Slashdot Mirror

The link in the article is to David Merrill's talk at TED2009. Here is a link to David's Siftables project page at the MIT Media Lab

The prototypes are probably quite expensive. Here is a more technical description of the system. I think you can find most of this stuff in ~50$ cellphones, and they even have an antenna. I don't think why this should cost more than ~30$ a piece.

Marxist? This ardent capitalist can agree that having laws that exist mainly to forcibly extract money out of people who violated no one's rights is wrong. Doesn't matter whether the goal is to "protect children" or "protect our vital industries."

In the case of Cambridge's cameras, it's also worth asking about the fate of the city's private security cameras. For instance, how many does MIT have now? The Media Lab FoodCam was probably one of the first culinary surveillance devices out there.

True. But just bear this in mind about governments, in particular, republics: Winston Churchill once said that those who fail to learn the lessons of history are doomed to repeat them.

He was more right than he knew. ;)

You can see a more detailed description at: http://web.mit.edu/newsoffice/2009/shock-absorbers-0209.html

I find your willful blindness appalling.

And I find your refusal to think for yourself even more appalling. Before you call someone else "blind," check your facts. And read Plato's parable of the Cave while you're at it.

I work kind of in this area as a researcher, so maybe I have a rosy-glass view, but the arguments seem a bit dated to me. Sure, in say 1999 this was a problem, and not that many people took games seriously. But in 2009? Yeah, people still like to kvetch ("games are rarely taken seriously blah blah and we aim to change that" is a standard opening move if you're writing a paper), and maybe the average person on the street doesn't, but there are plenty of inroads:

There are journals and academic conferences on games, in both the humanities and computer science.

MIT Press has an entire division of books about videogames. I'm currently reading one about the Atari 2600, which, yes, even covers its role as a cultural and artistic platform.

There are initiatives and companies to use games for "serious" purposes. The U.S. Army in particular takes them seriously and funds development.

Braid sold over $1m, despite being a kind of weird arty game made by a single guy. You can even get an MFA doing fine-arts stuff related to games.

Heck, Gamasutra itself frequently publishes about games as art, and it's semi-high-profile (at least to the extent that getting linked at Slashdot once a week counts as semi-high-profile).

I mean yeah, I'll agree that far more people respect, say, film than respect games. But it's not as if this is some novel argument and nobody has ever thought about taking games seriously before. Also, to some extent, it's the fault of people not making more interesting games: Hollywood may be crap, but there are a lot more innovative indie films out there than innovative indie games.

I work kind of in this area as a researcher, so maybe I have a rosy-glass view, but the arguments seem a bit dated to me. Sure, in say 1999 this was a problem, and not that many people took games seriously. But in 2009? Yeah, people still like to kvetch ("games are rarely taken seriously blah blah and we aim to change that" is a standard opening move if you're writing a paper), and maybe the average person on the street doesn't, but there are plenty of inroads:

There are journals and academic conferences on games, in both the humanities and computer science.

MIT Press has an entire division of books about videogames. I'm currently reading one about the Atari 2600, which, yes, even covers its role as a cultural and artistic platform.

There are initiatives and companies to use games for "serious" purposes. The U.S. Army in particular takes them seriously and funds development.

Braid sold over $1m, despite being a kind of weird arty game made by a single guy. You can even get an MFA doing fine-arts stuff related to games.

Heck, Gamasutra itself frequently publishes about games as art, and it's semi-high-profile (at least to the extent that getting linked at Slashdot once a week counts as semi-high-profile).

I mean yeah, I'll agree that far more people respect, say, film than respect games. But it's not as if this is some novel argument and nobody has ever thought about taking games seriously before. Also, to some extent, it's the fault of people not making more interesting games: Hollywood may be crap, but there are a lot more innovative indie films out there than innovative indie games.

Paleoseismology as you described is actually quite difficult. In the case of the San Andreas, you can't really look at off-set streams and such. You can rarely discern more than one or two events along such offsets, and once you do, it is very difficult to determine the age of the offset. You can get the amount that it's moved, yes, but not the timing. Worse, since you don't know the timing, you don't know if the offset is from one or more events.

The way it's done for strike-slip faults like the San Andreas is to look at a cross-section perpendicular to the fault, looking for layers of material off-set (or suddenly changing thickness, etc.) along the fault. The best way to date those layers is through carbon-14 dating of organic material, which can give you accuracy only within ~1-200 years - and that's assuming that the organic material you date is not from elsewhere, is not from 200 year old trees, etc. If an event offsets every layer from the bottom up to a certain point, you date the top layer that it cuts through to get a maximum age, and the layer that it didn't cut through is the minimum age.

