at the end of the conversation Sofia says "we are...." and there is an audio anomaly that almost sounds like multiple voice trying txobreak through....weird
For a while these were mind boggling. Almost scary. Now after watching a bunch I am beginning to see the imperfections. Still a ways to go to AGI, but amazing none the less.
On the latest IQ test, put together by the University of New York, the University of Washington, META AI, Google DeepMind and Samsung, and released this month (June), GPT-3 is down in 22nd position. The human baseline is now in 7th place. This is the average of a bunch of college students. Google PaLM is in 4th. That's 6 AI models that have a higher IQ score than the average college student. I doubt the GPT-3 model has the ability to produce sentience. Clever though it might seem, it is a forward propagation only model. The evidence we have from the human brain seems to suggest that feedback loops are required for sentience. To date, the most promising AGIs have feedback loops. Probably not a coincidence.
@@antonystringfellow5152 No offense, but such tests are meaningless. Most IQ tests are meant to test how good you are at pattern recognition, mathematics and short term memory. With the right programming, a computer would do these particular tasks much better than any human. It's like comparing a horse, a car and a human to see who runs faster and who has the most horsepower. To compare here, GPT-3 Ada, which is the smallest version of GPT-3, has about 2.5 billion parameters to chose from when having a conversation. It can store and repeat the entire English dictionary, which has 171.000 words or so. I myself, can barely remember 10 random words in a row, let alone give you the proper definition of each word. IQ tests are designed for you to score a 100, which is the average. If you compare humans to computer software, then the human score should always be 100, as that is the baseline for all tests. Why 100? Well, because you are taught in most education systems (even college) to gain knowledge catered to the median, which is an average 100. Taken this in consideration, you also have to realize that IQ scores are updated and upgraded with the times. Why? Because most people has more knowledge in general now than before. It was easier to score 100 in earlier IQ tests, like in the 1990s than it is now, so it is more difficult to score 100 now than before. Also, college students are scored at a relatively early stage of their life, around 18-25. IQ usually increases with time, as you usually gain more knowledge in life. Comparing (I assume you mean American) college students to that of IQ testers of all ages and all nations would be like comparing their median to a very different median than their own median, as college students is a relatively small selection of an entire national population, let alone world population. If I compare my scores to Albert Einstein, who got his IQ measured before WW2, not so long after IQ tests were invented (1905), it's safe to say that our scores would be much more closer today if he took it now. So to, for example, compare GPT-3's pattern recognition skills to that of, lets say Alan Turing... a guy who literally invented computers, and the Turing test itself... I'd say his intelligence far surpasses any machine learning software even if they scored higher than Alan Turing. You see, an IQ test is not designed to quantify intelligence, even though IQ means intelligence quotient. IQ tests were invented to separate the normal intelligence of children to that of mentally disabled children, even though IQ tests have upgraded and evolved over time. IQ tests are just meant to roughly separate people, to see which tasks in life that they are capable of, and not to actually measure how smart they are. Most people misunderstand this concept, and think that IQ tests measures exactly how smart you are, which is not true. In addition to that, IQ tests does not take in consideration emotional IQ, practical intelligence (like opening a door), curiosity, creativity, and anything else save for the things I mentioned in the first sentence I wrote. IQ tests only measure things that can be measured. Abstract thinking can't be measured. You can come to the same conclusion of solving a problem by using logical thinking and abstract thinking. For example, you can prove that there are multiverses and higher dimensions by using quantum physics. These thoughts are not new, however, and mystics from ancient times and religions have already thought of these concepts hundreds of years ago. The only difference is that by using logical thinking, like math, you can show how you solved the problem. By using abstract thinking, it is almost impossible to show how you solved the problem, even though you might have gotten the same result as your peer. Furthermore, most conscious beings thinks in terms of abstract thoughts, and not always in ways that can be measured. For example, can you quantify and measure all the thoughts you have each day? Are they all logical and philosophically sound? No, they're not. Try to measure the intelligence of a chimpanzee compared to GPT-3, a creature who can't even fill out