.

Even if their particular universe model isn't correct, there must be other models in which similar phenomena could occur. Interesting stuff.

I wonder if Web 3.0 will miss Silicon Valley.

I've found that the less technical and "webby" people are, the more they understand Web 3.0

A lot of that is that, in Web 3.0, perhaps 70% of what's necessary for Web 2.0 is superfluous... Perhaps nice to have, but frankly, the next generation systems need a community the same way that "The Terminator" needs people.

19 upvotes but no comments?

image search that works.

Look for something on Google image search and typically the same image will show up in 25% of the first page of results and precision can be anywhere between 10-75%.

Image search is still so bad today that people don't even believe it can be made better. Defined in a certain narrow way, the Ookaboo API is > 99% accurate, however, there are common cases (categorizable) where its view of reality doesn't match other people's. For instance, Ookaboo might show you a picture of a person that is associated with a place instead of the place itself, or it might show you a picture of something that an artist made rather than show you a picture of the artist. We've got answers to that under development

http://ookaboo.com/

did web and API simultaneously -- mobile will happen when revenue or investment comes in and I can hire a team to do it.

The API has been a tough sell, however, because it's doing something entirely new. It's also got the issue that I don't know how to monetize it, whereas the web can monetize just fine. On the other hand, landing just one major API user can send me enough traffic that my competitors won't know what hit them...

Oh yeah, kids who get "You have to be a doctor" parenting have a 50-50 or so chance of succeeding when at least one "doctor" is a parent. Some have brilliant careers, and others screw up totally and end up destitute.

Damn... I had a boss that wrote significant parts of a product and we could always tell that he wrote part of it because he'd repeat a 30-line block of code 7 times, rather than writing a subroutine.

There are cases where we all violate the DRY principle... Sometimes it makes sense to subroutinize something the fourth or fifth time rather than the second time. But overall, DRY is good.

I've got a friend who's the youngest deacon in the Catholic Church in the diocese that includes Syracuse. He practices what he preaches, so he's had 11 children.

He tells me that each child has an individual personality once they hit the ground. Some kids are easy to discipline, and other ones aren't. Overall, parents of ADD kids use harsher discipline than other parents, however, it appears that they're driven to do that by their children's behavior, not that harsh discipline causes ADD.

My 8-yr-old can sometimes be the best behaved and charming child and sometimes he can be a total embarrassment. We've had arguments in public escalate to having the cops show up and having child protective services knock on our door, so I know that harsh discipline isn't a viable way to get perfect discipline in public.

How about a service where you can pay people to do it for you, maybe, say, 10 cents an image?

I've seen signup success rates between 30-97% with conventional login systems. There are two things that appeal to me about Facebook:

(i) people take their Facebook identities seriously, so they're less likely to be griefers, and (ii) the like mechanism to push back updates about items that have changed

The person who succeeds best in our culture is the one with Narcissistic Personality Disorder. NPD people are great at telling others what they want to hear, and, unfortunately, that gets you further than any other skill.

If there's a reason that our civilization will go the way of the dinosaurs, this may be it.

"Brush your teeth" never worked me. I'd get hungry and eat anyway. Two servings of whey protein powder a day, on the other hand, completely nukes my interest in snacks.

As for HN and such, I find my interest waxes and wanes. I definitely have days where it's a big distraction, particularly when I'm waiting 10 seconds for something to compile and come back ten minutes later. Then there are days when I'm too focused on coding, marketing or whatever I'm doing for it to be a big distraction.

This is why I'm likely to use Facebook Connect. Sure, some people hate Facebook (maybe I lose 20% of sign-ups) but giving people 30 million choices causes them to freeze up (lose 80% of sign-ups)