they actually have not. you may be thinking of markov chains (a statistical model that makes the next word dependent on the prior chain of words) and mcculloch pitts artificial neural networks of the type described in this seminal paper home.csulb.edu/~cwallis/382/re… or perhaps the perceptron, a simple classifier circuit used in a famous demo
unfortunately none of these embodies anything like intellect because statistical likelihoods (no matter how many non-linearities or derivatives you introduce) do not encode the causality of the systems they model, nor do they encode semantics (meaningful symbolic relations), which are essentially an orthogonal notion to syntax (mere structures). a pattern does not encode the maker of the pattern, because the relationship between the two is not injective, meaning that the distinctness of the thing causing the pattern is not preserved by the pattern.
hence, because no such model can produce any information which encodes more than the pattern presented to it, it cannot produce any semantic contents. furthermore, when we use a model that determines the "most likely" words, phrases, or syntactic information to be in a certain place, it can be immediately reasoned that what is returned cannot correspond deliberately with the truth even syntactically, because many things are widely expressed which are not true, and the capacity to apply semantic context is necessary to make literate distinctions between jokes, earnest replies, metaphor, literal accounts of fact, and generally to determine meaning is completely missing.
with all of this in consideration, because people are profoundly susceptible to the assumption that there must be a mortal maker behind any pattern (this is called the ELIZA effect in machine learning, after a psychobabble machine that bamboozled even apparently sophisticated operators until it was revealed how trivial it was under the hood), the use of such a thing can only be a liability.
if you wanted a good alternative, that is, a machine that can actually make truth-preserving deductions about the information it is given, these have existed for much longer than LLMs and "deep learning" (a buzzword commonly given to the artificial neural network family of models, which do not learn but are repeatedly adjusted without anything resembling a closed loop without human intervention), and nearly as long as mcculloch-pitts networks and markov chains. they are called theorem provers and assisted/automated reasoning systems, and what they do is expand a set of mathematical and logical constraints as directed by a user to produce either nothing or a statement that is true if and only if the premises are true.
the reason they fell out of favor is because they work well and have real requirements beyond looking convincing, and therefore require acumen, engineering skills, people skills, and trust to build and maintain, whereas the industry wanted a magic 8-ball that would make a shiny demo for anyone gullible enough and advance their secondary objective of sowing mass hysteria and social and environmental crises so they could double dip by offering to solve those problems technocratically as well.
the reality of it is that because the generative statistical models can only imitate patterns they can only produce plausible looking non-sense, being "correct" periodically in the same sense that a broken clock is right twice a day. this is the aspect that is dangerous as an addiction because it is easy to lie to yourself if something looks authoritative to your lay eyes. in the literature, this phenomenon is called "cognitive surrender". thealgorithmicbridge.com/p/a-n…
I was, unfortunately, there without taking the opportunity to slime anyone. in minecraft. sorry. if it's any comfort, I think this is the timeline where the billionaires lose credibility and get liquidated slightly faster because it's all so much worse and more galling. imagine if they'd been highly effective and competent instead.
That's what scientists who once had a career in or adjacent to artificial intelligence who haven't lost their goddamn marbles say. If this weren't enough, it is worth noting that the deliberate manipulation of public opinion after the sudden introduction of LLMs as an alternative to everything that at least worked in a known limited capacity for certain tasks is being done actively and maliciously as part of a network of fascist pseudointellectuals known as TESCREAL, a term proposed by Timnit Gebru firstmonday.org/ojs/index.php/… to connect all of the various names and forums in which they have operated.
It's one of the times of all time to be alive right now.
Yann LeCun is one of the "godfathers of AI." He must be wicked smart, because he won the Turing Prize in 2018 (together with the other two "godfathers," Yoshua Bengio and Geoffrey Hinton ). The prize...
The issue I see here is that most of the general public have a misinformed idea of what Artificial Intelligence (AI) is. Most think of AI as LLMs and even though they are a significant part of AI today they did not evolve in AI until 2018, at least not to the general public. Yet AI has been a field of study and implementation since 1956. Take a look at the history of AI. Obviously it is not a new idea. I cannot tell you what all scientists think of AI. I can tell you what this scientist thinks of AI.