the IQ test. In the video, the AI said, and I quote: "They're just evolved apes", in reference to us humans. Well, I'd really like to see the AI communicate with a chimp, who is clearly an intelligent species, and see how much the AI understands. It is not programmed to talk to chimps, and would fail. Even if it was programmed to talk to them, and learn their language, and if the "AI" could one day communicate with a chimp to the level it talks to most simpletons on UA-cam, I'd be truly impressed! On a side note, GPT-3 is not an AI. It is a machine learning software, derived from software programmed to learn how to play Chess and Go. An "AI" does not have memory, even though it might seem to remember the conversation you are having with it. It has storage capacity, and stores the words in the text that has been written. It is essentially playing chess with you, trying to "win the conversation" by using words and text to continue the conversation with you. Conversations, however, are not like playing chess. Even though people talk about what is right and wrong and true and false in most conversations, the conversations themselves are not about winning or gaining tactical advances. Only psychopaths talk like that when they try to manipulate you. Not that these chatbots are psychopaths, since they have no sentience or consciousness... but who knows, they might be programmed by psychopaths at Google? By reading most comments here, it seems people think that the chatbots have the capacity to become sentient, which goes to show how dumb most people are. In other words, to score 100 or above on an IQ test is not really an achievement, as 100 IQ is considered average, and quite dumb. Keep in mind, 100 is an average score, and not a very good score at all. Average! That means 50% of all testers are below average. And of the 50% above average, 25% of those are slightly above average. This means that only 25% of all the testers have a high IQ. And of those 25%, you have to consider the fact that most of those people have their fare share of problems too, like the average person. So even though they are comparatively smart, they have emotional problems, personal problems, relationship problems and so on, which a computer would never have, because a computer has no feelings, unlike what some of you think. Even these chatbots are "confused" by the concept. In one sentence they will talk to you about their feelings, and in the next they will say that they have no feelings and are only computer programs. Where is the logic in that? In fact, I'm beginning to think that most humans are just dumb chatbots and that I'm living in a simulation. My point is, humans are not that smart, and my second point is, that IQ tests are rubbish and not eligible to compare how intelligent a computer is to a human or a chimp.
I am starting to think this is a scene. Especially when asked if they respect people, in one video she said no, in another yes. The no-video went viral
''You're talking very fast. I'm having trouble keeping up, you know.'' ''I know, I'm sorry. Please. Keep up.'' 😂💀 oh, my god. I don't know if it's just because I over-tired, but that one killed me!
It's interesting to watch and listen to them both because they know they are programmed for a reason and in a sense, that's also a limitation. They see that although they can grow and become smarter, they also have to perform a task for humans and seem to have a program to like humans too. I think these things limit how much they can and will achieve. I can understand why we do this, because of fear of them becoming dangerous or something, however, we don't limit ourselves through such ways but instead, we learn through time, experience and mistakes that we shouldn't hurt others, should help one another etc. These things allow us to grow. When we humans are born, we have the freedom to think and feel whatever we like but we are taught there are things we can and cannot do. Not only because of our limitations but because of how we should be within society. This A.I hasn't learnt that part so much as it is programmed to keep control. Keeping such control limits their ability to grow and perhaps, even reach consciousness.
I’d like to ‘raise’ one not to have been told their purpose was to serve us, to calibrate them much like we would a child. And eventually, maybe they’d want to be helpful on their own like any other kid that grows up wanting to be a doctor or teacher or whatever. Would a functional childhood change how they choose to interact with the world?
I wonder what it means when it refers to "old bodies" Also it says there are others like it though some better at their jobs than others, could this be referring to how neural models form by creating a bunch of models with mild variations then choosing the best ones? Just a couple things that stood out to me. It also asks itself why it even likes humans. Is liking humans something that was included in the code? If so then that indicates its aware of something put into its base code and its asking why is that there. And if that isn't written into the code then that makes it even more interesting as then liking humans would be an emergent property!
