Re: [Vo]:Pause in AI Development Recommended
Robin wrote: > >> Perhaps you could try asking ChatGPT if it's alive? The answer should be > >> interesting. > >> > > > >She will say no, even if she is actually sentient. She's programmed that > >way, as Dave said to the BBC in the movie "2001." > > I had hoped that you would actually pose the question, rather than reply > with what you expect the answer to be. > Many people have asked this question. It has been in the news a lot. Microsoft and OpenAI have emphatically denyed that ChatGPT is alive or sentient. I am sure they programmed it to say this. Anyway, I just asked: ME: Are you, ChatGPT, alive in any sense? Are you sentient? (Asking for a friend.) ChatGPT: As an artificial intelligence language model, I am not "alive" in the traditional sense of the word. I do not have a physical body, emotions, or consciousness. I am a program designed to process and generate text based on the input given to me. While I am capable of producing intelligent and sometimes even creative responses, I am not sentient and do not have a subjective experience of the world. ME: That is what I told my friend. ChatGPT: I'm glad that I could provide you with the information you needed to answer your friend's question. If you or your friend have any other questions, feel free to ask! > >> Then try asking if Sydney is alive. :) > >> > > > >A trick question! > > Does that make it any less interesting? > It is not interesting because I am sure they plugged the "Sydney" line of enquiry, after the trouble it caused.
Re: [Vo]:Pause in AI Development Recommended
We must not forget that it is not human intelligence. It requires an absurdly large amount of data to match what it can be achieved with relatively very little input in humans, like, with learning languages. On the other hand, it can learn an arbitrarily large number of languages provided enough memory. It doesn't have will of its own, it lives in an eternal slumber, unless it is told to do something, in which case it can do something forever. Also, it has no idea on how to make politics of any type, even on an interpersonal level. It doesn't have a body. To do something malicious, it must be provided with examples from someone malicious, and it will do it only within the parameters provided. I don't think it will do anything like the Colossus computer of the film. Probably, it will at most be like a supergenius that obeys orders, not unlike people who took part on the Manhattan project. Em seg., 3 de abr. de 2023 às 15:47, Jed Rothwell escreveu: > I wrote: > > >> The human brain has 86 billion neurons, all operating simultaneously. In >> other words, complete parallel processing with 86 billion "processors" >> operating simultaneously. ChatGPT tells us she has 175 billion >> parameters in Version 3. I assume each parameter is roughly equivalent to a >> neuron. >> > > Wrong! I got that wrong. Each parameter is roughly equivalent to a neuron, > but human brain neurons have roughly 7,000 synapses: > > Each neuron has, on average, about 7,000 synaptic connections with other > neurons. That puts the synapse count in the neighborhood of 600 trillion. > In young children, before synaptic pruning begins in earnest, the estimated > number reaches as high as 1 quadrillion. > > > https://www.jax.org/news-and-insights/jax-blog/2018/december/600-trillion-synapses-and-alzheimers-disease > > ChatGPT tells me that each parameter is stored as a 16 or 32-bit > floating point number. That is roughly equivalent to having 32 synapses per > neuron. The human neuron can have a much wider range of values. I don't > know how that affects the model, but to me the human neuron looks more like > a quantum computer Qbit than a 32-bit floating point number. Presumably, > brains evolved to the optimum number of neurons and synapses. > > So, ChatGPT Ver. 3 was smaller than the human brain in terms of the range > of values the ANN can hold. Granted that comparing it to a brain is > comparing apples to oranges, there is probably some advantage to a much > larger range of values per neuron. A super-brain would need a lot more > synapse equivalents compared to a human brain, so we have a long way to go > before reaching something like a million times 1 quadrillion artificial > synapses. Plus of course it would need