What happened to deepfakes?

Waiting on GPUs to hit 3nm with chiplets so that they can deliver roughly 2-3x performance over current in the same TDP as current. So that you can throw the DNN models at that performance threshold and massively improve the fakes.

roll

>brain is clearly digital
>let's call it analogue because lol

rolling

REEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE

Attached: flat,550x550,075,f.u1.jpg (550x550, 28K)

>Deepfakes will be viable in a maybe not-so-distant future where quantum computing is solved

Not really. It'll be viable when researchers come up with better methods for it.

Serious people were never actually really bothered about it because you can already shop/manipulate videos. Deepfakes just makes it easier/faster.

Rollan

kek, rip

How odd...still rollin.

Amy Adams a cute

celebrities don't get me wet.

What do you mean no one talks about them? There's still a lot of activity on a main github project, github.com/iperov/DeepFaceLab and you still see the occasional OC from reddit or elsewhere.

rooling

youtube.com/watch?v=NPxZjZks7bw
It's great

I beg to differ. Just like gender whereby people identify as a range of values with infinite variants in between a plethora of characteristics, the brain operates the same way.

I honestly want the 0

young johnny depp and jensen ackles make out deepfake when?

they take too long and most can't be bothered to do them, and there's no guarantee that the end result will be good anyway.
it's a good start to this sort of tech, but it has a ways to go. as of now, it's just a hobbyists thing.

anyone have resources on this? last time i tried googling it was all censored

duckduckgo it

>duckduckgo it
Don't even try to make it into "Google it".
Say it the proper way, "Go Duck, yourself"

My mistake, sorry user.

We still have the occasional thread on /gif/

amy is so pretty.

Roll.
Btw redpill me on buying Jow Forums pass

Yesterday someone was on the radio talking about how they're being used as a defence in court
>I never said that! It's a deepfake!
The interviewer was all
>But I'm sure we have the super advanced technology needed to detect them, right? I mean, computer geniuses lol!
The person being interviewed basically told him "No, stop living in a bubble where everything's fine and justice prevails, you might be able to prove something's a fake, from manual examination, but if someone says it's a fake, and you have to prove it's real, you're fucked"

all the kids who were playing around with that technology got CIA'd and are probably in gitmo.

>It'll be viable when researchers come up with better methods for it.
Pic related.

Attached: maxresdefault.jpg (1280x720, 146K)

24 hours for 5-10 minutes is pretty damn good actually. How long of a video do you really need? And why do you need to mass produce them unless you are making money off of them, in which case a render farm will greatly reduce render time or at least allow it to be parrallelized.

That being said it still looks off at certain points which is probably the biggest issue. It's close but it's not close enough to a lot of people.

>he doesn't know it's already being used by all the major players
It's all over yt if you open your eyes, it's easy af to spot.. The shittier easier to spot ones have skin tone imperfections around the neck line.

Attached: these faces are AI generated and not real.jpg (1536x1956, 348K)

This gives me an idea
>Deepfake parody videos

Nah, thats unrelated. Also incredibly misleading clickbait garbage. Deepfakes results are unpredictable and can't scale, as big mouse showed in their star wars films.

...

Do not buy.

that ai must have had only 56% accuracy

Funny first nvidia GAN use dataset hollywood actors, but because misrepresentation Nvidia + Flickr build new dataset with average persons

What? When they first became a thing I had a 380 and I could make them perfectly fine on Linux with the guides that were posted.

Can't wait for GAN based JAV

I really will love made GAN with pretty girls instagram, but no GPUs, maybe in 3 to 5 years

>Unlimited cuties generator

Probably because the press is the largest user of it. Fabricating huge headlines just got simpler.

There is a rumour that actors play multiple acting roles, of both genders.

Attached: Kenny Powers.jpg (1200x675, 55K)

What's the current popular tool for plebs who just want to gather a few GBs of mugshots and have a bit of fun with them? It seems like all the information about it just vanished.

Attached: Screen_Shot_2017_07_13_at_1.09.20_PM.0[1].jpg (1200x800, 162K)

People shitpost about sex robots like they did about flying cars, but this is where the future of adult entertainment is at.

It's exactly where it's always been (hint: not reddit) but normies realized it was tough.

>brain is clearly digital
No it's not. Neuron firing rate is a very analog phenomenon.

ok

E THOTS BLOWN THE FUCK OUT
>youtube.com/watch?v=isjQgTuUnwU

rolly

Shouldn't you be discussing how to make them, instead of where to find them on youtube? When did Jow Forums flood this shitpit of a board with brain damaged 14 year olds?

Well, do you have any sources and references or will you just present a re-hashed version of the /leftypol/-boogeyman meme?

lol
The US is so trash.

pft, wait till 2020.
there's going to be lots of political deepfakes.

It's somewhat outdated and written with /gif/ users in mind, but take a look at pastebin.com/btEBfEyX

rooooooollllliiinngggg

P-pretty cool user!
Note that I don't really know anything about computers or even deep-learning/AI though, I don't even know how to compile anything, heck, I don't even use linux.
But I am working on it, thanks for your reply!

