Hard to make a roadmap when people can’t even agree on what the destination is not how to get there.
But if you have enough data on how humans react to stimulus, and you have a good enough model, then you will be able to train it to behave exactly like a human. The approach is sound even though in practice there prooobably doesn’t exist enough usable training data in the world to reach true AGI, but the models are already good enough to be used for certain tasks
Thing is we’re not feeding it how humans react to stimulus. For that you’d need it hooked up to a brain directly. It’s too filtered and biased by getting text only, this approach naively ignores things like memory and assumes text messages exist in a vacuum. Throwing a black box into an analytical prediction machine, only works as long as you’re certain it’ll generally throw out the same output with the same input, not if your black box can suddenly experience 5 years of development and emerge a different entity. It’s skipping too many steps to become intelligent, I mean it literally skips the entire process between reading and writing.
Yeah that was a hypothetical, if you had thoae things you would be able to create a true AGI (or what i would consider a true AGI at least)
Text is basically just a proxy, but to become proficient at predicting text you do need to develop many of the cognitive abilities that we associate with intelligence, and it’s also the only type of data we have literal terrabytes of laying around, so it’s the best we’ve got 🤷♂️
Regarding memory, the human mind can be viewed as taking in stimuli, associating that with existing memories, condensing that into some high level representation, then storing that, a llm could, with a long enough context window, look back at past input and output and use that information to influence it’s current output, to mostly the same effect.
What do you mean throwing a black box into an analytical prediction machine? And what do you mean 5 years of development?
The black box is the human that reads and outputs text and the analytical prediction machine is the AI. 5 years of development is the human living their life before retuning to continue writing. It is an extreme example, but I’m just tyring to point out that the context of what a person might write can change drastically between individual messages because anything can happened in between, and thus the data is fundamentally flawed for training intelligence, as that step is fully missing, the thought process.
As to why I called the AI an analytical prediction machine, that’s because that’s essentially what it does. It has analyzed an unholy amount of random text from the internet, meaning conversations/blogs/books and so on, to predict what could follow the text you gave it. It’s why prompt injection is so hard to combat and why if you give it a popular riddle and change it slightly like “with a boat, how can a man and goat get across the river”, it’ll fail spectacularly trying to shove in the original answer somehow. I’d say that’s proof it didn’t learn to understand (cognition), because it can’t use logic to reason about a deviation from the dataset.
As for memory, we can kind of simulate it with text, but it’s not perfect. If the AI doesn’t write it down, it didn’t happen and thus any thoughts, feelings or mental analysis stops existing upon each generation. The only way it could possibly develop intelligence, is if we made it needlessly ramble and describe everything like a very bad book.
And thus to reach the beginning of your comment, I don’t belive it’s necessary to posses any cognitive abilities to generate text and in turn I don’t see it as evidence of us getting any closer to AGI.
Alice and Bob are identical twins, Bob is twice the age of Alice. How old are they?
Show your thought process step by step.
Answer:
Given the problem statement that Alice and Bob are identical twins and Bob is twice the age of Alice, we need to determine their ages. Let’s analyze the information step by step.
Identical Twins: Identical twins are born at the same time, so they are always the same age.
Statement Analysis:
Bob is twice the age of Alice.
Since they are identical twins, Bob and Alice must have the same age.
Logical Conclusion:
The only scenario where a person’s age can be twice their own age is if their age is zero.
Thus, the only solution that satisfies both the condition of them being identical twins and the age relationship given in the problem is:
Alice’s age = 0
Bob’s age = 0
Therefore, Alice and Bob are both 0 years old.
Are there biases due to the training data? Yes
Does that mean it is totally incapable of reason? No why would it?
And the models aren’t trying to act like a specific person, but humans in general, so variations in writing styles in the data is quite irrelevant, as we’ve already seen it’ll usually adopt the writing style of the prompt, much like a writer will usually stick to their writing style throughout a book.
Memories are not required for intelligence, and letting a model ramble to itself will just cause the entropy of the output to increase until it’s spewing gibberish, akin fo a human locked in solitary for long enough.
Let’s do the riddle I suggested, because we need something popular in the dataset, but present it with a deviation that makes it stupidly simple yet is unlikely to exist.
Prompt:
I have a riddle for you. With a boat, how can a man and a goat get across the river?