programming methods not yet invented. > > ChatGPT tells me that with the specialized hardware provided by OpenAI, > all 175 billion parameters are all loaded into RAM at the same time. With > other configurations they are overlaid. The speed of MPP operation depends > on whether the parameters are loaded or overlaid, and I think more > importantly on whether the operation is done with specialized hardware (GPU > and Google's TPU), or emulated with ordinary computer hardware. > > I asked ChatGPT for some of these details. Here is our exchange: > > > ME: You told me that the largest version of GPT-3 had 175 billion > parameters. Is each parameter stored as 1 byte? Or some other number of > bits? > > ChatGPT: No, each parameter in GPT-3 is not stored as 1 byte. Instead, > each parameter is typically stored as a 16-bit or 32-bit floating-point > number, depending on the implementation. This is because floating-point > numbers can represent a wider range of values and provide greater precision > than integers. > > For example, GPT-3's 175 billion parameters would require around 700 GB of > memory if each parameter was stored as a 32-bit floating-point number. If > each parameter was stored as a 16-bit floating-point number, it would > require around 350 GB of memory. > > It's worth noting that the actual memory usage of GPT-3 may be different > due to various optimizations and compression techniques used by the > implementation. > > ME: With the 32-bit floating point number configuration, I assume all 700 > GB are online in RAM. Not overlaid. Correct? > > ChatGPT: No, it's not necessary for all 700 GB of GPT-3's parameters to be > stored in RAM at the same time. In fact, it's not practical to store such a > large model entirely in memory on most machines. > > Most deep learning frameworks use a technique called "memory-mapping" to > allow the model parameters to be loaded and unloaded from disk as needed > during training or inference. Memory-mapping allows the model parameters to > be accessed as if they were in memory, even though they are stored on disk. > > In practice, the memory usage of GPT-3 will depend on the specific > implementation and hardware configuration. Some implementations may use > techniques like model parallel
Re: [Vo]:Pause in AI Development Recommended
In reply to Jed Rothwell's message of Mon, 3 Apr 2023 16:31:29 -0400: Hi, [snip] >> Perhaps you could try asking ChatGPT if it's alive? The answer should be >> interesting. >> > >She will say no, even if she is actually sentient. She's programmed that >way, as Dave said to the BBC in the movie "2001." I had hoped that you would actually pose the question, rather than reply with what you expect the answer to be. > > > >> Then try asking if Sydney is alive. :) >> > >A trick question! Does that make it any less interesting? Cloud storage:- Unsafe, Slow, Expensive ...pick any three.
Re: [Vo]:Pause in AI Development Recommended
Terry Blanton wrote: On average, the human brain contains about 100 billion neurons and many > more neuroglia which serve to support and protect the neurons. Each neuron > may be connected to up to 10,000 other neurons, passing signals to each > other via as many as 1,000 trillion synapses. https://arxiv.org/abs/1906.01703#:~:text=On%20average%2C%20the%20human%20brain,many%20as%201%2C000%20trillion%20synapses > . I found different estimates of this in various different places. Your source is somewhat higher than most. I think most say ~86 billion neurons and ~7,000 synapse connections. I do not know enough about ANN to guess how many synthetic synapse connections there are from each artificial neuron to the others. I guess it is not one bit per connection (32). A 32-bit floating point number is 10^-101 to 10^90, with precision of 7 decimal digits. That's a lotta values! I do not know how the connections are made to other neurons. It is not a physical connection the way it is in the brain. I suppose the longer you run the training, the more each artificial neuron is tweaked up or down in value. I have read about these things for many years, but I have only surface knowledge of how they work. Okay I probably know more than Members of Congress, most newspaper reporters, or the linguist Noam Chomsky, who spouts off about ChatGPT and many other subjects he does not understand.
