Artificial intelligence still has some way to go

Both funny, and prompting the question why would you have such a thing on your house?

Karl otm

Anyone see that bbc future technologies series from the last several years? Suggested the next war may be transhumans vs non modded peoples, like x men level stuff

the next war is probably going to look depressingly like Saudis bombing the shit out of Yemeni civilians far more than it will look like transhumans vs non modded peoples, and it will probably start less than a year from now. and another one will start the year after that.

The next war will have nothing to do with AI, and everything to do with natural idiocy

Yeah agree

combining this

https://www.youtube.com/watch?v=7Dg49wv2c_g

with this

https://www.youtube.com/watch?v=GuZGK7QolaE

equals a new circle of hell

see you all there

the use applications in the top video are so purposefully naive - if you team loses the big game, you can switch the jersey so that you're wearing the winning team's jersey!!!!

ooooooor, you could make the target a major political figure, make the major political figure say exactly what you want them to say in lifelike fashion, and send it to a bunch of people who still use aol accounts as their primary email addresses to confuse them!

NVIDIA's new vid2vid is the first open-source code that lets you fake anybody's face convincingly from one source video. prior "face2face" stuff was either cartoonish or proprietary. interesting times ahead... https://t.co/JsPVVa3xwa pic.twitter.com/AFhpeObd8N

— Gene Kogan (@genekogan) August 21, 2018

(btw i realize this stuff isn't AI, but it seems relevant)

(also, maybe facial recognition is not a very good long-term security strategy for unlocking phones)

I was recently wondering whether it was possible to truly articulate your research for the next ten years (a common question in grant applications) with just one example. I came up with this:

I’ve been thinking about the following problem: pic.twitter.com/P3hztqShZ3

— Allen Goodman (@0x00B1) October 3, 2018

perhaps a CV solution to spot the ball would be a good first step?

(has anyone done that, actually?)

I get the feeling lately that humanity will never be over taken by the machines because we will never make the machines - we will instead follow the path of Dune and strat breeding superintelligent ladies

the primary goal of all tech research is to make anime real

Amongst many other things it now occurs to me that dancing would be a lot easier if we were quadrupeds
https://www.youtube.com/watch?v=kHBcVlqpvZ8

no one's stopping you, friend

okay, the robot looking over its shoulder at the camera while it lewdly wiggles its but is cracking me up.

butt

i always fearmonger about this shit and no one cares, but

upgrading meat puppet to HD. as soon as i can get a good voice model, i’m going to make a parallel state of the union address pic.twitter.com/YPcuqxJMJB

— Gene Kogan (@genekogan) October 30, 2018

it won't be the end of the world, but in a few years things like this are going to work much, much better and the amount of disinformation out there is going to be absolutely insane. the solution is...some sort of embedded token in media files that "proves" the identity of whoever posted it...or something?

It'd be like possible to do some sort of certificate signing on every piece of media or whatever, but some asshole is going to insist on doing it with blockchain

anyway easier imo to stop consuming information

I know that we all hate the block chain, but since I'm a dummy could someone explain why it is so bad and hated?

personally i've always thought it could have legitimate use as a way of enforcing contracts (or confirming identity, in this case), but then again, i'm a dummy 2

I mean it's not not a like interesting-ish thing from a distributed systems point of view but like bitcoin now uses more electricity than Peru to manufacture fictional libertarian value tokens

oh yeah, i mean cryptocurrency is fucking terrible, particularly the kind that relies on mining. but that's not blockchain.

utilities are pretty excited about blockchain, here's an example:

https://www.greentechmedia.com/articles/read/wepower-is-the-first-blockchain-firm-to-tokenize-an-entire-grid

is it the crypto specific application or blockchain itself that's not efficient at scale?

the issue is that there's an incentive to join the distributed mining operation, and therefore people do in an attempt to turn electricity into libertarian value tokens, and the way the proof-of-work system works guarantees that a bunch of those miners will be burning a lot of cycles racing to verify transaction blocks but failing to do so.

idk I don't understand it but I instinctively loathe it

you guys know what MNIST is? it's 9.5MB of data. it would cost ~$100,000 to put that on the blockchain. so quite aside from the fact that the cryptocurrency applications are a scam, the actual practical distributed storage situation is ... not good.