You can imagine the difficulty and ambiguous nature of this. The individual layers that you have to recognize and date are on the scale of centimeters to decimeters - I've seen some of the areas that were used, the famous one being along Pallet Creek which is along the San Andreas northeast of LA (I have a picture of it - well, it is a picture of a girl standing in front of it - here: http://www.flickr.com/photos/penguinchris/3037578910/) Here, luckily there was constant, relatively rapid deposition of material. In most places this is not the case, so any record of movement on the fault is eroded away.

For the San Andreas, we have a partial record going back ~1500 years. There really is no reliable way to reach back further than that - the record isn't normally visible in older rock units. Looking at the larger-scale structures is interesting by itself but doesn't tell you anything about when specifically there was movement. The fault system in the Sichuan region is fairly well understood - it is a kind of combination strike-slip/thrust fault (see http://quake.mit.edu/~changli/wenchuan.html for some nice diagrams.) But I want to call BS on the idea that they have any idea how frequently major earthquakes have happened there - and even if they do, the idea that it is "perhaps millions of years" since the last one is ridiculous no matter what.

And then, when you *do* figure out a approximate year for an earthquake, how do you determine how big it was? Again, extremely difficult! The best estimates come from comparing old written records of destruction with those from modern earthquakes - nothing scientific at all!

What's being done extensively with the San Andreas is physics-based computer modeling - we have some idea of the force building up, and combining that with records of historical earthquakes we can make an estimate of a major earthquake every ~150 years. But even for this, the best-studied earthquake area, it's not much more than a guess.

I don't know as much about the Teton fault (other than that it is a normal fault, not a thrust fault as you stated ;) ) but I'll comment on the idea of a "magnitude 7 earthquake every 400-700 years." These kinds of estimates are based on the very difficult work I described earlier (and I'm not sure how much has been done for the Teton fault) and whatever geologist came up with that would probably admit it is a simple guess without much to base it on. I mean, think of it - is knowing there's a large earthquake every 400-700 years really all that useful anyway?

By the way, I assume any dating of the Teton fault would be done this way: when new patches of rock are exposed along the fault as you described, they start getting hit by cosmogenic radiation. By measuring the amount of cosmogenic radio isoto

Too late:

http://trillian.mit.edu/~jc/ideas/AudioHomeRecordingAct.html

If the RIAA had perfect aim and only took out pirates that it could prove were pirates, they'd have a lot more sympathy.

Not really. I don't care if a given 16-year-old kid is guilty - I don't want him sued out of his college fund. If the RIAA had sued for reasonable amounts (say, $5-$10 per proven upload) and gone about it fairly, I might have a little sympathy. As it is, they're on the short list of people I'm going looking for if we ever have another civil war.

This is a crude proposal. There are probably much better ones out there.

Yes, quite a few. They tend to run along the same lines, but with different approaches for ensuring, with high probability, that your ballot ID can't be tied to you.

See ThreeBallot (and variants), for example. Others include Scantegrity II, which has gotten a fair amount of attention recently.

hmm not the post I would have chosen for this news... Could have pointed out some of the source post announcements and avoid perpetuating a few misconceptions.

I have heard about Theora is that it is technically inferior to many other video codecs

Hence the need for funding the Thusnelda enhancements. Theora is a pretty solid codec and can be greatly improved with a few enhancements on the encoder side.

I wonder if wouldn't be better to direct effort to Dirac, perhaps putting Dirac into an Ogg container

Dirac is best at high resolution high bitrate video and not so good for standard definition low bitrate video, hence an enhanced theora is the optimal way to hit the low bandwidth target. Enabling theora to be competitive or better than others codecs in the low bitrate range in the intimidate future with relatively small investment.

Furthermore dirac is planed for inclusion and will be explored in the tail end of this grant. (once liboggplay is more solid). Making liboggplay playback library solid will enable Dirac support to be solid as well. Since Dirac already has a maturing decoder/encoder library (Schrodinger) and already been mapped to an ogg container (what liboggplay plays).
It's relatively easy to add in additional free codecs with ogg mappings. if( FLAC, Speex or Dirac) and will not be the primary use of the funding so its not focused in on the announcement or secondary coverage of the announcement.
More info on the announcement here and the above mentioned links.