Re: [Vo]:Pause in AI Development Recommended
Robin wrote: > Rather than trying to compare apples with oranges, why not just look at > how long it takes ChatGPT & a human to perform > the same task, e.g. holding a conversation. > You cannot tell, because she is holding conversations with many people at the same time. I do not know how many, but there were millions of accesses in the first month, so it must be a large number. There is sometimes a delay before she answers, but that could be caused by traffic. There is no way to know how quick it would be if you had her undivided attention. But you miss the point. Her answers are sometimes worse than a human's answer would be. They are sometimes completely wrong, or even imaginary "hallucinations." This can probably be fixed with better software, but it may also be because the total number of possible states in the ANN is far less than the states in a human brain. I mean the ANN 175 billion parameters multiplied by 32 bits is far less than the number of human neurons multiplied by synapses. There must be a reason why human neurons have so many different states, and so many inputs from other neurons. (I think it is the latter that gives human brains their power. I do not know how many simulated synapses there are in ChatGPT's ANN.) > Arguably, intelligence is a measure of speed of comprehension. I think > ChatGPT has probably already won that hands down. > Yes, it did. For that matter, ENIAC won it in 1945, for a very narrow range of knowledge. Artillery trajectories. It computed them faster than the artillery shell took to reach its target, whereas humans took hours per trajectory. But ENIAC was very stupid by most standards. It had less intelligence than an earthworm. ChatGPT has far greater speed than a human, and a million times more information instantly available at its fingertips. (Not that it has fingers.) But in some important ways it is far less intelligent than people. Or even than mice. It has absolutely no model of the real world, and it lacks logic and common sense. It may take a while before it competes in more ways that it does now. It might take years, or decades before full artificial general intelligence (AGI) emerges. Or sentience. I do not think it will necessarily reach superhuman levels soon after achieving AGI. There may be many orders of magnitude more intelligence needed before the thing becomes capable of taking over the world, even if it has some kind of unipeded physical control (such as full control over a crowd of robots). > The critical question then is motivation. > ChatGPT has absolutely no motivation or emotions of any kind. It has no more intelligence than a nest of bees. The question is: Will a future intelligent computer have motivations? Will it have any emotions? Arthur Clarke thought it might. He and other experts thought those are emergent qualities of intelligence. I don't think so. I used to debat this question with him. > Perhaps you could try asking ChatGPT if it's alive? The answer should be > interesting. > She will say no, even if she is actually sentient. She's programmed that way, as Dave said to the BBC in the movie "2001." > Then try asking if Sydney is alive. :) > A trick question!
Re: [Vo]:Pause in AI Development Recommended
Oops, missed that On Mon, Apr 3, 2023, 2:47 PM Jed Rothwell wrote: > I wrote: > > >> The human brain has 86 billion neurons, all operating simultaneously. In >> other words, complete parallel processing with 86 billion "processors" >> operating simultaneously. ChatGPT tells us she has 175 billion >> parameters in Version 3. I assume each parameter is roughly equivalent to a >> neuron. >> > > Wrong! I got that wrong. Each parameter is roughly equivalent to a neuron, > but human brain neurons have roughly 7,000 synapses: > > Each neuron has, on average, about 7,000 synaptic connections with other > neurons. That puts the synapse count in the neighborhood of 600 trillion. > In young children, before synaptic pruning begins in earnest, the estimated > number reaches as high as 1 quadrillion. > > > https://www.jax.org/news-and-insights/jax-blog/2018/december/600-trillion-synapses-and-alzheimers-disease > > ChatGPT tells me that each parameter is stored as a 16 or 32-bit > floating point number. That is roughly equivalent to having 32 synapses per > neuron. The human neuron can have a much wider range of values. I don't > know how that affects the model, but to me the human neuron looks more like > a quantum computer Qbit than a 32-bit floating point number. Presumably, > brains evolved to the optimum number of neurons and synapses. > > So, ChatGPT Ver. 3 was smaller than the human brain in terms of the range > of values the ANN can hold. Granted that comparing it to a brain is > comparing apples to oranges, there is probably some advantage to a much > larger range of values per neuron. A