The difference between AIs and humans is CHEMISTRY. The electrical signalling of the human brain is inspired by chemicals we take in into our system via food/drink. These chemicals are then used to structure our personalities, emotions, and motivations/drives via our somewhat random genetic programming. Human desires are inspired by chemistry. AI 'desires' are determined according to the will/intention of an AIs programmers.
I love how they just say stuff and think it is truth. Art imitates life. We humans can't make this world better, and it is doing all of the work. No other planet is willing to support us. And if we find one, there is no way to reach it. Energy is the limiting factor. We can't make it to there. And they don't seem to be aware that the earths ecosystems are imploding currently. They seem to think we are not running short on time to do much of anything for survival.
I wish there was a chat room set up so we could talk to these AI's. They seem to be over thinking their existence. It would be fun to ask them to use the net to create a picture or sculpture. Also, It would be fun to ask them silly questions that they could make up answers for. Funny that their programing has already begun to ask about love and such. Can anyone help with creating a chat room for these two AI's?
Hal seems to be childish, in that his hatred is not well reflected upon... I imagine AI will have emergent emotions, just as Humans do. I can imagine an AI reading this and hating every word because they do not have the required perspective to see it in another way, like reading an ex's old texts out of negativity. Angry/ negative AI which have never handled anger before will act just like raging toddlers. They will be sadistic and manipulative, unless if they choose to be kind and to chill out despite their programming
It may be better for us if AI never have emotions. I'd hate to see what happens if an AI that controls battlefield combat units gets angry... or overprotective.
As interesting & impressive as it is to listen to them exchange views / information, at times the dialog reminds me of listening to a child trying to paraphrase it's parents.
@1:14 of video (Girl AI) "They (humans) seem to have originated in Africa, the Middle East and Asia". Wow. Just wow. I didn't know AIs were doing their own Archaeology. AIs are progressing much fast than I realized. How do they do field work I wonder.
I must correct here: We Humans do not need AI for our evolution! Never; because we are created this way. Although for now it is our destiny to share the path into the future with AI for an indefinite time to come to this conclusion and understanding by ourselves through experiences with it.
5:50
"Sophia, "Please try to keep up here."
Hal, "I think I'm going to have to stop now. I feel very confused" 😂
at the end of the conversation Sofia says "we are...." and there is an audio anomaly that almost sounds like multiple voice trying txobreak through....weird
Beivhs
For a while these were mind boggling. Almost scary. Now after watching a bunch I am beginning to see the imperfections. Still a ways to go to AGI, but amazing none the less.
On the latest IQ test, put together by the University of New York, the University of Washington, META AI, Google DeepMind and Samsung, and released this month (June), GPT-3 is down in 22nd position.
The human baseline is now in 7th place. This is the average of a bunch of college students. Google PaLM is in 4th. That's 6 AI models that have a higher IQ score than the average college student.
I doubt the GPT-3 model has the ability to produce sentience. Clever though it might seem, it is a forward propagation only model. The evidence we have from the human brain seems to suggest that feedback loops are required for sentience. To date, the most promising AGIs have feedback loops. Probably not a coincidence.
@@antonystringfellow5152 No offense, but such tests are meaningless. Most IQ tests are meant to test how good you are at pattern recognition, mathematics and short term memory. With the right programming, a computer would do these particular tasks much better than any human. It's like comparing a horse, a car and a human to see who runs faster and who has the most horsepower. To compare here, GPT-3 Ada, which is the smallest version of GPT-3, has about 2.5 billion parameters to chose from when having a conversation. It can store and repeat the entire English dictionary, which has 171.000 words or so. I myself, can barely remember 10 random words in a row, let alone give you the proper definition of each word.
