i take it back, my sub is totally different...!

I look at the hopper once in a blue moon.
If there is "poaching" from the hopper it's because it's possible to submit a link that is in the hopper, which poetv should see as a dupe, but doesn't. Big surprise.

call Christopher Walken, quick!

same here

It looks like what they did was measure the brain response to a huge library of youtube clips. Play a clip to a person getting an mri, then use software to determine which videos produced similar patterns earlier. The final image seems to be just an amalgam of the software's best guesses for which clip was being watched.

It's still impressive, but it would have been more honest to just show the single most likely clip instead of the amalgam. It seems to be largely influenced by the most accurate clip, so they would probably get a pretty successful match rate.

Uh, this is not BS...

Why does it look like a guy in a tee-shirt on the right when Steve Martin in uniform is on the left?

Additional video: http://bit.ly/o08uEv

In this one, you can see the machine's "best guess." It's the 2nd column, the one sitting just to the right of the "average" (blurry) images.

You'll see that the best guess is frequently totally different. This is because of the incredible random variation of human bold signal. The blurred result is impossible without consistently averaging MANY matches -- and for best results, this could be an average of hundreds or thousands of videos. (Or more, or less, depending on the kind of machine learning they're using.)

But in all seriousness, the implications of this are enormous. It provides a very useful platform that can accomplish all sorts of secondary mind reading.

I'll admit this is my own amateur speculation, but take this for example:

Watch the link and notice how each subject has their own trends in matches, with consistent individual patterns of inaccuracy. E.G. Subject 1's best average image is always darkest of the three. It most likely represents a consistent tendency for the activity on part 2 of the experiment (the test part, probably performed a day later) to be more correlated with darker-colored videos.

This means that the general patterns of activity of his visual cortex were different, but in a non-random way. Non-random means a predictor variable exists, a variable that could reflect the subject's current mood, energy level, or hormone/chemistry levels. Who knows?

The reason this would be valuable is that of all parts of the brain we understand, we understand the visual cortex -the most-. Mood circuitry, for example, is struck with dozens of barriers to accurate pattern-recognition: heavy blood flow in that part of the brain creating flow artifacts, ... but the visual cortex is solidly predictable. Hence, another possible indirect method of lie detection; the best currently available is GSR. Cool.

That's all cool simillion, but when am I going to star in baleen movies where I'm just banging everybody I want to.

This isn't really dream reading. This is taking a map of the parts of your brain being stimulated by what you're seeing. I'd be very curious to see what results they get when they're hook up to a person experiencing REM sleep. It might just be nonsense.

Not that I know anything about this subject matter, but I think dreaming is largely a result of activity in a completely separate part of the brain.

Short answer, when this tech is perfected, you can film yourself banging everyone you want to and then watch that film.

Stop laughing! You think the Worldwide Mad Deadly Gangster Computer God hasn't surgically altered your Frankenstein-enslaved body at night to further the Communist "Black Wave" of the future!?!1?

You are incorrect, sir. But it would be bitchin'. I'd like to see some of my crazy ass dream output so I could finally go "see? SEE? I told you!"

Then what is it really doing?...