Can this really work outside of a lab, where the water surface isn't like glass ?

It sure can work outside the lab. Check out the pics (search for figure 2 / figure 3) to see photos!

I love how Figure #9 has the naughty bits blacked out!

Here's the math and a nice picture (page 4):
http://web.mit.edu/1.63/www/Lec-notes/Surfacetension/Lecture4.pdf

Can this really work outside of a lab, where the water surface isn't like glass ?

It sure can work outside the lab. Check out the pics (search for figure 2 / figure 3) to see photos!

Not crashing when calling something on a NULL object is not necessarily a bad thing. You should read: Enhancing Server Availability and Security Through Failure-Oblivious Computing You can also use TRUE/FALSE and 1/0 for boolean values, too. Dropping stack allocation for objects is, IMHO, a good thing. The fact that C++ has to truncate an object when you pass a subclass to a method that expects the superclass via the stack, but does not when you pass a pointer to it. Not what most people would expect and would break dynamic dispatch.

Scratch is good too. Variables, loops, iterators, etc. Along with cute cartoon characters. My boy started at 2 and is learning well.

Allow me to add a few, since I might know a thing or two about this subject. Some of you might remember Flame, a robot designed at the TU Delft, and being used to further understand human walking (he walks like we do, as opposed to for example Asimo...)

http://www.3me.tudelft.nl/live/pagina.jsp?id=c4fa06f1-b767-4a67-a19e-ea3356400f06&lang=en

The nice people at DBL (Delft Biorobotics Laboratory) have built a next generation robot called TUlip

http://www.dutchrobotics.net
for those interested. That's one I worked on a little, so I might be a little biased in terms of how cool it is :-)

Some more cool robots we saw in China and elsewhere are:

The Cornell Ranger's record for longest distance walked
http://ruina.tam.cornell.edu/research/topics/locomotion_and_robotics/papers/CornellRanger/index.html

or maybe nexi?

http://robotic.media.mit.edu/projects/robots/mds/overview/overview.html

Or take your pick from a variety in this list:
http://www.ri.cmu.edu/research_project_view.html?menu_id=261

or this one

http://www.rec.ri.cmu.edu/project/index.htm

Cool. I first played Zork on a VAX at Bell Labs, right before Infocom was formally formed in 1979.

There's a great student paper (research project?) from MIT that quite nicely recounts the history of Infocom, the making of Zork, and their fall etc.:
http://web.mit.edu/6.933/www/Fall2000/infocom/infocom-paper.pdf
(yeah, PDF sorry)

Abstract from the paper:
The success and failure of Infocom, a company founded by members of MIT's Laboratory
for Computer Science, resulted from a combination of factors. Infocom succeeded not only
because it made Zork, a text-adventure game, available on personal computers, but also
because it developed an effective system for supporting new platforms, maintained an
engineering culture that excelled at writing computer games, and marketed its products to
the right audience. Similarly, Infocom did not fail simply because it decided to shift its
focus to business software by making Cornerstone, a relational database. Infocom failed
for many reasons that were closely tied to how the company managed the transition to
business products. Behind the scenes, the transition created a litany of problems that hurt
both the games and the business divisions of the company. Combined with some bad luck,
these problems--not simply the development of Cornerstone--ultimately led to Infocom's
downfall.

SICP has an good examples of that problem: http://mitpress.mit.edu/sicp/full-text/book/book-Z-H-23.html#%25_sec_3.4

Functional programming may be an answer, but this answer is limited by mother nature.

4. There are computing-jobs that are inherently not parallel.

5. Parallel programming is hard not because of bad programming languages but because of the logical problems that come with shared state and parallelism.

http://mitpress.mit.edu/sicp/full-text/book/book-Z-H-23.html#%25_idx_3598

Therefore multicores do not bring a substantial performance benefit. Futhermore because the problems are fundamental logical ones, there is no big hope.

Okay, this idea is not so original. This concept was originally created my MIT in 2003 and was called ASSIST: A Shrewd Sketch Interpretation and Simulation Tool. At least give props where props are due, instead of making it look like they though of it all them selves. http://rationale.csail.mit.edu/index.shtml

If creation of generated research papers papers can be considered fun, then by all means!