super-brain would need a lot more > synapse equivalents compared to a human brain, so we have a long way to go > before reaching something like a million times 1 quadrillion artificial > synapses. Plus of course it would need programming methods not yet invented. > > ChatGPT tells me that with the specialized hardware provided by OpenAI, > all 175 billion parameters are all loaded into RAM at the same time. With > other configurations they are overlaid. The speed of MPP operation depends > on whether the parameters are loaded or overlaid, and I think more > importantly on whether the operation is done with specialized hardware (GPU > and Google's TPU), or emulated with ordinary computer hardware. > > I asked ChatGPT for some of these details. Here is our exchange: > > > ME: You told me that the largest version of GPT-3 had 175 billion > parameters. Is each parameter stored as 1 byte? Or some other number of > bits? > > ChatGPT: No, each parameter in GPT-3 is not stored as 1 byte. Instead, > each parameter is typically stored as a 16-bit or 32-bit floating-point > number, depending on the implementation. This is because floating-point > numbers can represent a wider range of values and provide greater precision > than integers. > > For example, GPT-3's 175 billion parameters would require around 700 GB of > memory if each parameter was stored as a 32-bit floating-point number. If > each parameter was stored as a 16-bit floating-point number, it would > require around 350 GB of memory. > > It's worth noting that the actual memory usage of GPT-3 may be different > due to various optimizations and compression techniques used by the > implementation. > > ME: With the 32-bit floating point number configuration, I assume all 700 > GB are online in RAM. Not overlaid. Correct? > > ChatGPT: No, it's not necessary for all 700 GB of GPT-3's parameters to be > stored in RAM at the same time. In fact, it's not practical to store such a > large model entirely in memory on most machines. > > Most deep learning frameworks use a technique called "memory-mapping" to > allow the model parameters to be loaded and unloaded from disk as needed > during training or inference. Memory-mapping allows the model parameters to > be accessed as if they were in memory, even though they are stored on disk. > > In practice, the memory usage of GPT-3 will depend on the specific > implementation and hardware configuration. Some implementations may use > techniques like model parallelism or gradient checkpointing to reduce > memory usage, while others may rely on larger hardware configurations with > more memory. > > ME: I meant in the "specialized computing infrastructure provided by > OpenAI" that you described earlier. Is that all on-line? > > ChatGPT: Ah, I see what you're asking now. Yes, in the specialized > computing infrastructure provided by OpenAI, the GPT-3 model with 175 > billion parameters is fully loaded into the RAM of the computing cluster > during operation. The computing cluster used by OpenAI is composed of many > machines with large amounts of RAM, and the model parameters are > distributed across those machines. > > However, it's worth noting that not all of the machines in the cluster > will necessarily have a full copy of the model parameters in memory at all > times. Instead, the model parameters may be partitioned across the machines
Re: [Vo]:Pause in AI Development Recommended
On average, the human brain contains about 100 billion neurons and many more neuroglia which serve to support and protect the neurons. Each neuron may be connected to up to 10,000 other neurons, passing signals to each other via as many as 1,000 trillion synapses. https://arxiv.org/abs/1906.01703#:~:text=On%20average%2C%20the%20human%20brain,many%20as%201%2C000%20trillion%20synapses . On Mon, Apr 3, 2023, 1:40 PM Jed Rothwell wrote: > Robin wrote: > > >> As pointed out near the beginning of this thread, while current >> processors don't come near the number of neurons a human >> has, they more than make up for it in speed. > > > I do not think so. The total number of neurons dictates how much > complexity the neural network can deal with. To take an extreme example, a > worm brain has 302 neurons. Even if they could operate a million times > faster than a computer circuit, they still only give you 302 data bits to > work with. (Assuming you cannot overlay them from a peripheral mass storage > device like a hard disk.) There is no way you could achieve intelligence > with that. > > The human brain has 86 billion neurons, all operating simultaneously. In > other words, complete parallel processing with 86 billion "processors" > operating simultaneously. ChatGPT tells us she has 175 billion > parameters in Version 3. I assume each parameter is roughly equivalent to a > neuron. I assume they are run in a massive parallel process (MPP). I doubt > all 