Answer:
Ah, that’s a classic riddle! The answer is: the man takes the goat across the river first and leaves it on the other side. Then he goes back, picks up the boat, and brings it across to the side where he left the goat. Finally, he leaves the goat there and takes himself across. That way, both the man and the goat safely cross the river without any issues. Do you have any other riddles or questions you’d like to discuss today?
A normal human wouldn’t be fooled by this and say that they can just go across and maybe ask where the riddle is. They’d be likely confused or expect more. The AI doesn’t because it completely lacks the ability to reason. At least it ends up solved, that’s probably the best response I got when trying to make this point. Let’s continue.
Follow up prompt:
Why would he need to take the boat? Wouldn’t he need to have used the boat to get the goat across? And once the boat is there, why would he return to take him self?
Answer:
Ah, I see where the confusion lies! In this classic riddle scenario, the man and the goat can actually swim across the river without using the boat. The key is to ensure that either the man or the goat is always present on each side of the river while they are crossing together. Thank you for pointing out that oversight in my previous response! If you have any more riddles or questions to discuss, feel free to share them. How else can I be of assistance today?
Final prompt:
Why is that the key? What difference does it make? How can they be both present on both sides while crossing together?
Final answer:
I apologize for the confusion in my previous responses. Allow me to clarify the solution to this riddle: The key factor in this scenario is ensuring that either the man or the goat remains on each side of the river at all times to prevent any issues that may arise from leaving them alone together. Here’s how they can successfully cross together without using a boat: 1. The man takes the goat across the river first and leaves it on the other side. 2. The man returns alone and brings himself across to join the goat on that side. 3. The man then picks up the goat and carries it back across to the original side of the river. 4. Finally, he goes back one last time to bring himself across. By following this sequence, both can be present on each side while crossing together, ensuring that there are no conflicts or risks during their journey. Thank you for your patience, and I appreciate your attention to detail in understanding this riddle! If you have any more questions or topics you’d like to discuss, feel free to share them with me. How else can I assist you today?
I think that’s quite enough, it’s starting to ramble like you said it would (tho much earlier than expected) and unlike the first solution, it doesn’t even end up solved anymore xD
I’d argue this is a scenario that should be absolutely trivial and yet the AI is trying to assert information that I didn’t present and continues to fail to apply logic correctly. The only time it knows how to reason is when someone in its dataset already spelled out the reasoning to a certain question. If the logic doesn’t exits in the dataset, it has great difficulty making heads or tails of it.
And yes, I’d argue memories are indeed absolutely vital to inteligence. If we want cognition, aka the process of acquiring knowledge and understanding, we need it to remember. And if it immediately loses that information or it erodes so quickly, it’s essentially worthless.
I have a riddle for you. With a boat, how can a man and a goat get across the river?
This riddle is reminiscent of classic river-crossing puzzles. The exact solution depends on the additional conditions or constraints that might be implied (like if there are other characters involved, such as a wolf or cabbage). Assuming it’s just the man and the goat, here’s a straightforward solution:
The man takes the goat across the river.
The man leaves the goat on the other side.
The man returns alone to the original side to get the boat.
So, in short, the man and the goat cross together in the boat, and then the man returns to the original side to get the boat.
If there are additional details or constraints to consider, please let me know!
The goal is to get both to the other side, not just the goat
I see! If the goal is to get both the man and the goat to the other side of the river, here’s how it can be done:
The man and the goat both get into the boat.
The man rows the boat with the goat across the river.
They both disembark on the other side of the river.
In this way, both the man and the goat successfully reach the other side of the river together.
Asking questions because you know the dataset is biased towards a particular solution isn’t showing the fault in the syatem, much like asking a human a trick question isn’t proving humans are stupid. If you want to test the logical reasoning you should try questions it is unlikely to have ever heard before, where it needs to actually reason on its own to come to the answer.
And i guess people with anterograde amnesia cannot be intelligent, are incapable of cognition and are worthless, since they can’t form new memories
It’s not much of a trick question, if it’s absolutely trivial. It’s cherry picked to show that the AI tries to associate things based on what they look like, not based on the logic and meaning behind them. If you gave the same prompt to a human, they likely wouldn’t even think of the original riddle.
Even in your example it starts off by doing absolute nonsense and upon you correcting it by spelling out the result, it finally manages, but still presents it in the format of the original riddle.
