dats

The total effective tax-rate (national+municipal) in Japan for entry-level workers is under 20% unless you work a ridiculous amount of overtime, and even then it wont be much more. Unlike US jobs, most salaried positions in Japan pay overtime after the 40th hour.

I also work in Japan, and I've always been treated politely. Of course, there's the occasional politician who announces that all crime in Tokyo is committed by foreigners, but real people tend to be a lot more level-headed.

The only remarkable thing about the equity being asked for is the anti-dilution provision. I don't think he's suggesting you ask for liquidation preference or participating preference (which wouldn't make any sense).

Anti-dillution is there to prevent unscrupulous partners from buying tons of equity (issuing new shares) at absurdly low prices just to dilute you.

Have you seen this? http://uebb.cs.tu-berlin.de/~magr/pub/Transformers.en.html

It's the best intro to monad transformers I've come across.

BitTorrent. Ashwin worked for Goldman Sachs.

Well, Agarwal says the remaining is split on an "as converted" basis which implies non-participating preferred. So it seems right to me!

Yeah, the papers only discuss how they combine known algorithms[1]. They could be speeding up the process using any combination of these:

1. Using the order of the images to reduce matching from O(N^2) to something closer to O(N) since people are more likely to take pictures sequentially instead of randomly. I'm going to run some tests next time I boot into Windows.

2. Running keypoint detection as the images are uploaded and potentially slowing down the upload speed to make it appear more seamless.

3. New keypoint detection algorithm for constant-time speedup. SURF, for instance, is about 5x faster than SIFT.

4. Using GPUs. The keypoint detection and matching steps are perfect for running on parallel GPUs. A couple recent papers showed 10-15x speedup for keypoint detection and KD-tree construction and search on GPUs, compared to similarly priced CPUs.

Any other ideas?

[1] There's also a new paper called Skeletal graphs for efficient structure from motion that discusses a neat graph algorithm for the SfM step.

I think he means that it's a (Windows) plugin for your browser. Both the picture uploader and the viewer are Windows-only for now.

The really impressive thing about Photosynth is how far it has come in the last year. It used to take hours to weeks to create synths from hundreds of pictures. The processing time is dominated by

1. Keypoint detection (SIFT or a similar algorithm)

2. Keypoint matching for each pair of images with approximate nearest neighbor (ANN) search on kd-trees

3. Structure from motion to recover rotation and the relative location of photos

I've noticed intelligence is respected but conformity is respected much, much more.

From a recent article titled "Why Apple isn't Japanese"

"One notorious case in point involves Shuji Nakamura, the brilliant scientist who invented a revolutionary energy-saving blue-diode light source only to find himself mired in years of litigation as he struggled to extract royalty payments from the company that had profited from his invention. Nakamura ultimately abandoned Japan for California. Fasol recalls asking scientists at the University of Tokyo if they considered his departure a blow. " 'No, not at all,' they told me. 'It might be good to have someone more ordinary'." "

http://www.newsweek.com/id/73236/output/print

I know you'd rather be there in person, but here's the next best thing. Last year, Berkeley did a one-day Ruby on Rails course and put the videos online.

http://radlab.cs.berkeley.edu/wiki/Ruby_on_Rails_1-day_cours...

http://www.youtube.com/view_play_list?p=4910EFAA8600A920

EDIT: You mentioned getting your questions answered quickly. I'm constantly amazed at the number of smart and responsive people in IRC channels. You might want to give that a shot.

It's not a joke. From the latest 10-Q linked in the article:

Finite-lived intangibles include developed and core technology, carrier contracts, trade name, and subscribers.

This reminds me of Brian Lent. He sort of missed out on both Yahoo and Google. http://blogoscoped.com/archive/2008-01-22-n20.html. Edit: some choice quotes.

"When Filo asked Lent if he would like to join Yahoo! as employee No. 1, in order to keep the founders on their toes with his skepticism, he laughed. 'You couldn’t pay me enough money to work for a company called Yahoo!' he recalls saying at the time."

"In early 1996, Lent explains, “We all said, ’There will never be another Yahoo!’”"

Last year someone benchmarked a simplified raytracer in a few different languages--primarily to show off OCaml. Later, someone wrote the raytracer in 7 lines of K. http://www.nsl.com/k/ray/ray.k. The original comparison: http://www.ffconsultancy.com/languages/ray_tracer/index.html

I work in Tokyo, and I've noticed the only people who seem to have Facebook accounts are students who studied abroad and those who spend a lot of time socializing with foreigners.

Maybe they meant for it to be available in certain countries but erroneously included the US (for 20 minutes).

Yup. It's prevalent in Tokyo and seems to be very effective.

I think this only works for certain countries. It also applies only if you live outside the US for at least 330 days the first year, but those days don't have to be consecutive.

I'm saying something different though. If you have 2 projects, you won't be able to split your time as effectively so you may have 40% focus for each, and so on. As you take on more and more projects, the time it takes to get back into the groove increases. You are likely to spend a lot of time moving between projects and this will disrupt your overall focus.