We are in the AI Singularity
- Thread starter ClaytonB
- Start date
- Joined
- Nov 5, 2010
- Messages
- 39,958
ClaytonB
Member
- Joined
- Oct 30, 2011
- Messages
- 10,227
This sermon is SPOT ON... this is a must-watch...
Artificial Intelligence is not "just another technology", like the Gutenberg Press. A Big Tech executive recently suggested it's the most significant invention since fire. However, the unbelieving world does not really understand why AI is such a big deal. The two primary reasons that AI is a big-deal are (a) AI is based on a facsimile of the image of God in man and (b) AI uniquely fulfills several apocalyptic prophecies in a way that no other technology ever has or ever will.
Point (a) is the most overlooked by the unbelieving world -- "So what? We're copying your 'God'. What of it?" But these very same people will turn around and sue their competitors into a smoking hole in the ground for "copyright violations". Some copyright lawyers recently copyrighted a hard-drive containing an exhaustive enumeration of every single 8-note melody (and put use-rights in the public-domain) to make the point that copyright law is out-of-control. Before copyright, how were copyrights enforced? What stopped people from copying one another? There are two answers -- on the one hand, nothing stopped them, people copied each other like mad and, on the other hand, performers were constantly scrambling to present something to the public audience that was original. So, ticket-sales / entrance-fees drove people to actually develop original content instead of just copying the old songs and shows that had been played a thousand times before.
The words we're searching for, here, are worship and glory. A performer who performs an enjoyable and original act receives applause (a form of worship, when you think about it), because he is worthy. Originality (creativity), then, is a manifestation of glory. In particular, it is a manifestation of the glory of the residual image of God within man. And this is precisely the thing which modern AI is trying to hijack and put under lock-and-key as the private property of XYZ Corp. The image of God in man is not the property of M$, it's not the property of OpenAI, it's not the property of DeepMind or Google or Apple or Amazon or Netflix or Baidu or Tencent or Tik-Tok or Huawei or fill-in-the-blank. No man on earth, no organization of men, no 'political movement' or other movement has any property claim to the image of God, whatsoever.
In the Apocalypse, God is going to bring the biggest Intellectual Property lawsuit ever, and the defendants of that lawsuit will be every single person who has ever lived. Your "religion" won't matter, your political-party won't matter, your nationality won't matter, your ethnicity won't matter... everyone who has ever lived will be named in God's IP lawsuit and ordered to appear before his judgment seat. And everyone will know for an absolute fact that they are guilty, from the very moment they are served with papers. No one will be able to raise any defense on any basis, no matter who they are, no matter what they've "gone through", and no matter how "perfect" they falsely imagine they are. The only escape will be in Jesus. Those who are building this Beast AI system are doing so precisely in order to take on God in the Apocalypse. Just read the book! That's what it's all really about.
Wake up! Repent! The Kingdom of God is at hand!
Artificial Intelligence is not "just another technology", like the Gutenberg Press. A Big Tech executive recently suggested it's the most significant invention since fire. However, the unbelieving world does not really understand why AI is such a big deal. The two primary reasons that AI is a big-deal are (a) AI is based on a facsimile of the image of God in man and (b) AI uniquely fulfills several apocalyptic prophecies in a way that no other technology ever has or ever will.
Point (a) is the most overlooked by the unbelieving world -- "So what? We're copying your 'God'. What of it?" But these very same people will turn around and sue their competitors into a smoking hole in the ground for "copyright violations". Some copyright lawyers recently copyrighted a hard-drive containing an exhaustive enumeration of every single 8-note melody (and put use-rights in the public-domain) to make the point that copyright law is out-of-control. Before copyright, how were copyrights enforced? What stopped people from copying one another? There are two answers -- on the one hand, nothing stopped them, people copied each other like mad and, on the other hand, performers were constantly scrambling to present something to the public audience that was original. So, ticket-sales / entrance-fees drove people to actually develop original content instead of just copying the old songs and shows that had been played a thousand times before.