AI is a practical tool that can be used to make your research a lot more productive. AI can parse data sets far faster than humans can. With that said keep in mind that Natural Intelligence (NI) is not perfect and AI is modeled after NI, based on an Artificial Neural Network that can be trained (this is a generalization not a precise definition). Just like our own NI it can be taught wrong, it can produce bad results, just like us. It can even produce mistakes faster than any team of engineers could! Yet most of the time it produces useful results. So it is a vastly useful tool and one of the greatest achievements of our era.
I do not use general purpose corporate offerings for AI. I use domain specific LLMs that I train locally (fine tune not base train) and train ML architectures such as random forests, small CNNs. They have a far better accuracy rate as they are not trying to evaluate every possible evaluation as they are limited in scope by the domain they are trained in. I see this as the optimal way for a data science research. Also it is great for code analysis though I personally do not have AI generate code. I do not have an issue with those that do. In my opinion any one that has a code base of over 100,000 lines of code and is not using AI analysis is being irresponsible.
The anti-AI collective aka the general public have a very misinformed view of what AI is. Machine Learning (ML) is used to produce an LLM but an LLM is not the extent of what can be done with ML. AI is not chatbots on some corporations unregulated Data Center trained by nefarious methods. Contrary to popular belief. No, it does not take a data center to train a LLM, base training requires a lot of compute power but fine tuning it can be done at home with a rather modest data science rig, using no more power than a gaming computer. LLMs are not typically built on others Intellectual Property this is an attribute to how unregulated corporation train their LLMs not most data scientists. The whole topic is very skewed. What those AI Data Centers are for is to provide a service, to monetize AI, for the general public and corporations. They are not necessary to train AI nor to implement it. And training smaller traditional ML architectures can be done exclusively on local science data rigs.
I am against unregulated corporation doing whatever they please. I am against the way AI Data Centers are being developed with no concern for the environment. I am a humanitarian I am against pretty much anything that is a negative impact on humanity. I am not against AI or ML or LLMs. They are just tools that can be very useful to a data scientist or even the layperson providing they know something about what they are researching in order to vet the information they get.
It seems like everyone has forgotten all the bad advice given by obscure websites and the peer ask and answer sites that excel in being rude but often produce unreliable results even more so than AI. Because people associate LLMs with the behavior of unregulated corporations they do not like LLMs despite the fact that they have no idea what an LLM is. They just know they hate how corporation are monetizing them. I guess it has not occurred to them to address the actual problem and to push for regulations on the corporation in their demographic? Why address a problem by looking at the root of that problem when we can just use derogatory slurs about something we have no idea what it even is? Yes, I was being sarcastic as well as passive aggressive, my apologies.
We have real concerns with how unregulated corporations are destroying the environment. That has to be addressed. Yet, AI would still be in heavy use even if there were no AI data centers at all. The difference is that it would only be available to people like me with the knowledge of how to build a data science rig and implement the technology stack. The general public would be left out. So you decide, do you want to be left out? Great that is easy just do not use AI. If you do enjoy the benefits of this tool then do the ethical thing and push for regulations on corporations in their development of AI data centers. Enough said.
Same here ... Along with the fear the "AI" tech boys have instilled into the older population... I have said to many older people that are worried "as long as we can cut the power, everything should be fine"
I'm not against AI, I am against these tech bros that think they can do what they want, put larger parts of the population in fear and that they have no proof of concept
Yes, I tried to cover all of that in my comment but it is just a lot to go over. The facts still remain that the real issue is with how corporations are unregulated not AI itself. Though there is no disconnect for the anti-AI collective. They cannot distinguish between the fire that burns down the apartment building or the campfire. The blame the corporations unethical use of the technology on the technology. This, to me, is just unreasonable.
You drive an EV, or are you destroying our atmosphere by riding public transportation (that is not EV) or using a vehicle with an internal combustion engine? You did not see that building burn and you still endorse the use of fire. Come now, who is the hypocrite? Who exactly is uninformed?