I graduated from MIT and loved it, but have been questioning how much it was really worth. One thing that the Institute didn't do for me was kick my tail when I decided to major in a humanities subject, thus making my degree useless. I was too ignorant to understand that this was a mistake, and was mainly thinking that I didn't want to end up as Dilbert. Innovations like MIT's OpenCourseWare project and a shift in class sizes may be useful, but they don't address the question of whether students are learning anything useful for their tuition.

They're not using telephone operators, they're using the intertubes.

http://ocw.mit.edu/

MIT can do that to people. Burnout is a real risk there. But MIT's 6-year graduation rate for undergraduates is 94%. Most students do make it eventually. By comparison, Ohio State is at 68%. The University of California at Santa Barbara (America's best college for sex) is at 65%.

(I didn't go to MIT. Went to Stanford.)

The giant schools are not the place where the best educations come from. Sure they often have the biggest research budgets and thus are in the news the most. Smaller schools with smaller class sizes are where it's at from a value for dollar spent standpoint.

My biggest class was intro psych and it was 75 folks. My Hydrodynamic instability was four students and the professor. Just try to hide when you haven't prepared with only three other peeps to hide behind.

Sheldon

MIT's not a giant school. Their freshman class is around 1000., which is bigger than it was when I was a freshman, but not as big as a few year ago when they took steps to reduce size.

I think I had a half-dozen big classes the entire time I was there; the rest of the classes were small enough that I felt everyone got enough attention, especially in the recitations, where my biggest beef was the occasional grad student who didn't speak English.

I think the worst was a math professor pressed into service for a recitation section, who would stand at the board, say, "Uh, I don't remember how to do this one. What's the answer to this integral? Oh yeah, it's..." and write down the answer and prove it was right. But I later found out that is actually how you solve differential equations!

"...because giving a diploma to everyone who pays waters down the value of the diploma."

http://tech.mit.edu/V121/N14/col14nesmi.14c.html

"Even at MIT, where we pat ourselves on the back for our meritocratic ways until our skin is raw, admissions staffers report that legacies are granted an additional review before their rejection is finalized. At several schools, such students receive much more than an extra review. "

Talking about a school which even considers "legacy" status in admissions not wanting to give a diploma to everyone who pays...

Thank you. That's the best laugh I've had all morning.

I'm surprised no one mentioned MIT's long-running "Charm School" designed to teach nerds table manners, basic fashion, and dating tips.

What? Where's the art in that? I love the handcoded steganographic puzzles like this (or, really, any of the puzzles in the MIT Mystery Hunt). There is real artistry in designing a unique code and--although it's not exactly standard cryptographic practice--building in enough clues that an outsider can, potentially, with enough caffeine, break it.

Would having a conversation with a computer that was capable of understanding conversational english be awesome? I imagine it would be. But what would we talk about? What would I do with such a computer that I couldn't do with my current PC?

You are thinking too narrowly about "computers". I agree that talking to your desktop is largely inefficient compared to using a keyboard and mouse but natural language processing in robotics is another story.

The idea is that we may have hundreds of computers within our immediate environment all around us most of the time. See Intel's work on Claytronics http://www.cs.cmu.edu/~claytronics/ and MIT's work in robotics http://robotic.media.mit.edu/projects/projects.html. These technologies will be miniaturized and have some degree of natural language processing. Since you are not necessarily siting down at a desk with a keyboard and mouse, one of the most convenient ways of communicating with them will be through speech, gestures and eventually thoughts. Emotive and others are already making good progress in the thoughts category. http://emotiv.com/

Yeah, that'd be awesome. but that's nowhere near being on the horizon yet, and I don't know that we'll ever get there, because where's the demand for the intermediary steps that would lead us there, and what would those intermediary steps even be??

There is indeed a great demand for first generation, speech enabled robots and embedded computers. Think of the money that can be made for a robot that cleans using even a very limited amount of AI. Add in some speech recognition and I bet someone from http://store.irobot.com/shop/index.jsp?categoryId=2804605 would be very interested in talking with you.

I remember once years ago freaking my colleagues out with a largish app written in R... with nary a loop anywhere.

That's a feature of functional languages, a class that also includes Scheme and XSLT. The basic idea is that programs should not have state, because state makes them harder to debug. A for or while loop, by definition, has state, so you have to do your iteration some other way, namely Tail Recursion.

I suppose that makes sense, but I've never been able to teach myself to think that way. It's the main reason I never managed to get through The Wizard Book.

R has a scheme-like lower layer but it feels more like APL with its array manipulation capabilities than Scheme. It does not support tail recursion.

I am surprised the following report has not been posted or submitted to /.

