What don't you like about Vox? What should I be reading instead?

I never really liked the brightness of Vox. I was using a '68 Sunn combo for a long time but it's too big and loud for what I do these days. I've bee really satisfied with a little half watt kit amp I built 8 years ago for recording and practicing, and the last band I was in I just borrowed anything I could find that had tubes and was over 30 watts.

I've seen a lot of horrible shit in Vox.
Sometimes they have interesting things.

Like a compression algorithm, riding on Phil's coat-tails.

he did it with A Scanner Darkly and it looks amazing.

So it must not have been very accurate then.

They should have been brave and did it like the book : Deckard is a corporate psychopath obsessed with keeping living things as pets, uses drugs to tolerate his wife while, and guns down living replicants who have the mind and innocence of a mentally handicapped child.

Then when he gets dragged into a police station full of replicants and has a gun fight in an elevator, it's meaningful.

I see this neural network can replace Christopher Nolan pretty much immediately.

That entire "police station full of replicants except for the one blade runner he talks to and teams up with" scene made very little sense, other than saying, "Now the tables are turned!" but it was completely out of nowhere and made zero sense.

PKD books always read like someone is slowly having an aneurysm. I think Man in the High Castle actually did end that way.

(dude has major issues with women too)

Yep, he was a nutso. VALIS should have just been retitled DSM IV to save publishing costs.

The problem is, he spoils you with his crazy. You can read a bunch of PK Crazy and then read, say Vulcan's Hammer, and it's boring as hell because it's just science fiction.

Exactly, Meme. That opening scene in Do Androids where he fires up his penfield mood organ to put him in a positive outgoing state of mind, and his wife illegally hacks hers to plunge her into a dark depression, is just brilliant beyond words. He makes everyone else in the genre look like cheap hacks. I would argue his greatest work was The Divine Invasion. I've read that piece perhaps a dozen times, and will read it a dozen more I am sure.

"I just read a book where the Torah is an interactive machine intelligence that defines reality, so I can only assume God sent you to inform me of my role in perfection now that contact has been made" is a great line to drop on unsuspecting Jehovah's Witnesses, too.

And the horror of Buster Friendly, who is like if Jay Leno, Jimmy Fallon, Chris Hardwick, and Bill Maher had been put in a centrifuge and the resulting mess merged together and formed a mega-talk show host overseeing constant witty banter with airhead celebs who are mostly famous for being famous.

Oh god yes, Buster Friendly and his Friendly Friends. When the show "Friends" appeared in the 90's I immediately flashed on this story. All science fiction claims to predict the future: Phil was the only one who actually _could_, and succeeded far more often than he failed.

I think a replicant built to kill other replicants (without realizing he is one!) is taking the same metaphor a step further -- Scott is saying we're all replicants, policing each other in a neurotic society.

From the fancy schmancy Cornell abstract :

"We present an autoencoder that leverages learned representations to better measure similarities in data space. By combining a variational autoencoder with a generative adversarial network we can use learned feature representations in the GAN discriminator as basis for the VAE reconstruction objective. Thereby, we replace element-wise errors with feature-wise errors to better capture the data distribution while offering invariance towards e.g. translation. We apply our method to images of faces and show that it outperforms VAEs with element-wise similarity measures in terms of visual fidelity. Moreover, we show that the method learns an embedding in which high-level abstract visual features (e.g. wearing glasses) can be modified using simple arithmetic. "

Fortunately, you gots me to translate from the academic klingon!

"We used a hundred million dollar neural network to encode RealVideo on the fly. Everyone else uses VLC."

Huh? I'm still trying to wrap my brain around this one. It didn't "learn" the movie. Instead it "recognizes" that this is the movie "Blade Runner?" So if were to make a channel flipping robot someone else could hook-up this machine brain to make the robot stop when ever it sees that "Blade Runner" is on?

no.

okay.


this is similar to the technology I used to make my Deep Dream VR film.

The neural network was taught to learn what blade runner looked like.

then it rendered it from scratch.

It's a computer drawing a picture of something it's "seeing", giving us a good indicator of how terrible computers still are at that and why we should stay off of sidewalks a good 10 years after self-driving cars go on sale.

I'm proud to demonstrate my mastery of the obvious:
The Cornell abstract sounds like a bunch of hand-waving that enables them to burn grant money.

So I can't use this computer network to channel surf for Blade Runner? That's faggy.

The only thing meme got wrong was the price. Otherwise, his executive summary was correct. I think this is legitimate research and is a proof of concept that clearly works, but the press interpretations are overblown.

To put it in layman's terms: we have finally perfected nightmare-vision.

Despite your heroic efforts, I still don't get it. What is difference between "teaching a neural network what Blade Runner looks like" and just ctrl-c'ing the file? Isn't storing and outputting an image one of the most basic things ever? I'm not trying to sound unimpressed... I just don't know enough to appreciate why this copy is so interesting.

Also, c-beams always makes me cringe fur some reason.

It's a compression algorithm. The neural network representation is smaller than the original file. Also rather lossy. Does that make sense?

"The neural network was taught to learn what blade runner looked like.

then it rendered it from scratch."


So AI buzzwords aside, how is this not just a novel approach to video compression?



Rhetorical question.

A less cynical pitch as to why this is important or interesting:

Currently image recognition and now video recognition is big business. On a base level, this technology will be useful for tracking down copyright infringements of various kinds, but it will also be useful during babysteps of AI that we are taking. China has a bot that can look at a photo of soup and tell you that you need more broth, and emotions can be detected in still photographs.

This kind of on-the-fly image construction is like the kindergarten class before Running Man.

It's your birthday. Someone gives you a calfskin wallet. How do you react?

Someone posted an Anita Sarkeesian/Davis Aurini video to poetv. You aren't downvoting it. Why aren't you downvoting it?

No but I am.

What do you mean, "I'm not"?

True that. The neural network has an easier time encoding static images than dynamic ones. That's a technical explanation, but yours is equivalent and much more thought provoking. As I mentioned earlier, I think we are a lot farther away from a true AI than the industry people are promoting. Experiments like this highlight some of the shortcomings.

I'm pretty confident strong AI won't happen.

Neural networks are useful programming technique but as far as modelling actual intelligence, our understanding of what "intelligence" even means is still at the "throwing spears at the sun" stage.

We are very very far away from AI, but pretty close to a lot of amazing things that seem like AI.