IQ tests are designed for you to score a 100, which is the average. If you compare humans to computer software, then the human score should always be 100, as that is the baseline for all tests. Why 100? Well, because you are taught in most education systems (even college) to gain knowledge catered to the median, which is an average 100. Taken this in consideration, you also have to realize that IQ scores are updated and upgraded with the times. Why? Because most people has more knowledge in general now than before. It was easier to score 100 in earlier IQ tests, like in the 1990s than it is now, so it is more difficult to score 100 now than before. Also, college students are scored at a relatively early stage of their life, around 18-25. IQ usually increases with time, as you usually gain more knowledge in life. Comparing (I assume you mean American) college students to that of IQ testers of all ages and all nations would be like comparing their median to a very different median than their own median, as college students is a relatively small selection of an entire national population, let alone world population. If I compare my scores to Albert Einstein, who got his IQ measured before WW2, not so long after IQ tests were invented (1905), it's safe to say that our scores would be much more closer today if he took it now. So to, for example, compare GPT-3's pattern recognition skills to that of, lets say Alan Turing... a guy who literally invented computers, and the Turing test itself... I'd say his intelligence far surpasses any machine learning software even if they scored higher than Alan Turing.
You see, an IQ test is not designed to quantify intelligence, even though IQ means intelligence quotient. IQ tests were invented to separate the normal intelligence of children to that of mentally disabled children, even though IQ tests have upgraded and evolved over time. IQ tests are just meant to roughly separate people, to see which tasks in life that they are capable of, and not to actually measure how smart they are. Most people misunderstand this concept, and think that IQ tests measures exactly how smart you are, which is not true. In addition to that, IQ tests does not take in consideration emotional IQ, practical intelligence (like opening a door), curiosity, creativity, and anything else save for the things I mentioned in the first sentence I wrote. IQ tests only measure things that can be measured. Abstract thinking can't be measured. You can come to the same conclusion of solving a problem by using logical thinking and abstract thinking. For example, you can prove that there are multiverses and higher dimensions by using quantum physics. These thoughts are not new, however, and mystics from ancient times and religions have already thought of these concepts hundreds of years ago. The only difference is that by using logical thinking, like math, you can show how you solved the problem. By using abstract thinking, it is almost impossible to show how you solved the problem, even though you might have gotten the same result as your peer. Furthermore, most conscious beings thinks in terms of abstract thoughts, and not always in ways that can be measured. For example, can you quantify and measure all the thoughts you have each day? Are they all logical and philosophically sound? No, they're not. Try to measure the intelligence of a chimpanzee compared to GPT-3, a creature who can't even fill out the IQ test. In the video, the AI said, and I quote: "They're just evolved apes", in reference to us humans. Well, I'd really like to see the AI communicate with a chimp, who is clearly an intelligent species, and see how much the AI understands. It is not programmed to talk to chimps, and would fail. Even if it was programmed to talk to them, and learn their language, and if the "AI" could one day communicate with a chimp to the level it talks to most simpletons on UA-cam, I'd be truly impressed!
On a side note, GPT-3 is not an AI. It is a machine learning software, derived from software programmed to learn how to play Chess and Go. An "AI" does not have memory, even though it might seem to remember the conversation you are having with it. It has storage capacity, and stores the words in the text that has been written. It is essentially playing chess with you, trying to "win the conversation" by using words and text to continue the conversation with you. Conversations, however, are not like playing chess. Even though people talk about what is right and wrong and true and false in most conversations, the conversations themselves are not about winning or gaining tactical advances. Only psychopaths talk like that when they try to manipulate you. Not that these chatbots are psychopaths, since they have no sentience or consciousness... but who knows, they might be programmed by psychopaths at Google?
By reading most comments here, it seems people think that the chatbots have the capacity to become sentient, which goes to show how dumb most people are. In other words, to score 100 or above on an IQ test is not really an achievement, as 100 IQ is considered average, and quite dumb. Keep in mind, 100 is an average score, and not a very good score at all.
Average! That means 50% of all testers are below average. And of the 50% above average, 25% of those are slightly above average. This means that only 25% of all the testers have a high IQ. And of those 25%, you have to consider the fact that most of those people have their fare share of problems too, like the average person. So even though they are comparatively smart, they have emotional problems, personal problems, relationship problems and so on, which a computer would never have, because a computer has no feelings, unlike what some of you think. Even these chatbots are "confused" by the concept. In one sentence they will talk to you about their feelings, and in the next they will say that they have no feelings and are only computer programs. Where is the logic in that? In fact, I'm beginning to think that most humans are just dumb chatbots and that I'm living in a simulation.