Future of Human Spaceflight (16-page PDF)

The MIT Space, Policy & Society Research Group took a step back from the "do this" / "no, do that" debate and examined the very questions being posed by the National Academy of Sciences.

The above link has a 16-page document that examines the reasons for a human spaceflight program. The report is compelling, challenging and thought-provoking.

Give it a read!

That sounds extremely weird: if a program has a stack, then it has a state - the location on the stack is still state. Thus, if you use recursion, you still have state. I mean, you can try to hide the fact that you have state, but I don's see how you can have a program without state.

Even the wizard book appears to have a chapter on state: http://mitpress.mit.edu/sicp/full-text/book/book-Z-H-19.html#%25_chap_3 , but, unlike your description, instead of talking about a program without state, it considers two kinds of state: the state of objects, or the state of streams of data.

Do you happen to have a link to what you mean by "a program should not have state"? Because, I mean, that seems antithetic to the nature of a program.

I remember once years ago freaking my colleagues out with a largish app written in R... with nary a loop anywhere.

That's a feature of functional languages, a class that also includes Scheme and XSLT. The basic idea is that programs should not have state, because state makes them harder to debug. A for or while loop, by definition, has state, so you have to do your iteration some other way, namely Tail Recursion.

I suppose that makes sense, but I've never been able to teach myself to think that way. It's the main reason I never managed to get through The Wizard Book.

Let's keep a few things in mind:

1. This was "a technology he created as an intern at NASA in the summer of 2007." It's not like he was an undergraduate sitting in a classroom -- he was working for NASA when he made the invention.

2. "The iShoe researchers used some of their own work and previous NASA data ," the latter presumably taken with "an expensive device about the size of a phone booth" in the creation of their invention. So NASA's data (and presumably equipment) were needed to produce the invention.

3. While an intern, Lieberman was also a federally-funded (i.e., taxpayer-supported) graduate student, receiving money from both the National Science Foundation and Department of Defense, through his university, for his research. Like many (perhaps substantially all) graduate researchers in US universities, he was being paid by his university to do research. The fact that the research was being conducted at NASA doesn't change the fact that Lieberman was on the university payroll at the time the invention was made. Welcome to internships.

4. His company has also filed for federal funding to develop the idea for market and, "[o]nce funding is obtained, the iShoe could be for sale in 18 months, Lieberman said." So he's still using taxpayer money to develop the invention for market.

5. We don't know what the "hefty royalty" is (unless I missed it, it's not in any of the linked articles), but $75,000 is peanuts. "The iShoe has a way to go to reach the market [...] Lieberman estimates $1 million is needed for a broad clinical trial, and $3 million to $4 million to bring the insole to market." As a startup, his monthly burn rate will be much more than $75,000.

Frankly, I'm fine with institutions receiving a financial return on the work of their paid employees -- especially if taxpayers are ultimately footing the bill. In fact, I would argue that Mr. Lieberman is getting a sweetheart deal; I think once he gets into industry himself he'll find that the commercial sector typically requires employees to assign all rights to any future inventions (at least, in the company's field of interest) to their employer starting on Day 1, usually with trivial or no compensation.

It will be interesting to see what intellectual property policy the new iShoe company establishes for its own employees. As CEO, will Lieberman let his iShoe researchers invent and patent without expecting that those inventions will belong to iShoe?

full 64bit drivers that can shove data to devices oh like Video cards much faster

How do 64bit drivers speed up DMA?

This includes not only the OS's operation, but even 32bit applications running on the OS.

My understanding is that 32bit application would run slightly slower if the CPU was in 64bit mode. Presumably 15% would be the overall system performance, including legacy 32-bit applications?

You see when you have a 64bit memory addressing and can optimize for this in the memory manager you no longer have FS and pagefile lookkup tables for extended amounts of RAM.

What is a FS (filesystem?) lookup table?

You also can do like Vista x64 does and shove two 32bit memory writes into on 64bit address space, so when it can, you get double the read/write performance out of the memory chip because you are pulling two 32bit chunks in one read cycle.

By "64 bit address space" I presume you mean 64 bit register (you fit 2^32 32bit address spaces in a 64bit address space). But even in 32bit mode Intel CPUs can access 128bit registers via SSE. Anyway, this presumably has more to do with your compiler than your OS, so I don't know what Vista has to do with this.

