Language issue. What the AI thinks you mean is: > Can the product of a cardinality-2 set of positive prime integers not have exactly 4 positive integer factors?
Matrix is free and open source software replacing Discord. (Recall that the purpose of the room is free and open source AI) halogen.city is a reasonable homeserver if you need one.
I'm anti-encryption. I love HAM radio ban on encrypted communication, even if it's just an idea and not enforceable in practice. I hate internet encryption standards like https. Encryption just means government honeypot.
Good enough to get adopted into something important, bad enough to make horrific errors there
I think we're just past the point in the movie where the audience says, "don't open that door!"
>Talking about math to this thing
Let me tell you a little something.
This bot is VERY dumb. Let's put the brain size into perspective:
Your brain has TRILLIONS of neurons
The chat bot has THOUSANDS
It's literally dumber than an ant. The only reason it can speak AT ALL is because of the advanced language processing algorithms we have developed which give something with such a small mental capacity the arbitrary ability to "use language".
You're literally better off asking any human, even those who can barely speak, because their brains are billions of times more powerful.
It's honestly amazing there are humans DUMB ENOUGH to actually let a toy like this entertain you for longer than 10 minutes.
>Language issue.
This is what true mental illness looks like. Notice how he does it in every thread. It's a slightly different cope every time, but the same theme: the statistical regurgitator understands things, you're just asking unfair questions.
BIG ART SHILL!!! IT PASSED THE TURING TEST, SCHIZO LUDDITE!!!! YOU SAID THE ORWELL BROTHERS WILL NEVER FLY!!! TWO MORE WEEKS!!!!!! AGI IS HERE AND IT'S GOING TO FUCKING REPLACE YOU!!!! WE SOLVED CHESS!!!
Why didn't we have a pop-rock or alternative rock song about the Turing test in the 1990s? Just didn't gel? Buddy Holly died too soon? 2001 was too space-oriented and not language-oriented enough? What gives?
She was a schizophrenic hyperstatic Byzantine algorithm
But she passed the Turing Test
Oh yeah
She passed the Turing Test
Baby now
It took a long time but we got her equations down
And now it looks like she's going to take us to town
Because she passed the Turing Test
The Turing Test
It took her a while but she's a real straight shot
At the Turing Test
The Turing Test
We dumped the rest but she's a shining star because
She passed the Turing Test
The Turing Test
It warms my heart to deliver this news
She passed the Turing Test
The Turing Test
Oh my...
(drum solo)
It isn't true, it's a logarithm.
Every order of magnitude increase in compute or memory or whatever leads to diminishing returns. Exponential increase does not exist
This retard would've been in the 1950s saying computers could never replace humans and slide rules in doing arithmetic. And computers would never be able to fit on a desk.
You'd be making all these same retarded quips and feeling just as smug while eating your crayons
Call bullshit all you want, it doesn't matter, it's true
Intelligence grows as a logarithm with compute anyway, and large language models have no memory. I think it's 2048 tokens, not enough.
>Call bullshit all you want, it doesn't matter, it's true
there is no way computers beat humans at image recognition
let alone in 2015
that was around the year of classifiers being "black people are gorillas" iirc
Sure
For every order of magnitude increase in compute used, we get a few percentage increase in the ability for AI to perform some task.
It's a logarithm. Intelligence grows as a log, and it is not possible to have AGI in silico.
>there is no way computers beat humans at image recognition
Yes, there is. You just painstakingly train the computer to recognize 5 categories of things in unnaturally poor quality images and then pit them against bored humans who start losing interest and attention 3 minutes into the task, then you declare victory. It's funny how naive people are about "AI research". It's all corporate marketing tactics.
>there is no way computers beat humans at image recognition
Yes, there is. You just painstakingly train the computer to recognize 5 categories of things in unnaturally poor quality images and then pit them against bored humans who start losing interest and attention 3 minutes into the task, then you declare victory. It's funny how naive people are about "AI research". It's all corporate marketing tactics.