The words we're searching for, here, are worship and glory. A performer who performs an enjoyable and original act receives applause (a form of worship, when you think about it), because he is worthy. Originality (creativity), then, is a manifestation of glory. In particular, it is a manifestation of the glory of the residual image of God within man. And this is precisely the thing which modern AI is trying to hijack and put under lock-and-key as the private property of XYZ Corp. The image of God in man is not the property of M$, it's not the property of OpenAI, it's not the property of DeepMind or Google or Apple or Amazon or Netflix or Baidu or Tencent or Tik-Tok or Huawei or fill-in-the-blank. No man on earth, no organization of men, no 'political movement' or other movement has any property claim to the image of God, whatsoever.
In the Apocalypse, God is going to bring the biggest Intellectual Property lawsuit ever, and the defendants of that lawsuit will be every single person who has ever lived. Your "religion" won't matter, your political-party won't matter, your nationality won't matter, your ethnicity won't matter... everyone who has ever lived will be named in God's IP lawsuit and ordered to appear before his judgment seat. And everyone will know for an absolute fact that they are guilty, from the very moment they are served with papers. No one will be able to raise any defense on any basis, no matter who they are, no matter what they've "gone through", and no matter how "perfect" they falsely imagine they are. The only escape will be in Jesus. Those who are building this Beast AI system are doing so precisely in order to take on God in the Apocalypse. Just read the book! That's what it's all really about.
Wake up! Repent! The Kingdom of God is at hand!
Last edited:
- Joined
- Nov 5, 2010
- Messages
- 39,958
acptulsa
Member
- Joined
- Jan 2, 2008
- Messages
- 75,396
Interesting that there isn't one man in it. Seems AI doesn't go around in the real world, putting its life on the line crossing bridges, taking shelter in buildings, fixing its own fiber optic cable. It just listens to the noise.
No wonder it can't see half the human race.
Last edited:
ClaytonB
Member
- Joined
- Oct 30, 2011
- Messages
- 10,227
Brace Yourselves: Artificial General Intelligence Is Just Around The Corner
Been holding off on this particular warning for a while now. But the time has come to sound the warning on this because I don't think that Google is making idle "marketing" claims, here. General-purpose thinking (the kind of thinking we do when solving a puzzle or assembling the pieces of Ikea furniture, etc.) can be thought of as a kind of game, in the sense of game-theory.
An LLM is also a kind of tree, but of words. At each branch of that "word tree", you need to make choices about how to "move forward", as in a game of chess. And the LLM does this by what is really an extremely sophisticated and fine-tuned probability table. It doesn't feel like a probability table but it really is just one, giant, probability table that has been highly compressed (like a Zip-file). The transformer model is really acting as a kind of "compression algorithm" and it is the compression that makes LLMs feel intelligent. The compression algorithm is mapping related "concepts" in a very high-dimensional space, in order to cram the training signals received during training into a seemingly impossibly-small space... as little as 4 gigabytes for some reasonably-performant LLMs.
The sequence of "all possible thoughts" can itself be thought of as a giant branching tree like this, but the problem with current LLMs is that they simply "roll the dice" at every branch. Those rolls are not really over the space of "the next word", they are over the space of all compressed "embeddings", one of which will be chosen, and decoded to a word (which will be the next word). The embeddings contain the probability information, but the key is to realize that there are vastly more embeddings than there are words, and the embeddings themselves are something more akin to thought vectors. We have no idea what is in these "thought-vectors", so they are not quite the same thing, but they behave like that.
What Google is doing is they are applying the same kind of framework underlying AlphaZero to this branching thought-tree. AlphaZero was designed to learn a game without any human training. So, what they are probably doing is using a foundational language model (perhaps Bard) as the "rule-book", and they are training AlphaZero against some score metric to make "legal moves" within that rule-book, to maximize the score metrics. Basically, when AlphaZero says, "The chicken was slivy" the LLM fires a rule-violation because "slivy" doesn't make any sense in this context (or any context), and so AlphaZero has to choose a different "move" (a syntactically valid word). So, it tries some other word, e.g. "The chicken was running" and the LLM now allows it. But AlphaZero is not being scored by the LLM, the LLM is just the syntactic referee, like an English teacher scoring your essay purely on the correctness of the grammar while ignoring the actual topic of discussion in the essay. Nevertheless, since an LLM does have both semantic and syntactic understanding of language, this is a very well-educated English teacher, so your essay needs to really make sense, even though you may make factually-false claims in the essay as long as the English teacher doesn't particularly know about that topic. This brings us to the scoring -- the actual scoring will be based on some kind of factual semantic knowledge, such as the many semantic AI tests that are already in use for scoring LLMs. The difference is that AlphaZero + LLM can do massively better than LLM alone can. Instead of topping out at 50% on various logic puzzle tasks, AlphaZero + LLM is going to be able to do as well as an intelligent human would do if they are really concentrating, and probably much better than the human average. In short, when they launch it, this will be the message: AI is now smarter than humans.