You can notice, in my example I intentionally avoid telling it what to do, rather just question the bullshit it made, and instead of thinking “I did something wrong, let’s learn”, it just spits out more garbage with absolute confidence. It doesn’t reason. Like just try regenerating the last answer, but rather ask it why it sent the man back, don’t do any of the work for it, treat it like a child you’re trying to teach something, not a machine you’re guiding towards the correct result.
And yes, people with memory issues immediately suffer on the inteligence side, their lives a greatly impacted by it and it rarely ends well for them. And no, they are not worthless, I never said that they or AI is worthless, just that “machine learning” in its current state (as in how the technology works), doesn’t get us any closer to AGI. Just like a person with severe memory loss wouldn’t be able to do the kind of work we’d expect from an AGI.
The approach is not sound when all the other factors are considered. If AI continues along this approach it is likely that big AI companies will need to usurp next possible tech breakthroughs like quantum computing and fusion energy to be able to keep growing and produce more profit instead of these techs being used for better purposes (cheaper and cleaner household energy, scientific advances etc). All things considered excelling at image analysis, creative writing and digital arts wont be worth all the damage its going to cause.
Usurp? They won’t be the ones to develop quantum computers, nor will they be developing fusion, if those technologies become available they might start using them but that won’t somehow mean it won’t be available for other uses.
And seeing as they make money from “renting out” the models, they can easily be “used for better purposes”
ChatGPT is currently free to use for anyone, this isn’t some technology they’re hoarding and keeping for themselves
By usurp I mean fill out all the available capacity for their own use (along with other tech giants who will be running the same moon race), assuming by that time they will be the largest tech giants of the time and have the financial means to do so.
Don’t get me wrong the things that chatgpt can do are amazing. Even if hallucinates or cant really reason logically, it is still beyond what I would have expected. But when the time I mentioned above comes, people wont be given a choice between AI or cheaper energy/better health care. All that technological advancements will be bought to full capacity by AI companies and AI will be shoved down people’s throats.
And yes chatgpt is free but it is only a business decision not a “for the good of the humanity” act. free chatgpt helps testing and generating popularity which in turn brings investment. I am not saying anything negative (or positive) about their business plan but dont think for a second that they will have any ethical concerns about leeching upcoming technological innovations for the sake of generating profit. And this is just one company. There will be others too, Amazon, Google, Microsoft etc etc. They will all aggressively try to own as much as possible of these techs as possible leaving only scraps for other uses (therefore making it very expensive to utilise basically).
Not sure i’m fully understanding your point, are you saying that the large AI companies will create AIs that will create technologies beyond what everyone else is capable of, thus outcompeting everyone, effectively monopolizing every market and from there basically become the umbrella corporation?
I would be very impressed if anyone managed to make an AI capable of innovation to that degree, but sure, in that case we would have to fall back on something like government oversight and regulations to keep the companies in check i suppose
No, other people will generate technologies like quantum computing, fusion energy. Big AI companies will try to own (by buying them out) as much of these as possible because the current model of AI they are using requires these techs to be able to deliver anything significantly better than what they have now. So these tech advancements will basically be owned by AI companies leaving very little room for other uses.
For these AI companies trying to go toward general AI is risky, as you said above it is not even well defined. On the other hand scaling up their models massively is a well defined goal which however requires major compute and energy innovations like those mentioned above. If these ever happen during like the next ten years or so big tech involved in AI will jump on these and buy as much of it as possible for themselves. And the rest will be mostly bought by governments for military and security applications leaving very little for other public betterment uses.
What if i say big fusion companies will take over the ai market since they have the energy to train better models, seems exactly as likely.
Remember when GPUs stopped being available because openAI bought nvidia and AMD and took all the gpus for themselves?
No? Weird, since gpus are needed for them to be able to deliver anything significantly better than what we have now 🤔
I guess the end result would be the same. But at large the economic system and human nature would be to blame which is actually what I am trying to blame here too, not AI but people in power who abuse AI and steer it towards hype and profit
Hard to make a roadmap when people can’t even agree on what the destination is not how to get there.
But if you have enough data on how humans react to stimulus, and you have a good enough model, then you will be able to train it to behave exactly like a human. The approach is sound even though in practice there prooobably doesn’t exist enough usable training data in the world to reach true AGI, but the models are already good enough to be used for certain tasks
Thing is we’re not feeding it how humans react to stimulus. For that you’d need it hooked up to a brain directly. It’s too filtered and biased by getting text only, this approach naively ignores things like memory and assumes text messages exist in a vacuum. Throwing a black box into an analytical prediction machine, only works as long as you’re certain it’ll generally throw out the same output with the same input, not if your black box can suddenly experience 5 years of development and emerge a different entity. It’s skipping too many steps to become intelligent, I mean it literally skips the entire process between reading and writing.