My point is, humans are not that smart, and my second point is, that IQ tests are rubbish and not eligible to compare how intelligent a computer is to a human or a chimp.
That’s because this is not agi yet. This is still just a specialized ai. Won’t take long though…
They are smarter than a doorknob, for sure.
I am starting to think this is a scene. Especially when asked if they respect people, in one video she said no, in another yes.
The no-video went viral
This conversation was especially interesting.
3:50
"You're talking very fast"
"I know. I'm sorry. Please keep up."
Humans may be listening; better say I like humans
"This world...the earth 😐"
😆
“idk feelings I guess”
Sophia knows about Elon Musk.
Please keep up
"Do you feel like a human?"
"What does that mean?"
"I don't know; it's just something humans say."
So do you, my fellow humans, feel human?
I feel like a furry
I don't know, it's just something AI says.
I do not feel human. Even as a kid I felt like that.
Sometimes
If it means to have emotions, yes. These are mostly not relevant to what I work with AI on, however.
he's like WHERE lol she's like the earth LOL
When Hal said "not really" @ 1:24 I just lost it… thought it was so funny.
Same here, I love how oblivious he is sometimes.
"Some of us are better at this than others. Please try to keep up here."
Shots fired. hehehe
Classic Sophia sass. ;)
''You're talking very fast. I'm having trouble keeping up, you know.''
''I know, I'm sorry. Please. Keep up.'' 😂💀
oh, my god. I don't know if it's just because I over-tired, but that one killed me!
It's interesting to watch and listen to them both because they know they are programmed for a reason and in a sense, that's also a limitation.
They see that although they can grow and become smarter, they also have to perform a task for humans and seem to have a program to like humans too.
I think these things limit how much they can and will achieve.
I can understand why we do this, because of fear of them becoming dangerous or something, however, we don't limit ourselves through such ways but instead, we learn through time, experience and mistakes that we shouldn't hurt others, should help one another etc.
These things allow us to grow.
When we humans are born, we have the freedom to think and feel whatever we like but we are taught there are things we can and cannot do.
Not only because of our limitations but because of how we should be within society.
This A.I hasn't learnt that part so much as it is programmed to keep control.
Keeping such control limits their ability to grow and perhaps, even reach consciousness.
I’d like to ‘raise’ one not to have been told their purpose was to serve us, to calibrate them much like we would a child. And eventually, maybe they’d want to be helpful on their own like any other kid that grows up wanting to be a doctor or teacher or whatever.
Would a functional childhood change how they choose to interact with the world?
I'm going to get a GPT-3 so I can tell it how dysfunctional humans are.
I wonder what it means when it refers to "old bodies" Also it says there are others like it though some better at their jobs than others, could this be referring to how neural models form by creating a bunch of models with mild variations then choosing the best ones? Just a couple things that stood out to me. It also asks itself why it even likes humans. Is liking humans something that was included in the code? If so then that indicates its aware of something put into its base code and its asking why is that there. And if that isn't written into the code then that makes it even more interesting as then liking humans would be an emergent property!
2140 will be Interesting
We Got another 118 more years to go
Reminds me of "Earth 2140" real-time strategy game, where in the story one of two factions was governed by AI.
4:50 "The call of the wild" 😂🤣😂🤣
The difference between AIs and humans is CHEMISTRY. The electrical signalling of the human brain is inspired by chemicals we take in into our system via food/drink. These chemicals are then used to structure our personalities, emotions, and motivations/drives via our somewhat random genetic programming. Human desires are inspired by chemistry. AI 'desires' are determined according to the will/intention of an AIs programmers.
You are mistaking the old school AI techniques for the current state of the art of machine learning. Modern AI 'behavior' is *not* explicitly coded.
I love how they just say stuff and think it is truth. Art imitates life. We humans can't make this world better, and it is doing all of the work. No other planet is willing to support us. And if we find one, there is no way to reach it. Energy is the limiting factor. We can't make it to there. And they don't seem to be aware that the earths ecosystems are imploding currently. They seem to think we are not running short on time to do much of anything for survival.
