2,303 karma · joined May 9, 2009
http://www.tudou.com/programs/view/XH5W4vffBbY
Sorry that the first 10 seconds are shady japanese yogurt commercials, last time I looked the videos were $300 on ebay reportedly because PTJ had gone around buying up all the tapes (maybe the glasses?)
Better take another look at Dodd Frank
http://bigthink.com/series/62#!selected_item=4845
He starts at 8 mins in, and although the whole video is 2 hours long, he makes a bunch of major points in the first 3 to 4 minutes of his talk, after which he answers questions for 3 to 5 minutes.
His central point is that there is a lot of potential value in getting the fixed costs of the search business down, and that he would prefer to invest there, rather than in playing the zero sum game of trying to take away make share in the $25 Bln search business.
Also worth noting: anybody who can lower the fixed costs of search by an order of magnitude is also going to be in a position to make money from businesses that have similar cost structures.
The successful ones that are a result of an initial good idea are a strict subset of the ones that choose a promising market.
The YC alums are the editors, that's why some things end up dead and others don't
a blend of advertising and ingratiation
HN is YC marketing, it didn't really start out as an alumni magazine either, but once they realized that they could use it to, you know, sell ...
I am curious because it seems like almost any configuration & setup should be able to serve a blog post to the volume of request that come with being at the top of HN.
It looks like the sparsity of the matrix is going to be a much bigger challenge than the scale.
I understand about the focus being primarily on the approach, that makes sense; how are you intending to evaluate the results files?
One of the nice things about them is that they come at the problem from many different angles. Part of the reason summarization has not been particularly productized is because for a long time the standard approach has involved focusing on a narrow domain, training a model, etc. That gets the best results for the local problem, and focus is great for a startup, but that approach is prone to over-fitting, and it is not scalable, or extensible. Ultimately, it has held the whole category back. The solution is probably to take a bunch of concepts from related fields and combine them within the constraint of a scalable framework.
That's why I recommend these papers (beyond the fact that they are relatively approachable): you can almost sense that he's feeling different surfaces of the problem, trying to map texture, and find the right formula for a great general solution.
It would cut down on the need for comments like this: http://news.ycombinator.com/item?id=1815316