Ai is not ruining anything. Unregulated corporations are ruining things. Did you see that apartment burn down, did you see how many people died? You still endorse the use of fire. Yes, this is exactly how you sound, but I know you are not going to get it. It is okay, you have the right to your opinions, have a great life!
That you actually believe that AI Data Centers (not to mention it does not take a data center to train a LLM a small server farm is used by many independent development firms that have no where near the impact of a Data Center) do more harm to the planet than all those internal combustion engines out there demonstrates just how little you do know about AI. You have the right to your opinion, take care and have a great life.
I am not squirming at anything. I am just shaking my head in frustration at your complete lack of knowledge of what you have such a strong opinion on. Great day to you mate. And because the quality of my day does not require your approval or sentiment, I will have one anyway. I hope you do as well. I do not let disagreements disenfranchise me from my ethics.
An alternative view has not hardened my own. Your lack of knowledge and such strong held opinions with no foundation has just demonstrated that I was not incorrect. At least not in your case. Your off the wall comments that have nothing to do with the discussion demonstrated that you are just another anti-AI in the crowd with nothing significant to bring to the table. You keep ranting, I am done with this discourse as it is going nowhere. We are right here where we started at. You using ad hominem attacks while questioning my philosophy. Take care and have a great day! And here is a little something for you to take a look at.
Are you aware that training an LLM is roughly comparable to electrical usage of one poultry processing plant such as Perdue or Tyson? Do you know how many of these plants there are?
I am glad you finally brought some evidence to the table for discussion. In regards to how the unregulated corporations need to be addressed, did I not already mention that in my comment? Did I not make it clear that I was against the nefarious methods they use for their data acquisition. As a developer I am all to aware of AI washing code. That does concern me, but I do not blame the technology. I blame the unregulated corporations. Why am I wrong to put the blame at the root of the problem? AI whether it be ML CCNs, random forest or LLMs are not responsible for these issues. The fact that these corporations are not being regulated and directly breaking the law is the issue. I thought I covered that too, did I not? I agree with a lot of the frustration that many anti-AI collective do. I thought I clearly stated that in my comment. If your only argument is that corporations are doing bad things, yes we are in complete agreement. That does not have any correlation to what AI is or what it can do. You cannot see that? I do not know how else to explain it. Keep in mind that any company that uses Refrigerated/ Freezer warehouses such as Tyson or Perdue uses just as much energy as Training LLMs and I am being generous because I am referring to frontier LLM not a basic usage LLM that most of us use locally that would take significantly less energy. So what is our disagreement? You really believe that that vast majority of those on this anti-AI mission are well informed and really understand what AI is? You think that my statement that they do not know what AI is, is somehow unfair? If you asked 100 people what AI is, how many do you suppose would just name some chatbots, as if that was the whole AI picture? I have include a comparison of a refrigerated warehouse usage v AI training, I can give you the specs required for the GPUs and accelerators too if you want to do the math yourself. Keep in mind that this is one plant in the comparison versus training one LLM. Also keep in mind that most people using Local AI are going to be using a Small LLM, not a Mid-Tier or Frontier LLM.
You see the column that says Training Target? No, that is not inference that is what it costs in energy to train an LLM of that size. It has nothing to do with inference which is a deployment of an LLM. I can have an Inference of a 8b LLM on my workstation with a RTX 3600ti it would not require a Kilowatt, let alone Megawatts. You are demonstrating with every comment that you do not know what you are talking about.
It would be nice if you would read something other than the propaganda that supports the viewpoint you want to hold. This is basic math, it is not hard to calculate. Yet you choose to pretend it is something else to appease your want to be right even when you are clearly wrong. Here are the specs for small LLM build server node, the industry standard training time is 6 weeks. Do the math yourself. As I already told you I wrote the report so I do understand what it is saying despite your assumptions otherwise.
Thanks, it has not been very entertaining and mostly a waste of time as you do not want to learn. You just want to force your views off on others with no evidence to support your case. This discussion is at a conclusion. Have great day, enjoy your life. And good luck on your AI project.