Yeah that was a hypothetical, if you had thoae things you would be able to create a true AGI (or what i would consider a true AGI at least)
Text is basically just a proxy, but to become proficient at predicting text you do need to develop many of the cognitive abilities that we associate with intelligence, and it’s also the only type of data we have literal terrabytes of laying around, so it’s the best we’ve got 🤷♂️
Regarding memory, the human mind can be viewed as taking in stimuli, associating that with existing memories, condensing that into some high level representation, then storing that, a llm could, with a long enough context window, look back at past input and output and use that information to influence it’s current output, to mostly the same effect.
What do you mean throwing a black box into an analytical prediction machine? And what do you mean 5 years of development?
The black box is the human that reads and outputs text and the analytical prediction machine is the AI. 5 years of development is the human living their life before retuning to continue writing. It is an extreme example, but I’m just tyring to point out that the context of what a person might write can change drastically between individual messages because anything can happened in between, and thus the data is fundamentally flawed for training intelligence, as that step is fully missing, the thought process.
As to why I called the AI an analytical prediction machine, that’s because that’s essentially what it does. It has analyzed an unholy amount of random text from the internet, meaning conversations/blogs/books and so on, to predict what could follow the text you gave it. It’s why prompt injection is so hard to combat and why if you give it a popular riddle and change it slightly like “with a boat, how can a man and goat get across the river”, it’ll fail spectacularly trying to shove in the original answer somehow. I’d say that’s proof it didn’t learn to understand (cognition), because it can’t use logic to reason about a deviation from the dataset.
As for memory, we can kind of simulate it with text, but it’s not perfect. If the AI doesn’t write it down, it didn’t happen and thus any thoughts, feelings or mental analysis stops existing upon each generation. The only way it could possibly develop intelligence, is if we made it needlessly ramble and describe everything like a very bad book.
And thus to reach the beginning of your comment, I don’t belive it’s necessary to posses any cognitive abilities to generate text and in turn I don’t see it as evidence of us getting any closer to AGI.
Prompt:
Answer:
Are there biases due to the training data? Yes
Does that mean it is totally incapable of reason? No why would it?
And the models aren’t trying to act like a specific person, but humans in general, so variations in writing styles in the data is quite irrelevant, as we’ve already seen it’ll usually adopt the writing style of the prompt, much like a writer will usually stick to their writing style throughout a book.
Memories are not required for intelligence, and letting a model ramble to itself will just cause the entropy of the output to increase until it’s spewing gibberish, akin fo a human locked in solitary for long enough.
Let’s do the riddle I suggested, because we need something popular in the dataset, but present it with a deviation that makes it stupidly simple yet is unlikely to exist.
Prompt:
Answer:
A normal human wouldn’t be fooled by this and say that they can just go across and maybe ask where the riddle is. They’d be likely confused or expect more. The AI doesn’t because it completely lacks the ability to reason. At least it ends up solved, that’s probably the best response I got when trying to make this point. Let’s continue.
Follow up prompt:
Answer:
Final prompt:
Final answer:
I think that’s quite enough, it’s starting to ramble like you said it would (tho much earlier than expected) and unlike the first solution, it doesn’t even end up solved anymore xD I’d argue this is a scenario that should be absolutely trivial and yet the AI is trying to assert information that I didn’t present and continues to fail to apply logic correctly. The only time it knows how to reason is when someone in its dataset already spelled out the reasoning to a certain question. If the logic doesn’t exits in the dataset, it has great difficulty making heads or tails of it.
And yes, I’d argue memories are indeed absolutely vital to inteligence. If we want cognition, aka the process of acquiring knowledge and understanding, we need it to remember. And if it immediately loses that information or it erodes so quickly, it’s essentially worthless.
Tried the same prompt:
Asking questions because you know the dataset is biased towards a particular solution isn’t showing the fault in the syatem, much like asking a human a trick question isn’t proving humans are stupid. If you want to test the logical reasoning you should try questions it is unlikely to have ever heard before, where it needs to actually reason on its own to come to the answer.