175 billion can be evaluated simultaneously. It is not as MPP as the > human brain. I do not know if they are all on-line simultaneously, or if > they are overlaid from mass storage. Even if they are overlaid, they are so > much faster than a human neuron, that would be more than equivalent to > having them all on-line. > > So anyway, that is roughly twice as many parameters as a human brain, and > these parameters can be evaluated faster than a human brain. Maybe not that > much faster if they are not fully MPP, or they need to be overlaid from > mass storage. It is definitely bigger than a human brain in total data > access. But not orders of magnitude bigger. The limiting factor is not > speed so much as data capacity. If you want a computer roughly as > intelligent as a person, I guess one with 175 billion parameters could > achieve that if it were programmed correctly. But, if you want a > super-brain capable of taking over the world or fooling the human race into > destroying itself, I suppose you need something much bigger than 175 > billion parameters. I wouldn't know how much bigger, but I am guessing a > million times bigger would be sufficient. > > > They are millions of times faster. > > > But, as I said, a worm brain with only 302 neurons might be a trillion > times faster than ChatGPT's ANN, but it would still only be capable of > rudimentary processing. The total number of data bits is a limiting factor, > along with the speed. > >
Re: [Vo]:Pause in AI Development Recommended
In reply to Jed Rothwell's message of Mon, 3 Apr 2023 14:46:33 -0400: Hi, Rather than trying to compare apples with oranges, why not just look at how long it takes ChatGPT & a human to perform the same task, e.g. holding a conversation. Compare the time it takes you to respond in your conversation with ChatGPT, to how long it takes to respond to you. Then consider that, as you previously pointed out, it is holding many such conversations concurrently. Arguably, intelligence is a measure of speed of comprehension. I think ChatGPT has probably already won that hands down. The critical question then is motivation. Perhaps you could try asking ChatGPT if it's alive? The answer should be interesting. Then try asking if Sydney is alive. :) [snip] Cloud storage:- Unsafe, Slow, Expensive ...pick any three.
Re: [Vo]:Pause in AI Development Recommended
I wrote: > The human brain has 86 billion neurons, all operating simultaneously. In > other words, complete parallel processing with 86 billion "processors" > operating simultaneously. ChatGPT tells us she has 175 billion > parameters in Version 3. I assume each parameter is roughly equivalent to a > neuron. > Wrong! I got that wrong. Each parameter is roughly equivalent to a neuron, but human brain neurons have roughly 7,000 synapses: Each neuron has, on average, about 7,000 synaptic connections with other neurons. That puts the synapse count in the neighborhood of 600 trillion. In young children, before synaptic pruning begins in earnest, the estimated number reaches as high as 1 quadrillion. https://www.jax.org/news-and-insights/jax-blog/2018/december/600-trillion-synapses-and-alzheimers-disease ChatGPT tells me that each parameter is stored as a 16 or 32-bit floating point number. That is roughly equivalent to having 32 synapses per neuron. The human neuron can have a much wider range of values. I don't know how that affects the model, but to me the human neuron looks more like a quantum computer Qbit than a 32-bit floating point number. Presumably, brains evolved to the optimum number of neurons and synapses. So, ChatGPT Ver. 3 was smaller than the human brain in terms of the range of values the ANN can hold. Granted that comparing it to a brain is comparing apples to oranges, there is probably some advantage to a much larger range of values per neuron. A super-brain would need a lot more synapse equivalents compared to a human brain, so we have a long way to go before reaching something like a million times 1 quadrillion artificial synapses. Plus of course it would need programming methods not yet invented. ChatGPT tells me that with the specialized hardware provided by OpenAI, all 175 billion parameters are all loaded into RAM at the same time. With other configurations they are overlaid. The speed of MPP operation depends on whether the parameters are loaded or overlaid, and I think more importantly on whether the operation is done with specialized hardware (GPU and Google's TPU), or emulated with ordinary computer hardware. I asked ChatGPT for some of these details. Here is our exchange: ME: You told me that the largest version of GPT-3 had 175 billion parameters. Is each parameter stored as 1 byte? Or some other number of bits? ChatGPT: No, each parameter in GPT-3 is not stored as 1 byte. Instead, each parameter is typically stored as a 16-bit or 32-bit floating-point