And it's not over. There is more coming...
Been holding off on this particular warning for a while now. But the time has come to sound the warning on this because I don't think that Google is making idle "marketing" claims, here. General-purpose thinking (the kind of thinking we do when solving a puzzle or assembling the pieces of Ikea furniture, etc.) can be thought of as a kind of game, in the sense of game-theory.
An LLM is also a kind of tree, but of words. At each branch of that "word tree", you need to make choices about how to "move forward", as in a game of chess. And the LLM does this by what is really an extremely sophisticated and fine-tuned probability table. It doesn't feel like a probability table but it really is just one, giant, probability table that has been highly compressed (like a Zip-file). The transformer model is really acting as a kind of "compression algorithm" and it is the compression that makes LLMs feel intelligent. The compression algorithm is mapping related "concepts" in a very high-dimensional space, in order to cram the training signals received during training into a seemingly impossibly-small space... as little as 4 gigabytes for some reasonably-performant LLMs.
The sequence of "all possible thoughts" can itself be thought of as a giant branching tree like this, but the problem with current LLMs is that they simply "roll the dice" at every branch. Those rolls are not really over the space of "the next word", they are over the space of all compressed "embeddings", one of which will be chosen, and decoded to a word (which will be the next word). The embeddings contain the probability information, but the key is to realize that there are vastly more embeddings than there are words, and the embeddings themselves are something more akin to thought vectors. We have no idea what is in these "thought-vectors", so they are not quite the same thing, but they behave like that.
What Google is doing is they are applying the same kind of framework underlying AlphaZero to this branching thought-tree. AlphaZero was designed to learn a game without any human training. So, what they are probably doing is using a foundational language model (perhaps Bard) as the "rule-book", and they are training AlphaZero against some score metric to make "legal moves" within that rule-book, to maximize the score metrics. Basically, when AlphaZero says, "The chicken was slivy" the LLM fires a rule-violation because "slivy" doesn't make any sense in this context (or any context), and so AlphaZero has to choose a different "move" (a syntactically valid word). So, it tries some other word, e.g. "The chicken was running" and the LLM now allows it. But AlphaZero is not being scored by the LLM, the LLM is just the syntactic referee, like an English teacher scoring your essay purely on the correctness of the grammar while ignoring the actual topic of discussion in the essay. Nevertheless, since an LLM does have both semantic and syntactic understanding of language, this is a very well-educated English teacher, so your essay needs to really make sense, even though you may make factually-false claims in the essay as long as the English teacher doesn't particularly know about that topic. This brings us to the scoring -- the actual scoring will be based on some kind of factual semantic knowledge, such as the many semantic AI tests that are already in use for scoring LLMs. The difference is that AlphaZero + LLM can do massively better than LLM alone can. Instead of topping out at 50% on various logic puzzle tasks, AlphaZero + LLM is going to be able to do as well as an intelligent human would do if they are really concentrating, and probably much better than the human average. In short, when they launch it, this will be the message: AI is now smarter than humans.
And it's not over. There is more coming...
ClaytonB
Member
- Joined
- Oct 30, 2011
- Messages
- 10,227
ChatGPT in trouble: OpenAI sued for stealing everything anyone’s ever written on the Internet - OpenAI's ChatGPT and Sam Altman are in massive trouble. OpenAI is getting sued in the US for illegally using content from the internet to train their LLM or large language models
Similar threads