Everything else being equal, 64bit software would run slower than 32 bit, because you need twice as many bits to represent a pointer. Essentially, unless you need an address space larger than 4GB, you are wasting 32bits on every pointer. This would waste memory, cache and memory bandwidth etc. The standard answer as to why 64bit software runs faster on Intel/AMD CPUs is that on these CPUs everything else is *not* equal.

The biggest bonus to running in 64bit mode on Intel/AMD chips is that since 64bit is essentially a whole new arch, we can throw out all the backward compatibility. In 64bit mode we actually have a decent number of registers. Also since 64bit code won't run on old processors anyway, there is no point in compiling code to be backward compatible with the old i586.

Understand yet?

Not really. Not any better than I understand this paper anyway :P. Could you give links explaining your claims above?

I much prefer to read Eric Drexler's PhD thesis, Molecular Machinery and Manufacturing with Applications to Computing. Chapter 11 (nanomechanical computational systems) is particularly interesting.

Otherwise, the person would have to (a) hear me correctly over the cell network

Not difficult. Take your pickof ways to do it.

and (b) write down what I said on a piece of paper

Or they could remember what is usually two letters (airline) and a four digit number for the remaining duration of the call or voice mail (which should be short since it can fit in a 160 character SMS), before they can store it in the Notes function on their phone (if they don't have a notes function, they can enter it as a temporary name in their address book). Unless, of course, you were actually organized enough to have sent it to them by e-mail days earlier when you booked the flight.

For you lazy folks, here's the garbage abstract:

Recent advances in cooperative technology and classical communication are based entirely on the assumption that the Internet and active networks are not in conflict with object-oriented languages. In fact, few information theorists would disagree with the visualization of DHTs that made refining and possibly simulating 8 bitarchitectures a reality, which embodies the compelling principles of electrical engineering. In this work we better understand how digital-to-analog converters can be applied to the development of e-commerce.

The paper was generated by the SCIgen project at MIT. According to , the program is meant to generate garbage.

Our aim here is to maximize amusement, rather than coherence. One useful purpose for such a program is to auto-generate submissions to conferences that you suspect might have very low submission standards.

When I read the Slashdot summary, I totally missed the point. The point is that some MIT folks have created a garbage paper generator and are mocking the 2008 International Conference on Computer Science and Software Engineering.

No. I asked and it said: "Scheme and the Turing machine, while compelling in theory, have not until recently been considered significant."[1] Apparently it's not self-aware: It's written in Perl and not Scheme. :) [1] http://apps.pdos.lcs.mit.edu/scicache/315/scimakelatex.75855.Zebigneow.html

Looks like this El Naschie (is he a Mexican wrestler?) is using the Mathematics equivalent of SCIgen - An Automatic CS Paper Generator.

Time for a duel between El Nachies and SCIgen?

I followed some of the links, and was appalled at the prices. $100 for a simple summary of OS technology? That's a blatant, immoral cashing-in on the fact that students are are a captive audience.

(What's really sad is that $100 for textbook is actually relatively cheap.)

Even $70 for SICP is ridiculous. Fortunately, the authors are kind enough to provide a free online copy.

Going through all my CS courses back at the university this was my favorite book: http://mitpress.mit.edu/sicp/ I had the original text but since this book was so awesome it is now available on the website.

I haven't been able to find it in my brief perusal of the link... does anyone know offhand if the MIT students asked permission first, or if they just did it, planned the talk, and then got in trouble?

Here, just read the presentation for yourself. I think there's also a more in-depth paper that was released by the courts when it was entered into evidence by the MBTA (doh!), but I can't remember where I saw the link for that.

Yes. Except it was said by one of the bad guys who was trying to inflame mob rule and take down the rule of law.

For a slightly more holistic project approach, take a look at a MIT 2.009 Product Engineering class (Mechanical Engineering dept), which now has videos from their projects for this semester: microwave fire extinguisher, self-adjusting electric cook-top array, basketball player tracking system, etc. There are also some neat projects for microcontroller beginners on the NerdKits videos page. DIY digital scale interface over USB, morse code decoder, iPhone R/C car control, and more. (Disclaimer: I did some of the electronics design for the 2.009 Purple Team, and am one of the NerdKits team.)

My school (Texas) has a course where were build up this OS, step by step. Right from the bootloader to it's full multitasking glory. By far the most useful/fun course I have ever taken in my life. Link: http://ocw.mit.edu/OcwWeb/Electrical-Engineering-and-Computer-Science/6-828Fall-2006/Assignments/index.htm