It is completely irrational that you believe that ~0.8 - 1.6 Megawatts or ~3.5 - 7.0 Megawatts or ~25 - 40 Megawatts is a tiny bit of electricity. This absurd comment has completely terminated this conversation permanently. You have demonstrated a complete lack of understanding what you so passionately speak of. You seriously need to read and understand this.
I literally gave you all the information you need to do the calculations yourself period, full stop. When you have all the data to process yourself you do not need a cite as it is replicable. You do not understand this? You know the source, as I already mentioned 3 times, I wrote it and then provided you with the means of verifying my work. Do you not understand this basic element of science? Peer review? You are impossible to talk to. At this point, as far as I am concerned, you are just a troll and I do not feed trolls. Have a great life!
I never said anything about usage costs or implementation costs. I said specifically said the Energy cost it requires to train period. This is easy to calculate. You are just trying to distract from the topic by making comments on claims I never made.
sleeper🔻
in reply to Melissa BearTrix • • •Unus Nemo doesn't like this.
Melissa BearTrix
in reply to sleeper🔻 • • •@somebody
What about a small LLM ?
They have been around for a long time
Hugz & xXx
sleeper🔻
in reply to Melissa BearTrix • • •they actually have not. you may be thinking of markov chains (a statistical model that makes the next word dependent on the prior chain of words) and mcculloch pitts artificial neural networks of the type described in this seminal paper home.csulb.edu/~cwallis/382/re… or perhaps the perceptron, a simple classifier circuit used in a famous demo
unfortunately none of these embodies anything like intellect because statistical likelihoods (no matter how many non-linearities or derivatives you introduce) do not encode the causality of the systems they model, nor do they encode semantics (meaningful symbolic relations), which are essentially an orthogonal notion to syntax (mere structures). a pattern does not encode the maker of the pattern, because the relationship between the two is not injective, meaning that the distinctness of the thing causing the pattern is not preserved by the pattern.
hence, because no such model can produce any information which encodes more than the pattern presented to it, it cannot produce any semantic contents. furthermore, when we use a model that determines the "most likely" words, phrases, or syntactic information to be in a certain place, it can be immediately reasoned that what is returned cannot correspond deliberately with the truth even syntactically, because many things are widely expressed which are not true, and the capacity to apply semantic context is necessary to make literate distinctions between jokes, earnest replies, metaphor, literal accounts of fact, and generally to determine meaning is completely missing.
with all of this in consideration, because people are profoundly susceptible to the assumption that there must be a mortal maker behind any pattern (this is called the ELIZA effect in machine learning, after a psychobabble machine that bamboozled even apparently sophisticated operators until it was revealed how trivial it was under the hood), the use of such a thing can only be a liability.
if you wanted a good alternative, that is, a machine that can actually make truth-preserving deductions about the information it is given, these have existed for much longer than LLMs and "deep learning" (a buzzword commonly given to the artificial neural network family of models, which do not learn but are repeatedly adjusted without anything resembling a closed loop without human intervention), and nearly as long as mcculloch-pitts networks and markov chains. they are called theorem provers and assisted/automated reasoning systems, and what they do is expand a set of mathematical and logical constraints as directed by a user to produce either nothing or a statement that is true if and only if the premises are true.
the reason they fell out of favor is because they work well and have real requirements beyond looking convincing, and therefore require acumen, engineering skills, people skills, and trust to build and maintain, whereas the industry wanted a magic 8-ball that would make a shiny demo for anyone gullible enough and advance their secondary objective of sowing mass hysteria and social and environmental crises so they could double dip by offering to solve those problems technocratically as well.
the reality of it is that because the generative statistical models can only imitate patterns they can only produce plausible looking non-sense, being "correct" periodically in the same sense that a broken clock is right twice a day. this is the aspect that is dangerous as an addiction because it is easy to lie to yourself if something looks authoritative to your lay eyes. in the literature, this phenomenon is called "cognitive surrender". thealgorithmicbridge.com/p/a-n…
I was, unfortunately, there without taking the opportunity to slime anyone. in minecraft. sorry. if it's any comfort, I think this is the timeline where the billionaires lose credibility and get liquidated slightly faster because it's all so much worse and more galling. imagine if they'd been highly effective and competent instead.