>Note that I don't really know anything about computers or even deep-learning/AI though, I don't even know how to compile anything, heck, I don't even use linux.
Neither does your average /gif/ user.
Btw DeepFaceLab offers builds for Windows. No compiling from source, not even setting up a Python environment. It doesn't have a GUI, but you can execute it via a bunch of included batch scripts.

these people are so ugly you just KNOW that all NVidia did was scrape photos from Americans' Facebook pages to get a training set

You are underestimating how little I know.
I can't even into command prompt. But thanks for your reccomendations, before you reply any more, note that I am a 20 year old boomer and just came to Jow Forums out of interest and the hope that I would find some thread that had infos on how to install void, since I want to get into linux and get my old netbook back and running so I can use it as an airgapped system.

i always wondered what if some of the "deepfakes" are actually real and the "real ones" are actually faked like some of the emma watson stuff?

Roll

roll i love nick so i cant lose *dabs*

What does that have to do with what I said?

looks like robin from himym

just duck it

>that deepfake with Dillion Harpers face being replace with seth mcfarland

after seeing that, i couldnt go through another deepfake thread on /gif/.

The signal is a binary signal once you pass the threshold. It is firing rate, not e.g. firing intensity, that decides outcome.

it's such an interesting tool in the hands of such boring people; making them about actresses doing porn got old after the first ones

Wasn't the complicated thing about the brain the chemichals? If it was this simple wouldn't we have a better understanding of it?

Arbitrary chemicals can bind to various "ports" in neurons, causing the gate level to go up or down (or preventing reception of normal triggers, or acting as artificial triggers). But under normal circumstances, it's very simple with very few interactions involved and it's straightforward.
The complications arise because it's not easy to observe neurons on a good enough level to see the mechanics by which the gating happens. There are complex interactions between neuronal clusters that cause emergent behavior (e.g. neurons in the cortical column mimic sobel operators and form stacked blockwise operators). Moreover, there are many different types of neurons, but it's hard to tell which is which with conventional instruments with certainty. Other problems include determining the connectivity scheme between neurons (often, we have to rely on correlation, which as you might imagine is far fetched for an approximation at best).
Macro mechanisms are very complex to even define, let alone identify in the brain (e.g. memory). Moreover, recent results suggest some of the tasks we thought the brain was the sole involved party in mediating since forever might actually also be mediated by the entire cellular (non-nervous) body.
To this, add the fact you never observe perfect brains, you instead observed randomly flawed brains. Not to mention it's not clear how many variants of a flawless brain can exist.

>The signal is a binary signal once you pass the threshold
No it's not. Firing rate matters, in an analog manner.

Firing rate is an analog abstraction for a digital phenomenon. That's exactly my point: the firing rate matters, but firing rate is not analog.

It got shutdown by you know who.

Firing rate matters, but firing is not analog*

But firing rate is analog. A neuron is not just firing or not firing, but firing at some particular rate, and the interval is continuous, just like an analog signal is.

It IS in fact just firing or not firing. What causes it to fire at a specific rate is the dynamics of input -> potentiation -> rest -> steady -> input which is a much more macro phenomenon. The firing rate does not in and of itself affect other stuff, but until the receiving neurons pass their potentiation threshold, they will effectively accumulate input. That is where the firing rate becomes meaningful.

Gimme Nicholas desu

Thanks, really insightful post.
How fast do you think things are advanving in the field? Could we hope to get full understanding in our lifetimes?

It's not going anywhere near as fast as it could. I mostly attribute it to the lack of penetration of modern computational methods and automated tools (because they could help bypass issues with instruments and experimental methods). However there are also methodological issues (very ad-hoc experiments and hasty conclusions based on ideas at the beginning of the field's history that were never verified or put into question tend to rule the field). At the same time it's one of the more advanced fields in the life sciences in terms of acceptance of modernization.
Even if neurosci modernized overnight to integrate with current SOTA tools, we would still not reach a complete understanding of a human brain in our lifetime by a long shot, but I think it's possible to reach the level of a drosophila's brain for some value of 'understanding', and assuming we aren't completely wrong about the role of the brain in the first place.

What are those SOTA tools you mention? I cannot find it with google.
And what do you mean we could be wrong about the role of the brain?

For example, white matter segmentation is still done with random forests, crfs, or similar methods, whereas convnets would be infinitely more appropriate. It's the general theme: "we have to solve pathfinding, so let's use a greedy algorithm because we've never heard of dijkstra before" kind of thing.
As for role of the brain, probably the most convincing recent talk is
youtube.com/watch?v=RjD1aLm4Thg&feature=youtu.be&t=2330
which is some high-level overview of a single aspect thereof.

Attached: nice.gif (444x250, 3.04M)

Rolling for Nicholas Cage.

>it's a "retards think AI will simulate human brains" thread

AI will simulate dog brains and everyone is going to get a robot puppy

Idk what people are whining about in this thread. I regularly make DeepFakes for my own peepee touching times on my 1080Ti. Not a problem, and you're generally looking at a convergence time of 6 - 8 hours assuming that you reuse the same model for the same B target. It comes down to some pretty specific factors in the training sets and sets you want to replace, but you can generally make some pretty good content with a small amount of effort if you understand what you're doing.

It could be that the first AI will be a brain simulation, i.e. because no approximation model exists or can be found to be sufficient for imitating human principles (e.g. the ability to learn efficiently, or behavior and emotions, etc.)
That does not mean that's why anyone interested in AI is even remotely considering working on though. Regardless, your post is pointless.

Fuck.

>That does not mean that's why anyone interested in AI is even remotely considering working on though
It literally is the publically stated reason many people are working on AI, and it's a giant meme

t. literal inbred

the truth hurts I guess