This shit is fucking impressive. One of the few areas of computing that is rapidly advancing these days.
Machine Learning/Neural Networks/AI
Other urls found in this thread:
research.nvidia.com
twitter.com
I'm a SQL developer and I don't even understand this shit
If you only write SQL, shouldn't you be called a DBA?
GAN begin from 2014,NVIDIA(Karras 2017) having a lot power GPU.
research.nvidia.com
bump
here is a network I trained
That fake face is so handsome, I bet that NN is shallow. Lol!
If only we could generate faces that make people freak out because they remind us of demonic entities or nightmares. Imagine a generated picture so scary that you shit your fucking pants when you see it.
Not really, the main driver for ML is cheap GPUs. The techniques are mostly 10-20 years old
Does statistics have anything to do with SQL?
Pretty cool. Do you have more? How long did it take you?
SQL is often used for exploring data.
this is another type of model I trained
...
What did you train it on? Apartment interiors?
porn
What am I looking at here?
lol
GAN is new, fuck off. The only thing that's old is gradient descent.
>wow last 200 years of advancements in math are all irrelevant because addition existed before that.
It's neural network that learns on a large set of images and becomes able to create a new one, similar to those it learned.
That the guy in the image doesn't exist, it's just a computer generating an image of what it thinks a person looks like?
because big data and processing power is available
all developments which before were not feasible due to lacking data or power imo so hes partly correct with GPUs
Anyone got any spooky AI stories?
At Facebook they got two AI chat bots to talk to each other but over time they started using more and more twisted version of English, after a while, it was impossible to figure out what they were saying, so they had to be shut down.
Presumably, yes, a person like that has not been seen by the neural network.
>hes partly correct with GPUs
He's 100% incorrect because he made a 100% incorrect statement
>The techniques are mostly 10-20 years old
Why would you be expected to understand this shit if you're just a lowly SQL developer?
The spooky part, of course, is that bots simply broke down and stopped being useful, but media presented this as if bots became too smart and decided to use code language to communicate.
so half of his statement are 0%?
fuck off faggot
when will someone write a quality waifu desktop assistant that learns with experience?
His statement is incorrect. It's as simple as that. GAN is a new development and almost every good new neural network paper that comes out is not good because its author was able to use a lot of GPUs, but because he introduced some concept.
If you actually develop SQL, cool.
If you mean you use SQL, that is not development.
Because ai hasn't actually approved. Data sets have just become larger which is only good for very specific things like voice recognition but bots can't follow stories anymore today then they could 20 years ago. So everything made will end up sounding like a bot
this better be bait, fucking arrogant crud monkeys shit up /dpt/
>2015
JUST
not spooky but Tay was pretty interesting and had implications for limiting the potential thoughts of future AIs
The main lesson from Tay was "pick your training set carefully". Anyone who has taken one intro class on ML could have told you that.
AI definitely has improved. Not just on the data sets.
There are many examples. AlphaZero -if they didn't cheat- is one of many examples we see now.