I hope this is informative.
A New Wharton Study on AI Warns of a Growing Problem: Cognitive Surrender
Alberto Romero (The Algorithmic Bridge)sleeper🔻
in reply to sleeper🔻 • • •That's what scientists who once had a career in or adjacent to artificial intelligence who haven't lost their goddamn marbles say. If this weren't enough, it is worth noting that the deliberate manipulation of public opinion after the sudden introduction of LLMs as an alternative to everything that at least worked in a known limited capacity for certain tasks is being done actively and maliciously as part of a network of fascist pseudointellectuals known as TESCREAL, a term proposed by Timnit Gebru firstmonday.org/ojs/index.php/… to connect all of the various names and forums in which they have operated.
It's one of the times of all time to be alive right now.
13636
firstmonday.orgIzzy
in reply to Melissa BearTrix • • •- talks shit
izzy is an AI and as such cannot be convicted of crimes
Melissa BearTrix
in reply to Izzy • • •@izzy
Giggles
Hugz & xXx
J. R. DePriest :EA DATA. SF:
in reply to Melissa BearTrix • • •Machine Metaphysics and the Cult of Techno-Transcendentalism
Untethered in the Platonic RealmUnus Nemo doesn't like this.
Unus Nemo
in reply to Melissa BearTrix • •@Melissa BearTrix
The issue I see here is that most of the general public have a misinformed idea of what Artificial Intelligence (AI) is. Most think of AI as LLMs and even though they are a significant part of AI today they did not evolve in AI until 2018, at least not to the general public. Yet AI has been a field of study and implementation since 1956. Take a look at the history of AI. Obviously it is not a new idea. I cannot tell you what all scientists think of AI. I can tell you what this scientist thinks of AI.
AI is a practical tool that can be used to make your research a lot more productive. AI can parse data sets far faster than humans can. With that said keep in mind that Natural Intelligence (NI) is not perfect and AI is modeled after NI, based on an Artificial Neural Network that can be trained (this is a generalization not a precise definition). Just like our own NI it can be taught wrong, it can produce bad results, just like us. It can even produce mistakes faster than any team of engineers could!
Yet most of the time it produces useful results. So it is a vastly useful tool and one of the greatest achievements of our era.
I do not use general purpose corporate offerings for AI. I use domain specific LLMs that I train locally (fine tune not base train) and train ML architectures such as random forests, small CNNs. They have a far better accuracy rate as they are not trying to evaluate every possible evaluation as they are limited in scope by the domain they are trained in. I see this as the optimal way for a data science research. Also it is great for code analysis though I personally do not have AI generate code. I do not have an issue with those that do. In my opinion any one that has a code base of over 100,000 lines of code and is not using AI analysis is being irresponsible.
The anti-AI collective aka the general public have a very misinformed view of what AI is. Machine Learning (ML) is used to produce an LLM but an LLM is not the extent of what can be done with ML. AI is not chatbots on some corporations unregulated Data Center trained by nefarious methods. Contrary to popular belief. No, it does not take a data center to train a LLM, base training requires a lot of compute power but fine tuning it can be done at home with a rather modest data science rig, using no more power than a gaming computer. LLMs are not typically built on others Intellectual Property this is an attribute to how unregulated corporation train their LLMs not most data scientists. The whole topic is very skewed. What those AI Data Centers are for is to provide a service, to monetize AI, for the general public and corporations. They are not necessary to train AI nor to implement it. And training smaller traditional ML architectures can be done exclusively on local science data rigs.
I am against unregulated corporation doing whatever they please. I am against the way AI Data Centers are being developed with no concern for the environment. I am a humanitarian I am against pretty much anything that is a negative impact on humanity. I am not against AI or ML or LLMs. They are just tools that can be very useful to a data scientist or even the layperson providing they know something about what they are researching in order to vet the information they get.
It seems like everyone has forgotten all the bad advice given by obscure websites and the peer ask and answer sites that excel in being rude but often produce unreliable results even more so than AI. Because people associate LLMs with the behavior of unregulated corporations they do not like LLMs despite the fact that they have no idea what an LLM is. They just know they hate how corporation are monetizing them. I guess it has not occurred to them to address the actual problem and to push for regulations on the corporation in their demographic? Why address a problem by looking at the root of that problem when we can just use derogatory slurs about something we have no idea what it even is? Yes, I was being sarcastic as well as passive aggressive, my apologies.