Someone compiled a list of instances of AI doing what creators specify, not what they mean: https://t.co/OqoYN8MvMN pic.twitter.com/I1bgCFfO8c

— Jim Stormdancer (@mogwai_poet) November 7, 2018

this is why you should never tell your robot to cure world hunger

http://static.tvtropes.org/pmwiki/pub/images/tumblr_lm77uuW5G71qb98uxo1_500-2_dragged_6225.jpg

on the specific issue of deepfakes starting a war, i didn't find this essay as concretely comforting as i hoped i would, but it was interesting...

https://reallifemag.com/faked-out/

As long as mass media has existed in the West, there have been complaints about social acceleration, uncertainty, and the loss of a real, knowable world. In other words, our current conversations about the loss of reality are familiar; while each writer attempts to sound innovative, the concerns are evergreen. If the term “infocalypse” is useful, it is as a synonym for modernity, where truth is always two decades ago and dying today, and a new dark age always on the horizon.

i guess the two things i think are

1) infocalypse looks better on a page, but infopocalypse rolls off the tongue a little easier

2) as the article itself says near the end:

This analysis echoes philosopher Georges Bataille’s notion of “nonknowledge,” that the creation of knowledge always implies a corresponding creation of new fields of ignorance. Every revolution in information is also a revolution in misinformation. Information isn’t a light that shines down to answer questions; it also produces new unknowns, and possibilities for the unknown. Giving people more data and more information is as likely to create new readings and narratives than to align or “correct” beliefs. As Rob Horning recently put it, “any information, no matter how damaging it may seem to a particular side, can be put to any political use by any side; in fact no fact has any intrinsic meaning.” With each of modernity’s new technologies we are brought to recognize anew that belief precedes data rather than follows from it.

i don't know how to quantify the amount of information that is out there. not only the amount, but our ability to create it and share it, at a level of quality where an invented lie can more or less pass as something that could be true. but it seems like at least a metric shit ton to me. concerns about truth (and the ability capability to manipulate it) aren't new, but the potential scale for abuse seems much larger (and worrisome) now than ever before.

^^^

Someone took the neural net Yahoo trained to recognize NSFW pics and ran it "backwards" to generate images:

https://open_nsfw.gitlab.io/

whoa

the desert ones are really good

i was just gonna say that. very surreal

these are like, no based on specific images right? just representations of the schema the AI generated to recognize deserts, beaches, genitalia, etc? and that is based on like identifying patterns across billions of images right?

so we are looking into the dreamwork of AI, i guess. if i recognize what is happening, which i may not because this involved equations.

i think that's generally, right, although i must admit i skipped past all the text and went straight for the desert penis pics

god that shit is horrifying, reminds me of the artwork for Chris Cunningham's Rubber Johnny

it's extremely uncanny. a machine's vision of pornography, which doesn't even render actual bodies.

this is from someone with a dog in the race, and the numbers are shaky, but the basic point seems sound to me: no autonomous vehicles any time soon, except in very simple situations (or, if the super rich are able to swing it, unless we establish legal no go zones for non-autonomous vehicles)

https://medium.com/may-mobility/the-moores-law-for-self-driving-vehicles-b78b8861e184

That's a nice combination of simplified explanation and corporate self-promotion. I bet it is based on a presentation that CEO has given to dozens of VCs and banks.

I don't doubt any of it, but it does seem like a pretty convenient way for him to say essentially "It's okay that my company's cars are bad at autonomous driving (and aren't getting better fast enough), because everyone else's are too!"

yes, stipulated in my post (and his!). doesn't change the fact that he's right: it will be many decades before self driving cars actually work in the way typically claimed in the human environment we live in today.

I realize that, if this article were attempting to be rigorous, it wouldn't derive a 'Moore's Law for Autonomous Vehicles' by citing two data points for autonomous cars and one data point for human-driven cars and graphing them. It's miles from any kind of rigor. But what it suggests is more convincing to me than the claims that such fully autonomous vehicles will be arriving much sooner.

Between human performance (10⁸ miles per fatality) and the best-reported self-driving car performance (10⁴ miles per disengagement) is a gap of 10,000x. Put another way, self-driving cars are 0.01% as good as humans.

AYFKM

This ... only works if every disengagement would have resulted in a fatality.

I mean this is all messy stuff, I'm not saying his time estimates on general self-driving are wrong. But if Waymo removed its safety drivers early I think we'd see a lot more stuck vehicles than fatalities.

trump otm

In conversations on Air Force One and in the White House, Trump has acted out scenes of self-driving cars veering out of control and crashing into walls. https://t.co/RMUC3kcXKR

— Jonathan Swan (@jonathanvswan) March 17, 2019