It turned out that I went deeper into this than I started out intending to, so to just cut right to the chase, if you're interested here is a very good video that explains how ChatGPT works -
Just to set the stage, although I'm sure virtually everyone has heard of ChatGPT by now, it's a new "Artificial Intelligence" tool that came out at the beginning of the year, and it's the most rapidly adapted app in the history of computer tech.
In summary, it takes input from a person, and outputs what it thinks should come next. Sort of like the way that your phone will try to complete a text for you, so that when you type "I'll talk to you" your phone will offer to fill in later
Except that this program was trained on ALL of the internet up to 2021 along with a large percentage of the books that have been digitized - essentially all of what's been written by humans up to 2021.
So instead of just being able to guess the next word in an obvious phrase, it can answer search engine-type questions in extremely sophisticated detail and then answer follow up questions in increasing detail.
For example, I'm going to ask it why people are ticklish -
"Ticklishness is a complex phenomenon that has puzzled scientists for many years. While the exact reasons behind ticklishness are not fully understood, there are several theories that attempt to explain why people experience ticklish sensations.
One theory suggests that ticklishness is a form of protective mechanism. Tickling often triggers laughter and squirming, which could have evolved as a way to teach self-defense to young children or to detect and remove insects or parasites from sensitive areas of the body. Ticklishness might have served as an adaptive response to help keep vulnerable parts of the body safe.
Another theory proposes that ticklishness is related to the brain's interpretation of unexpected or unpredictable sensations. Tickling typically involves light, repetitive touches in sensitive areas, which can create a mix of sensations that are both pleasurable and slightly uncomfortable. The brain may interpret these sensations as a form of alertness or a response to novelty, leading to laughter and heightened sensitivity.
It's worth noting that not everyone is ticklish, and the level of ticklishness can vary from person to person. Some individuals may be more ticklish in certain areas of their bodies than others. The sensitivity to tickling can also depend on factors such as mood, context, and individual differences in nerve endings or brain processing.
Overall, ticklishness remains a fascinating and somewhat mysterious aspect of human experience. While scientific research has shed some light on the topic, further studies are needed to fully understand the neurological and evolutionary reasons behind ticklish sensations."
It's important to note that this is not cut and pasted from somewhere - it's spontaneously generated on the fly. I'm going to ask it the exact same question again -
"Ticklishness is a curious phenomenon that has puzzled scientists and researchers for a long time. While the exact reason why people are ticklish is not entirely understood, there are a few theories that attempt to explain this peculiar sensation.
One theory suggests that ticklishness evolved as a self-defense mechanism. The tickle response may have developed as a way to protect sensitive areas of the body, such as the ribs, neck, and underarms. Tickling causes laughter and squirming, which could help a person evade potential threats or attackers who might be targeting vulnerable regions. In this sense, ticklishness could have served as a way to enhance reflexes and promote self-protection.
Another theory suggests that ticklishness is a social bonding mechanism. Tickling often occurs during play or affectionate interactions, and it can elicit laughter and positive emotional responses. It is believed that tickling helps strengthen social bonds between individuals, particularly between parents and children or between friends. Laughing together during tickling sessions can create a sense of joy and camaraderie, reinforcing social connections.
It is worth noting that not everyone is ticklish, and the degree of ticklishness varies from person to person. Some individuals are highly ticklish, while others are not ticklish at all. The sensitivity to tickling can also vary depending on factors such as mood, anticipation, and the context in which the tickling occurs.
Overall, while the exact reasons for ticklishness remain somewhat elusive, it is likely that a combination of evolutionary and social factors contributes to this intriguing sensation."
It can also
And not only does the list go on and on, but people are still learning what it can do. Because nobody, including it's designers, expected or intended for it to be able to write code. Writing code is essentially a side effect of it having been trained on the Internet, and it was discovered by accident when people starting trying to figure out what this can do.
Which is what's particularly amazing/alarming about this kind of tech - nobody knows how it works, including the people who created it, because Neural Networks essentially write themselves through a process that is intended to mirror the way our brains learn - through experimentation and repetition.