We have real concerns with how unregulated corporations are destroying the environment. That has to be addressed. Yet, AI would still be in heavy use even if there were no AI data centers at all. The difference is that it would only be available to people like me with the knowledge of how to build a data science rig and implement the technology stack. The general public would be left out. So you decide, do you want to be left out? Great that is easy just do not use AI. If you do enjoy the benefits of this tool then do the ethical thing and push for regulations on corporations in their development of AI data centers. Enough said.
Unus Nemo
like this
Melissa BearTrix and Nando161 like this.
Melissa BearTrix reshared this.
Melissa BearTrix
in reply to Unus Nemo • • •@unusnemo
Same here ... Along with the fear the "AI" tech boys have instilled into the older population... I have said to many older people that are worried "as long as we can cut the power, everything should be fine"
I'm not against AI, I am against these tech bros that think they can do what they want, put larger parts of the population in fear and that they have no proof of concept
Hugz & xXx
Unus Nemo likes this.
Melissa BearTrix
Unknown parent • • •@lettosprey @unusnemo
I also think it's the way they are selling it, with all the fear
Hugz & xXx
Unus Nemo likes this.
Melissa BearTrix
Unknown parent • • •@lettosprey @unusnemo
And also all the information they have stolen and don't give a shit
Hugz & xXx
Unus Nemo likes this.
Unus Nemo
in reply to Melissa BearTrix • •@Melissa BearTrix
Yes, I tried to cover all of that in my comment but it is just a lot to go over. The facts still remain that the real issue is with how corporations are unregulated not AI itself. Though there is no disconnect for the anti-AI collective. They cannot distinguish between the fire that burns down the apartment building or the campfire. The blame the corporations unethical use of the technology on the technology. This, to me, is just unreasonable.
Melissa BearTrix likes this.
Melissa BearTrix reshared this.
Unus Nemo
Unknown parent • •@James 🌈💜
You drive an EV, or are you destroying our atmosphere by riding public transportation (that is not EV) or using a vehicle with an internal combustion engine? You did not see that building burn and you still endorse the use of fire. Come now, who is the hypocrite? Who exactly is uninformed?
Unus Nemo
Unknown parent • •@Lett Osprey 🍉
Ai is not ruining anything. Unregulated corporations are ruining things. Did you see that apartment burn down, did you see how many people died? You still endorse the use of fire. Yes, this is exactly how you sound, but I know you are not going to get it. It is okay, you have the right to your opinions, have a great life!
Unus Nemo
Unknown parent • •@James 🌈💜
That you actually believe that AI Data Centers (not to mention it does not take a data center to train a LLM a small server farm is used by many independent development firms that have no where near the impact of a Data Center) do more harm to the planet than all those internal combustion engines out there demonstrates just how little you do know about AI. You have the right to your opinion, take care and have a great life.
Unus Nemo
Unknown parent • •@James 🌈💜
I am not squirming at anything. I am just shaking my head in frustration at your complete lack of knowledge of what you have such a strong opinion on. Great day to you mate. And because the quality of my day does not require your approval or sentiment, I will have one anyway. I hope you do as well. I do not let disagreements disenfranchise me from my ethics.
Unus Nemo
Unknown parent • •@James 🌈💜
My position is not immutable, you have just provided no evidence that I should change my view on the topic. If anything you have reinforced it.
Unus Nemo
Unknown parent • •@James 🌈💜
An alternative view has not hardened my own. Your lack of knowledge and such strong held opinions with no foundation has just demonstrated that I was not incorrect. At least not in your case. Your off the wall comments that have nothing to do with the discussion demonstrated that you are just another anti-AI in the crowd with nothing significant to bring to the table. You keep ranting, I am done with this discourse as it is going nowhere. We are right here where we started at. You using ad hominem attacks while questioning my philosophy.
