I also used some of this free time to do a 4 day machine learning run on all current English Wikipedia articles.
I will always create an AMI for all of my future books - awesome for readers to be able to have everything all set up. Also, they can clone my AMI, and have a fast start to deploying anything in my book that they want to use for their own projects.
For Amazon, this seems like viral marketing at its best :-)
I think that it was very savvy of Amazon to do this because I ended up writing a lot of material in the book on using Amazon AWS (Elastic MapReduce, S3, etc.). I now spend lots of time experimenting and blogging about AWS, and I ask my customers to host on AWS because I am now so used to using it.
I used to host customer work on VPSs or cheap dedicated servers, but now Amazon is definitely my deployment platform first choice and recommendation. (BTW, I host some of my own stuff on AppEngine because it is even cheaper, but the extra effort of dealing with AppEngine makes it not as attractive for customer projects.)
1. Get a genome for the simplest bacteria you can find. 2. You'd need molecular simulation software, specifically DNA and cells
After you've gotten this, I would strip out pieces of the genome using the genetic algorithm (oh the irony) and see if you can slim the genome down to a bare-bones replicating cellular machine.
After that, just a little trial and error and you could be making designer organisms.
(Sorry if you wanted something realistic. My brain doesn't work like that =)
That sounds interesting from a computational perspective. Any good papers to read? "Designer organisms" also sounds interesting...
Seriously, though, probably build a Web indexer/search engine. I seem to keep coming back to the idea every few years and EC2 makes it possible - a million free hours would get me 114 instances for a year, long enough to prove whether any of several new approaches is worthwhile or not.
As for the 1,000,000 hours? A genetic algorithm that runs through all the publically available data on deals made by the government looking for anything out of the ordinary to flesh out corruption.
And I don't think 'genetic algorithm' means what you think it means. All a GA does is try to find the max/min of some function. It is basically just hill climbing with a bit of randomness and a catchy name.
To find 'fishy' transactions you would use some sort of unsupervised machine learning algorithm to identify patterns/clusters in data (and that's starting an entirely different discussion).
As for the generic algorithm, you are _also_ right.