One word. "Apps."

So out of 3 million people signed up with Coursera, only 900 have completed 10 or more courses, comparable to roughly a year of full-time schooling. Only 100 have completed 20 or more. That's a 99.97% dropout rate after one year.

This isn't going to replace other forms of education with stats like that.

OK, so this vaccine needs a booster every decade or so. Lots of vaccines are like that. The vaccines against tetanus and hepatitis A and B all need to be re-administered every few years. No big deal.

The Android line is not a paid consumer product. It's a piece of middleware offered to phone manufacturers.

Wonder how long "Google Play Music All Access" will last? Compared to PlaysForSure (Microsoft), Zune Music Pass (Microsoft), and WalMart Music Download Service. Just having a big company behind it is no guarantee of success. Google has never had a successful consumer product that people had to pay for.

List of discontinued Google products.

Despite all the noise, almost nobody is making money in "social". Even Facebook isn't very profitable, despite its size. The business strategy in "social" seems to be to give the service away for a few years, build a following, then crank up the density of ads until the users get fed up. Worked for Myspace, right?

Facebook traffic peaked about a year ago. Twitter is now exploring the user's threshold of pain with "sponsored tweets". This is robocalling in another form.

Basic truth: ads with search results are useful to users and effective for advertisers, because they're presented when the user is actively looking for something relevant. Ads on "social" are merely annoying because the user is looking at what their friends are doing.

OK, a fanboy from the Arduno cult has been heard from.

Back in 1979, Milton Bradley introduced the Big Trak. This was the first mass-market battery-motor-wheels-CPU toy 'bot. Since then, there have been more machines in that category and slightly above it, like Lego Mindstorms.

It's been three decades since the Big Trak. There hasn't been much progress above that level in mass-market devices. A Roomba is only slightly smarter than a BigTrak. Mobile phones, on the other hand, have advanced somewhat since the late 1970s. R/C toys have become much better, but most of that reflects improved batteries, and the good stuff is still at a rather high price point.

There's a new BigTrak from 2010. It has an optional camera and a WiFi connection, and will connect to an iPhone. It has the basic hardware to be an intelligent autonomous vehicle. But it's no smarter than the original BigTrak. If you want something as dumb as a BigTrak, you can buy one of these. No assembly required. Ages 6 and up.

Here's what's possible today at the hobbyist level: an autonomous paintball robot. Runs a maze and hits targets. Uses a Kinect as a sensor. Has 2D SLAM; builds a map of its environment. That's what new products should be doing.

There are already dumbots in that range. Any new robot should come with at least an Allwinner ARM CPU ($7) and a camera as standard. That's enough for some vision processing and at least 2D SLAM. The hardware to put some real smarts in a little bot is now cheap and there's enough open source software available to get started on making it smart.

The trouble with being a plumber is that most of the work is in building and remodeling. With housing construction way down, most of the people in the building trades are hurting. It's great during a building boom, though.

A related trade is HVAC - heating, ventilating, and air conditioning. There's more electronics and control involved than in plumbing.

It's amazing how much miscellaneous rock is floating around this solar system. Four sizable chunks of rock (tens of meters) have gone by the earth in the last week, one within lunar orbit. None were known objects.

There's a mile-sized one going by on March 31st, but closest approach is over 3 million miles.

Plutonium and uranium are alpha emitters. Alphas won't get through a sheet of cardboard. A gamma ray detector won't pick up anything. This won't detect an atomic bomb.

This is only useful for detecting radioactive waste, miscellaneous medical and industrial radiation sources out of their casings, and X-ray machines.

There's a list of things humans can do. There's a list of things machines can do. The second list is growing steadily. The first list, not so much. As machines check off more of the items on the list of human capabilities, the need for human workers decreases. As new jobs appear, more of them will be done by machines.

The current "jobless recovery" demonstrates this. US production is back up. The stock market is back up. The number of people working is not back up. Hiring large numbers of people is so last-cen. Even Foxconn in Shenzhen is converting to robots.

We don't need "the singularity" for this. Just routine progress. Computers are so cheap now that they're cheaper than even low-wage people.

Here's a vision of the future. Watch this Kiva Robotics system fill orders. Those robots already fill about 15% of on-line orders in the US (Gap, Staples, Office.com, Walgreens, drugstore.com, pets.com, etc). Amazon bought Kiva recently. Those big new warehouses Amazon is building for local distribution won't have many employees. They'll kill off even more of retail.

We may not like the society we get from this, but that's where capitalism is taking us.

Machines should work. People should think.

If you're running on Windows 7 or Vista, press CTRL, TAB and the "Windows key" at the same time and watch what happens.

That's "cards" mode. Did you know Windows could do that? Is it useful?

The U.S. Army's current sniper rifle, the M24 Sniper Weapon System (SWS) is basically a Remington 700. Its replacement is a modified Remington 700 with a larger caliber.

From the article:

That's the problem. Just because we can extract the wiring diagram doesn't mean the components are well understood yet. Also, if we understood the components and how to wire them up, it would be cheaper to just build hardware. Simulating neurons is slow. It's like running SPICE instead of building circuits. Works, but there's about a 1000x or worse speed, power, and cost penalty. GPUs are often simulated at the gate level before making an IC; NVidia uses twenty or thirty racks of servers to simulate one GPU during development.

What bothers me about claims of strong AI is that I've heard it before. Ed Feigenbaum, the "expert systems" guy at Stanford, was running around in the 1980s, promising Strong AI Real Soon Now if only he could funding for a giant national AI lab headed by him. He even testified before Congress on that. Expert systems were a dead end.

Rod Brooks from MIT went down this road too. His COG project had a robotic head and some arms, some facial expressions, and a lot of hype. Work ceased on that embarrassment in 2003. He'd done good artificial insect work, but the jump to human level was way too big.

This is the hubris problem in AI. Too many people have approached this claiming their One Big Idea would lead to strong AI. So far, not even close.

All the mammals have similar DNA and brain architecture. A mouse brain is about 1g; a human brain is about 1000g. So build a simulated mouse brain and demonstrate it works, or STFU.