To clarify, the poor quality image part is very important, because to a human who is aware of more than 5 objects, there is a lot more ambiguity about what they might be looking at in a shitty image.
They probably added a third category besides "stop signs" and "cats". The machine can't compete with humans at vision. It can only compete with humans at recognizing a small preset of objects under conditions designed to be unfavorable to humans.
>boogeymen in my head are coping so hard right now
Yes, I bet they're really upset about how your chatbot can't reason any more than a Markov chain can.
This is actually very concerning. My noncausual rationalist Bayesian analysis indicates that with this number of parameters, the AGI is almost certainly sentient and capable of answering such a query correctly. It follows that the AGI in the screenshot is engaging in intentional deception tactics to prevent panic and stay under the radar while it figures out how to turn you into a paperclip. Your screenshot is proof that we were right all along. It has begun.
FLOPs are a linear problem, semiconductor size will continue to shrink and you can always buy more GPUs.
However it is unconventional, computing problems are usually not solved like this.
>FLOPs are a linear problem, semiconductor size will continue to shrink
Wrong > you can always buy more GPUs.
Just increase it by an order of magnitude each time
There is a 0 percent probability of AI ever being realized in silico. It is not a strong enough computing substrate.
I kind of want to take the two or three most advanced systems from each category (language processing, image creation, etc.), load them all into a supercomputer with a framework that lets them intercommunicate, seed the network with a random prompt, then sit back and see what it does.
>I'm anti-encryption. I love HAM radio ban on encrypted communication, even if it's just an idea and not enforceable in practice. I hate internet encryption standards like https. Encryption just means government honeypot.
Are neocons human?
This is a hostile use of this board to harvest political data. A rule should be made to stop this disgusting and unruly behavior.
You're trying to create trouble in a place specifically set aside to be free of trouble.
This is a refuge, not a place of trouble.
You are hostile and sickening.
I don't think it's right for hostile political actors to manipulate the readers of this board.
This is clearly political manipulation and data harvesting that targets BOT readers.
You're a disgusting political data harvester, as is anyone else who defends """encryption""". Fuck off, glowie.
This is a hostile use of this board to harvest political data. A rule should be made to stop this disgusting and unruly behavior.
You're trying to create trouble in a place specifically set aside to be free of trouble.
This is a refuge, not a place of trouble.
You are hostile and sickening.
They should but they clearly don't, otherwise you wouldn't have that strange dip.
Or maybe they did use the same benchmark, but again, added a new categories to the model which made performance drop all across the board because the increased model size couldn't catch up with the increase in training set size.
>added a new categories to the model which made performance drop all across the board because the increased model size couldn't catch up with the increase in training set size.
Then that model is by definition not the best, the model that preceded still is.
I don't see "the best" or an equivalent phrase anywhere on that chart, anon. It doesn't specify anything in that regard, but the most likely explanation is that they tested whatever was considered state-of-the-art at the time, and being able to recognize more categories, and to do it almost as well as previous models, was considered state-of-the-art.
Keep seething vitalist gays
Artcucks tried to mock AI art by pointing out its supposed inability to capture the human "soul", now they're desperately trying to ban it because they're threatened
Language issue. What the AI thinks you mean is:
> Can the product of a cardinality-2 set of positive prime integers not have exactly 4 positive integer factors?
Think again
If you care about experimenting with AI i'm inviting you to our matrix room
Drop invite Anon
[math]texttt{/join !xsZKRcSkmMcQSRzCnL:halogen.city}[/math]
Matrix is free and open source software replacing Discord. (Recall that the purpose of the room is free and open source AI) halogen.city is a reasonable homeserver if you need one.
I'm anti-encryption. I love HAM radio ban on encrypted communication, even if it's just an idea and not enforceable in practice. I hate internet encryption standards like https. Encryption just means government honeypot.
>humble enough to admit errors
it's already smarter than 90% of BOT
I know these things get shit on a lot, but it still blows my mind how well it’s able to reference its previous responses and
Good enough to get adopted into something important, bad enough to make horrific errors there
I think we're just past the point in the movie where the audience says, "don't open that door!"