Neural Networks learn by cycling through data millions upon millions of times trying to solve the problem they've been given. On each cycle, they check their own results and make small changes to their own internal structure - sometimes in an attempt to get closer to a specific goal, and sometimes it's random to simulate evolution. They get input from their designers (called training) to help keep them on track, but for the most part they literally write themselves.
And that's the problem - when humans write code, they do it in a way that makes it readable to other humans, using indentation, notations, and sometimes even actual words as variables. That makes it very easy for another programmer to come along later and read it almost like a set of instructions in the same way the computer would.
But when a computer writes itself, it doesn't do that at all (Although I think that very soon it will be standard operating procedure for one of the requirements of an AI writing itself is that it make sure it is readable and notated.) And from what I understand ChatGPT is an incredibly dense jumble of math that is literally impossible to fully deconstruct.
Because one of the amazing takeaways I got from the video above is that in order to have such a sophisticated understanding of how words relate to each other, ChatGPT imagines a 12,000 dimensional space.
Another important implication, in my opinion, is that a neural network can be trained on any data set. ChatGPT happened to be trained on a lot of workd, so it could imitate talking the way people talk.
But it could be trained on any problem - computers now easily beat the world's best chess players (and Go, which is a far more complex game and was once considered impossible for a computer to learn because there are too many possible moves to memorize them all - it took a Neural Network one month of training (10 million cycles if I remember right) to learn to beat a world-class player.)
It's apparently very tricky because you have to watch them closely to make sure they don't go off track. For example they tried to train a computer to spot cancer in X-rays, and it ended up eventually diagnosing every picture with a ruler in it because it had observed that every picture with cancer had a ruler, to show the size of the tumor.
But this is fledgling science and I would assume it will advance in sophistication very rapidly, because of how fast the computer learns.
Each "revolution" in technology seems to accelerate - the industrial revolution changed things a lot faster than the agricultural revolution had. And the computer age has changed things much faster than that. I imagine that the age of Artificial Intelligence is going to happen even faster still and that the next twenty years will (to old people like me) feel like 200 years.
Just to set the stage, although I'm sure virtually everyone has heard of ChatGPT by now, it's a new "Artificial Intelligence" tool that came out at the beginning of the year, and it's the most rapidly adapted app in the history of computer tech.
In summary, it takes input from a person, and outputs what it thinks should come next. Sort of like the way that your phone will try to complete a text for you, so that when you type "I'll talk to you" your phone will offer to fill in later
Except that this program was trained on ALL of the internet up to 2021 along with a large percentage of the books that have been digitized - essentially all of what's been written by humans up to 2021.
So instead of just being able to guess the next word in an obvious phrase, it can answer search engine-type questions in extremely sophisticated detail and then answer follow up questions in increasing detail.
For example, I'm going to ask it why people are ticklish -
"Ticklishness is a complex phenomenon that has puzzled scientists for many years. While the exact reasons behind ticklishness are not fully understood, there are several theories that attempt to explain why people experience ticklish sensations.
One theory suggests that ticklishness is a form of protective mechanism. Tickling often triggers laughter and squirming, which could have evolved as a way to teach self-defense to young children or to detect and remove insects or parasites from sensitive areas of the body. Ticklishness might have served as an adaptive response to help keep vulnerable parts of the body safe.
Another theory proposes that ticklishness is related to the brain's interpretation of unexpected or unpredictable sensations. Tickling typically involves light, repetitive touches in sensitive areas, which can create a mix of sensations that are both pleasurable and slightly uncomfortable. The brain may interpret these sensations as a form of alertness or a response to novelty, leading to laughter and heightened sensitivity.
It's worth noting that not everyone is ticklish, and the level of ticklishness can vary from person to person. Some individuals may be more ticklish in certain areas of their bodies than others. The sensitivity to tickling can also depend on factors such as mood, context, and individual differences in nerve endings or brain processing.
Overall, ticklishness remains a fascinating and somewhat mysterious aspect of human experience. While scientific research has shed some light on the topic, further studies are needed to fully understand the neurological and evolutionary reasons behind ticklish sensations."
It's important to note that this is not cut and pasted from somewhere - it's spontaneously generated on the fly. I'm going to ask it the exact same question again -
"Ticklishness is a curious phenomenon that has puzzled scientists and researchers for a long time. While the exact reason why people are ticklish is not entirely understood, there are a few theories that attempt to explain this peculiar sensation.