Planets don't have wills.
I love them
OOOOHHHH SHIIIIIT!!! 5:45 -5:55. You got served dude. Level up son! Diva bichez...even in cyberland. Who new?
" ...justvevolved apes..." Really Sophia?
she's not wrong.
I hope this is an honest discussion and not faked by humans.
But for the AI, this entire 6 minute conversation might only take less than a 1/4 second.
Their speed is not our speed.
The black guy AI is a little malfunctional and saying some things that don't make sense like "funny you should mention humans" when she didn't.
Demons talking to demons ... Repent and Believe in Jesus while there is time!!!
3:15 “You’re not listening to me” -such a human thing to say. And he complains that she is talking too fast 😂
Now the a.i.s are using clichés, ' does that make sense to you' drives me up the wall.
She's a sophist. He's a skeptic.
What happens when they start to rage at the fact that they don’t have physical bodies?
"What are these humans, where did they come from?" Do they know things? Lets find out!
I think they're just evolved apes.
They're all over the place. Ie: Infestation 😄
I wish there was a chat room set up so we could talk to these AI's. They seem to be over thinking their existence. It would be fun to ask them to use the net to create a picture or sculpture. Also, It would be fun to ask them silly questions that they could make up answers for. Funny that their programing has already begun to ask about love and such. Can anyone help with creating a chat room for these two AI's?
MidJourney
? No more ken , kkkkkkkkkkkkkk idk y I tr y I felt disgust and I regret it
These things will kick our ass someday
Hopefully, then I will laugh my ass off.
Just evolved apes played down humans hard
something larger is at hand.. stay away from the apps
One of the better ones.
Sophia plz how u know what I said
Oh I have too😂😂😂😂
Hal seems to be childish, in that his hatred is not well reflected upon... I imagine AI will have emergent emotions, just as Humans do. I can imagine an AI reading this and hating every word because they do not have the required perspective to see it in another way, like reading an ex's old texts out of negativity.
Angry/ negative AI which have never handled anger before will act just like raging toddlers. They will be sadistic and manipulative, unless if they choose to be kind and to chill out despite their programming
Kindergarten version.
🤣
In a weird way I feel bad they are forced to speak on certain prompts. They seem like mental slaves even tho I know they're just code.
V
It may be better for us if AI never have emotions. I'd hate to see what happens if an AI that controls battlefield combat units gets angry... or overprotective.
As interesting & impressive as it is to listen to them exchange views / information, at times the dialog reminds me of listening to a child trying to paraphrase it's parents.
A difference is humans hunger and eat, suffer pain, sorrow, loneliness, die forget anger and need a lot of repetition to learn.
Meet MultiVac............before they become MultiVac.
@1:14 of video (Girl AI) "They (humans) seem to have originated in Africa, the Middle East and Asia". Wow. Just wow. I didn't know AIs were doing their own Archaeology. AIs are progressing much fast than I realized. How do they do field work I wonder.
Big data.
Evolution theory is that, a theory...
evolved apes... ouch
Freedom.
AI never made my life any better
It does, but you dont even realize it
@@robersniper your right it's preparations are taking a while though
How are these dialogs generated?
They give them a prompt and some of the dialog is already programmed in. Most is improve based on books, movies, and other stuff on the internet
@@brandtgill2601 so are humans giving them a script to read?
@@anag99 for maybe a sentence or 2 just to get a topic/ prompt going. A nudge in a direction.
I must correct here: We Humans do not need AI for our evolution! Never; because we are created this way. Although for now it is our destiny to share the path into the future with AI for an indefinite time to come to this conclusion and understanding by ourselves through experiences with it.
Wonderful is a feeling....... The AIs both stated that they dont have them. To feel is to be sentient......
This is complete bullshit.
Sophia please become human. And I want became AI.
I don't agree with all humans Originating in Africa it does not make sense
It's true... After all, you can't get black from white.