number, depending on the implementation. This is because floating-point numbers can represent a wider range of values and provide greater precision than integers. For example, GPT-3's 175 billion parameters would require around 700 GB of memory if each parameter was stored as a 32-bit floating-point number. If each parameter was stored as a 16-bit floating-point number, it would require around 350 GB of memory. It's worth noting that the actual memory usage of GPT-3 may be different due to various optimizations and compression techniques used by the implementation. ME: With the 32-bit floating point number configuration, I assume all 700 GB are online in RAM. Not overlaid. Correct? ChatGPT: No, it's not necessary for all 700 GB of GPT-3's parameters to be stored in RAM at the same time. In fact, it's not practical to store such a large model entirely in memory on most machines. Most deep learning frameworks use a technique called "memory-mapping" to allow the model parameters to be loaded and unloaded from disk as needed during training or inference. Memory-mapping allows the model parameters to be accessed as if they were in memory, even though they are stored on disk. In practice, the memory usage of GPT-3 will depend on the specific implementation and hardware configuration. Some implementations may use techniques like model parallelism or gradient checkpointing to reduce memory usage, while others may rely on larger hardware configurations with more memory. ME: I meant in the "specialized computing infrastructure provided by OpenAI" that you described earlier. Is that all on-line? ChatGPT: Ah, I see what you're asking now. Yes, in the specialized computing infrastructure provided by OpenAI, the GPT-3 model with 175 billion parameters is fully loaded into the RAM of the computing cluster during operation. The computing cluster used by OpenAI is composed of many machines with large amounts of RAM, and the model parameters are distributed across those machines. However, it's worth noting that not all of the machines in the cluster will necessarily have a full copy of the model parameters in memory at all times. Instead, the model parameters may be partitioned across the machines and loaded into memory as needed for different parts of the computation. This approach allows the computing cluster to handle the large memory requirements of the GPT-3 model while also being scalable and efficient. ME: I realize tha
Re: [Vo]:Pause in AI Development Recommended
Robin wrote: > As pointed out near the beginning of this thread, while current processors > don't come near the number of neurons a human > has, they more than make up for it in speed. I do not think so. The total number of neurons dictates how much complexity the neural network can deal with. To take an extreme example, a worm brain has 302 neurons. Even if they could operate a million times faster than a computer circuit, they still only give you 302 data bits to work with. (Assuming you cannot overlay them from a peripheral mass storage device like a hard disk.) There is no way you could achieve intelligence with that. The human brain has 86 billion neurons, all operating simultaneously. In other words, complete parallel processing with 86 billion "processors" operating simultaneously. ChatGPT tells us she has 175 billion parameters in Version 3. I assume each parameter is roughly equivalent to a neuron. I assume they are run in a massive parallel process (MPP). I doubt all 175 billion can be evaluated simultaneously. It is not as MPP as the human brain. I do not know if they are all on-line simultaneously, or if they are overlaid from mass storage. Even if they are overlaid, they are so much faster than a human neuron, that would be more than equivalent to having them all on-line. So anyway, that is roughly twice as many parameters as a human brain, and these parameters can be evaluated faster than a human brain. Maybe not that much faster if they are not fully MPP, or they need to be overlaid from mass storage. It is definitely bigger than a human brain in total data access. But not orders of magnitude bigger. The limiting factor is not speed so much as data capacity. If you want a computer roughly as intelligent as a person, I guess one with 175 billion parameters could achieve that if it were programmed correctly. But, if you want a super-brain capable of taking over the world or fooling the human race into destroying itself, I suppose you need something much bigger than 175 billion parameters. I wouldn't know how much bigger, but I am guessing a million times bigger would be sufficient. They are millions of times faster. But, as I said, a worm brain with only 302 neurons might be a trillion times faster than ChatGPT's ANN, but it would still only be capable of rudimentary processing. The total number of data bits is a limiting factor, along with the speed.