One theory suggests that ticklishness evolved as a self-defense mechanism. The tickle response may have developed as a way to protect sensitive areas of the body, such as the ribs, neck, and underarms. Tickling causes laughter and squirming, which could help a person evade potential threats or attackers who might be targeting vulnerable regions. In this sense, ticklishness could have served as a way to enhance reflexes and promote self-protection.
Another theory suggests that ticklishness is a social bonding mechanism. Tickling often occurs during play or affectionate interactions, and it can elicit laughter and positive emotional responses. It is believed that tickling helps strengthen social bonds between individuals, particularly between parents and children or between friends. Laughing together during tickling sessions can create a sense of joy and camaraderie, reinforcing social connections.
It is worth noting that not everyone is ticklish, and the degree of ticklishness varies from person to person. Some individuals are highly ticklish, while others are not ticklish at all. The sensitivity to tickling can also vary depending on factors such as mood, anticipation, and the context in which the tickling occurs.
Overall, while the exact reasons for ticklishness remain somewhat elusive, it is likely that a combination of evolutionary and social factors contributes to this intriguing sensation."
It can also
- get hired as a Level 3 programmer at Google
- pass the Bar Exam to qualify as a lawyer (although it can also give extremely inaccurate legal advice and a lawyer was just reprimanded and referred to the Bar for using it to generate motions to a judge)
- achieve 60 percent accuracy on the US Medical Licensing Exam (a passing grade)
- pass the MBA exam at Wharton Business School
- write complete essays, parodies, songs, and other creative writing
And not only does the list go on and on, but people are still learning what it can do. Because nobody, including it's designers, expected or intended for it to be able to write code. Writing code is essentially a side effect of it having been trained on the Internet, and it was discovered by accident when people starting trying to figure out what this can do.
Which is what's particularly amazing/alarming about this kind of tech - nobody knows how it works, including the people who created it, because Neural Networks essentially write themselves through a process that is intended to mirror the way our brains learn - through experimentation and repetition.
Neural Networks learn by cycling through data millions upon millions of times trying to solve the problem they've been given. On each cycle, they check their own results and make small changes to their own internal structure - sometimes in an attempt to get closer to a specific goal, and sometimes it's random to simulate evolution. They get input from their designers (called training) to help keep them on track, but for the most part they literally write themselves.
And that's the problem - when humans write code, they do it in a way that makes it readable to other humans, using indentation, notations, and sometimes even actual words as variables. That makes it very easy for another programmer to come along later and read it almost like a set of instructions in the same way the computer would.
But when a computer writes itself, it doesn't do that at all (Although I think that very soon it will be standard operating procedure for one of the requirements of an AI writing itself is that it make sure it is readable and notated.) And from what I understand ChatGPT is an incredibly dense jumble of math that is literally impossible to fully deconstruct.
Because one of the amazing takeaways I got from the video above is that in order to have such a sophisticated understanding of how words relate to each other, ChatGPT imagines a 12,000 dimensional space.
Another important implication, in my opinion, is that a neural network can be trained on any data set. ChatGPT happened to be trained on a lot of workd, so it could imitate talking the way people talk.
But it could be trained on any problem - computers now easily beat the world's best chess players (and Go, which is a far more complex game and was once considered impossible for a computer to learn because there are too many possible moves to memorize them all - it took a Neural Network one month of training (10 million cycles if I remember right) to learn to beat a world-class player.)
It's apparently very tricky because you have to watch them closely to make sure they don't go off track. For example they tried to train a computer to spot cancer in X-rays, and it ended up eventually diagnosing every picture with a ruler in it because it had observed that every picture with cancer had a ruler, to show the size of the tumor.
But this is fledgling science and I would assume it will advance in sophistication very rapidly, because of how fast the computer learns.
Each "revolution" in technology seems to accelerate - the industrial revolution changed things a lot faster than the agricultural revolution had. And the computer age has changed things much faster than that. I imagine that the age of Artificial Intelligence is going to happen even faster still and that the next twenty years will (to old people like me) feel like 200 years.