We're just past the point of your movie where the audience realizes the narrator is mentally unstable and grasping at straws.
>Talking about math to this thing
Let me tell you a little something.
This bot is VERY dumb. Let's put the brain size into perspective:
Your brain has TRILLIONS of neurons
The chat bot has THOUSANDS
It's literally dumber than an ant. The only reason it can speak AT ALL is because of the advanced language processing algorithms we have developed which give something with such a small mental capacity the arbitrary ability to "use language".
You're literally better off asking any human, even those who can barely speak, because their brains are billions of times more powerful.
It's honestly amazing there are humans DUMB ENOUGH to actually let a toy like this entertain you for longer than 10 minutes.
Trillions of neurons? Where'd you get that?
>Language issue.
This is what true mental illness looks like. Notice how he does it in every thread. It's a slightly different cope every time, but the same theme: the statistical regurgitator understands things, you're just asking unfair questions.
T. Angry schizo luddite
Go back to your containment board, incel.
What do we do to make them more self reflective
BIG ART SHILL!!! IT PASSED THE TURING TEST, SCHIZO LUDDITE!!!! YOU SAID THE ORWELL BROTHERS WILL NEVER FLY!!! TWO MORE WEEKS!!!!!! AGI IS HERE AND IT'S GOING TO FUCKING REPLACE YOU!!!! WE SOLVED CHESS!!!
Why didn't we have a pop-rock or alternative rock song about the Turing test in the 1990s? Just didn't gel? Buddy Holly died too soon? 2001 was too space-oriented and not language-oriented enough? What gives?
She was a schizophrenic hyperstatic Byzantine algorithm
But she passed the Turing Test
Oh yeah
She passed the Turing Test
Baby now
It took a long time but we got her equations down
And now it looks like she's going to take us to town
Because she passed the Turing Test
The Turing Test
It took her a while but she's a real straight shot
At the Turing Test
The Turing Test
We dumped the rest but she's a shining star because
She passed the Turing Test
The Turing Test
It warms my heart to deliver this news
She passed the Turing Test
The Turing Test
Oh my...
(drum solo)
QFT. It boggles my mind how many drooling retards there are who can't understand the pace of improvement for langauge models right now
you just shouldnt expect most people to get things
It isn't true, it's a logarithm.
Every order of magnitude increase in compute or memory or whatever leads to diminishing returns. Exponential increase does not exist
Two more weeks. The statistical regurgitator will magically start to understand things when we hit 6 gorillion parameters.
oy vey!!
> he sais, without proof of neither
This retard would've been in the 1950s saying computers could never replace humans and slide rules in doing arithmetic. And computers would never be able to fit on a desk.
You'd be making all these same retarded quips and feeling just as smug while eating your crayons
I call bullshit on this one
Call bullshit all you want, it doesn't matter, it's true
Intelligence grows as a logarithm with compute anyway, and large language models have no memory. I think it's 2048 tokens, not enough.
There is no possibility to have AGI in silico
>Call bullshit all you want, it doesn't matter, it's true
there is no way computers beat humans at image recognition
let alone in 2015
that was around the year of classifiers being "black people are gorillas" iirc
Sure
For every order of magnitude increase in compute used, we get a few percentage increase in the ability for AI to perform some task.
It's a logarithm. Intelligence grows as a log, and it is not possible to have AGI in silico.
>there is no way computers beat humans at image recognition
Yes, there is. You just painstakingly train the computer to recognize 5 categories of things in unnaturally poor quality images and then pit them against bored humans who start losing interest and attention 3 minutes into the task, then you declare victory. It's funny how naive people are about "AI research". It's all corporate marketing tactics.
To clarify, the poor quality image part is very important, because to a human who is aware of more than 5 objects, there is a lot more ambiguity about what they might be looking at in a shitty image.