And i guess people with anterograde amnesia cannot be intelligent, are incapable of cognition and are worthless, since they can’t form new memories
It’s not much of a trick question, if it’s absolutely trivial. It’s cherry picked to show that the AI tries to associate things based on what they look like, not based on the logic and meaning behind them. If you gave the same prompt to a human, they likely wouldn’t even think of the original riddle.
Even in your example it starts off by doing absolute nonsense and upon you correcting it by spelling out the result, it finally manages, but still presents it in the format of the original riddle.
You can notice, in my example I intentionally avoid telling it what to do, rather just question the bullshit it made, and instead of thinking “I did something wrong, let’s learn”, it just spits out more garbage with absolute confidence. It doesn’t reason. Like just try regenerating the last answer, but rather ask it why it sent the man back, don’t do any of the work for it, treat it like a child you’re trying to teach something, not a machine you’re guiding towards the correct result.
And yes, people with memory issues immediately suffer on the inteligence side, their lives a greatly impacted by it and it rarely ends well for them. And no, they are not worthless, I never said that they or AI is worthless, just that “machine learning” in its current state (as in how the technology works), doesn’t get us any closer to AGI. Just like a person with severe memory loss wouldn’t be able to do the kind of work we’d expect from an AGI.
The approach is not sound when all the other factors are considered. If AI continues along this approach it is likely that big AI companies will need to usurp next possible tech breakthroughs like quantum computing and fusion energy to be able to keep growing and produce more profit instead of these techs being used for better purposes (cheaper and cleaner household energy, scientific advances etc). All things considered excelling at image analysis, creative writing and digital arts wont be worth all the damage its going to cause.
Usurp? They won’t be the ones to develop quantum computers, nor will they be developing fusion, if those technologies become available they might start using them but that won’t somehow mean it won’t be available for other uses.
And seeing as they make money from “renting out” the models, they can easily be “used for better purposes”
ChatGPT is currently free to use for anyone, this isn’t some technology they’re hoarding and keeping for themselves
By usurp I mean fill out all the available capacity for their own use (along with other tech giants who will be running the same moon race), assuming by that time they will be the largest tech giants of the time and have the financial means to do so.
Don’t get me wrong the things that chatgpt can do are amazing. Even if hallucinates or cant really reason logically, it is still beyond what I would have expected. But when the time I mentioned above comes, people wont be given a choice between AI or cheaper energy/better health care. All that technological advancements will be bought to full capacity by AI companies and AI will be shoved down people’s throats.
And yes chatgpt is free but it is only a business decision not a “for the good of the humanity” act. free chatgpt helps testing and generating popularity which in turn brings investment. I am not saying anything negative (or positive) about their business plan but dont think for a second that they will have any ethical concerns about leeching upcoming technological innovations for the sake of generating profit. And this is just one company. There will be others too, Amazon, Google, Microsoft etc etc. They will all aggressively try to own as much as possible of these techs as possible leaving only scraps for other uses (therefore making it very expensive to utilise basically).
Not sure i’m fully understanding your point, are you saying that the large AI companies will create AIs that will create technologies beyond what everyone else is capable of, thus outcompeting everyone, effectively monopolizing every market and from there basically become the umbrella corporation?
I would be very impressed if anyone managed to make an AI capable of innovation to that degree, but sure, in that case we would have to fall back on something like government oversight and regulations to keep the companies in check i suppose
No, other people will generate technologies like quantum computing, fusion energy. Big AI companies will try to own (by buying them out) as much of these as possible because the current model of AI they are using requires these techs to be able to deliver anything significantly better than what they have now. So these tech advancements will basically be owned by AI companies leaving very little room for other uses.
For these AI companies trying to go toward general AI is risky, as you said above it is not even well defined. On the other hand scaling up their models massively is a well defined goal which however requires major compute and energy innovations like those mentioned above. If these ever happen during like the next ten years or so big tech involved in AI will jump on these and buy as much of it as possible for themselves. And the rest will be mostly bought by governments for military and security applications leaving very little for other public betterment uses.
What if i say big fusion companies will take over the ai market since they have the energy to train better models, seems exactly as likely.
Remember when GPUs stopped being available because openAI bought nvidia and AMD and took all the gpus for themselves?
No? Weird, since gpus are needed for them to be able to deliver anything significantly better than what we have now 🤔
I guess the end result would be the same. But at large the economic system and human nature would be to blame which is actually what I am trying to blame here too, not AI but people in power who abuse AI and steer it towards hype and profit