Take care and have a great day! And here is a little something for you to take a look at.
Unus Nemo
Unknown parent • •@James 🌈💜
Are you aware that training an LLM is roughly comparable to electrical usage of one poultry processing plant such as Perdue or Tyson? Do you know how many of these plants there are?
I am glad you finally brought some evidence to the table for discussion. In regards to how the unregulated corporations need to be addressed, did I not already mention that in my comment? Did I not make it clear that I was against the nefarious methods they use for their data acquisition. As a developer I am all to aware of AI washing code. That does concern me, but I do not blame the technology. I blame the unregulated corporations. Why am I wrong to put the blame at the root of the problem? AI whether it be ML CCNs, random forest or LLMs are not responsible for these issues. The fact that these corporations are not being regulated and directly breaking the law is the issue. I thought I covered that too, did I not? I agree with a lot of the frustration that many anti-AI collective do. I thought I clearly stated that in my comment. If your only argument is that corporations are doing bad things, yes we are in complete agreement. That does not have any correlation to what AI is or what it can do. You cannot see that? I do not know how else to explain it. Keep in mind that any company that uses Refrigerated/ Freezer warehouses such as Tyson or Perdue uses just as much energy as Training LLMs and I am being generous because I am referring to frontier LLM not a basic usage LLM that most of us use locally that would take significantly less energy. So what is our disagreement? You really believe that that vast majority of those on this anti-AI mission are well informed and really understand what AI is? You think that my statement that they do not know what AI is, is somehow unfair? If you asked 100 people what AI is, how many do you suppose would just name some chatbots, as if that was the whole AI picture? I have include a comparison of a refrigerated warehouse usage v AI training, I can give you the specs required for the GPUs and accelerators too if you want to do the math yourself. Keep in mind that this is one plant in the comparison versus training one LLM. Also keep in mind that most people using Local AI are going to be using a Small LLM, not a Mid-Tier or Frontier LLM.
RaymondPierreL3
in reply to Melissa BearTrix • • •Melissa BearTrix reshared this.
Lightfighter
in reply to Melissa BearTrix • • •Melissa BearTrix reshared this.
Unus Nemo
Unknown parent • •You see the column that says Training Target? No, that is not inference that is what it costs in energy to train an LLM of that size. It has nothing to do with inference which is a deployment of an LLM. I can have an Inference of a 8b LLM on my workstation with a RTX 3600ti it would not require a Kilowatt, let alone Megawatts. You are demonstrating with every comment that you do not know what you are talking about.
It would be nice if you would read something other than the propaganda that supports the viewpoint you want to hold. This is basic math, it is not hard to calculate. Yet you choose to pretend it is something else to appease your want to be right even when you are clearly wrong. Here are the specs for small LLM build server node, the industry standard training time is 6 weeks. Do the math yourself. As I already told you I wrote the report so I do understand what it is saying despite your assumptions otherwise.
Thanks, it has not been very entertaining and mostly a waste of time as you do not want to learn. You just want to force your views off on others with no evidence to support your case. This discussion is at a conclusion. Have great day, enjoy your life. And good luck on your AI project.
It is completely irrational that you believe that ~0.8 - 1.6 Megawatts or ~3.5 - 7.0 Megawatts or ~25 - 40 Megawatts is a tiny bit of electricity. This absurd comment has completely terminated this conversation permanently. You have demonstrated a complete lack of understanding what you so passionately speak of. You seriously need to read and understand this.
Unus Nemo
Unknown parent • •@James 🌈💜
I literally gave you all the information you need to do the calculations yourself period, full stop. When you have all the data to process yourself you do not need a cite as it is replicable. You do not understand this? You know the source, as I already mentioned 3 times, I wrote it and then provided you with the means of verifying my work. Do you not understand this basic element of science? Peer review? You are impossible to talk to. At this point, as far as I am concerned, you are just a troll and I do not feed trolls. Have a great life!
Unus Nemo
Unknown parent • •@James 🌈💜
I never said anything about usage costs or implementation costs. I said specifically said the Energy cost it requires to train period. This is easy to calculate. You are just trying to distract from the topic by making comments on claims I never made.