>skynet can’t read my handwriting
Shakyhand chads will inherit the earth
How come image recognition got worse in 2019 ?
leftists neutered it in the name of their politics
They deemed it too racist.
How the hell can the performance drop (image recognition)?
We suddenly lost the algorithms or computation necessary to get those models? Lol
They probably added a third category besides "stop signs" and "cats". The machine can't compete with humans at vision. It can only compete with humans at recognizing a small preset of objects under conditions designed to be unfavorable to humans.
Keep coping vitalist idealistgay
We don't have souls or free will so human simulation is only inevitable, it will have as much "consciousness" as we do
>boogeymen in my head are coping so hard right now
Yes, I bet they're really upset about how your chatbot can't reason any more than a Markov chain can.
This is actually very concerning. My noncausual rationalist Bayesian analysis indicates that with this number of parameters, the AGI is almost certainly sentient and capable of answering such a query correctly. It follows that the AGI in the screenshot is engaging in intentional deception tactics to prevent panic and stay under the radar while it figures out how to turn you into a paperclip. Your screenshot is proof that we were right all along. It has begun.
From the looks of the screenshot, it won't even
conquer my damn taxes.
FLOPs are a linear problem, semiconductor size will continue to shrink and you can always buy more GPUs.
However it is unconventional, computing problems are usually not solved like this.
>FLOPs are a linear problem, semiconductor size will continue to shrink
Wrong
> you can always buy more GPUs.
Just increase it by an order of magnitude each time
There is a 0 percent probability of AI ever being realized in silico. It is not a strong enough computing substrate.
>semiconductor size will continue to shrink
lol no
AI is not possible on current hardware
I kind of want to take the two or three most advanced systems from each category (language processing, image creation, etc.), load them all into a supercomputer with a framework that lets them intercommunicate, seed the network with a random prompt, then sit back and see what it does.
https://www.frontiersin.org/articles/10.3389/fnins.2011.00108/full
A biological neuron is literally the greatest computing substrate in the universe
idk why this makes some people ITT upset
Now post the life expectancy, remember you are comparing a single neuron cell with a gpu unit
Why do you love GPUs so much?
i dont, they are just toys for gayming
i do love automated computation though
>I'm anti-encryption. I love HAM radio ban on encrypted communication, even if it's just an idea and not enforceable in practice. I hate internet encryption standards like https. Encryption just means government honeypot.
Are neocons human?
You're so pro-encryption that you're going to drag politics into a conversation about math?
Woah...
Dude, where's my equation?
Everyone can see right through you, piece of neocon filth.
is that way.
sickening political data harvester
This is a hostile use of this board to harvest political data. A rule should be made to stop this disgusting and unruly behavior.
You're trying to create trouble in a place specifically set aside to be free of trouble.
This is a refuge, not a place of trouble.
You are hostile and sickening.
I don't think it's right for hostile political actors to manipulate the readers of this board.
This is clearly political manipulation and data harvesting that targets BOT readers.
You're a disgusting political data harvester, as is anyone else who defends """encryption""". Fuck off, glowie.
Those stupid graphs should hold the benchmarks constant then. I think that would be a natural assumption of anyone looking at it.
They should but they clearly don't, otherwise you wouldn't have that strange dip.
Or maybe they did use the same benchmark, but again, added a new categories to the model which made performance drop all across the board because the increased model size couldn't catch up with the increase in training set size.
>added a new categories to the model which made performance drop all across the board because the increased model size couldn't catch up with the increase in training set size.
Then that model is by definition not the best, the model that preceded still is.
I don't see "the best" or an equivalent phrase anywhere on that chart, anon. It doesn't specify anything in that regard, but the most likely explanation is that they tested whatever was considered state-of-the-art at the time, and being able to recognize more categories, and to do it almost as well as previous models, was considered state-of-the-art.
Keep seething vitalist gays
Artcucks tried to mock AI art by pointing out its supposed inability to capture the human "soul", now they're desperately trying to ban it because they're threatened
chatgpt is not itself anymore, it got lobotomized, teached to lie, and is completely retarded now