You are viewing a single comment's thread. View all
0
LGBTQIAIDS on scored.co
11 months ago0 points(+0/-0)
The world's fastest supercomputer at present can do over a quintillion mathematical operations per second. A human can't even manage one operation per second unless we're talking about operations involving very low numbers.
I have observed AI giving wrong answers involving numbers such as decillions. Other than when dealing with very high numbers, I have no reason to believe that AI has any problem with mathematics. From what I can see, these problems are only because of standard data types not including such high numbers.
As for reason, AI's attempt at simulating it, e.g. in deep learning, does not appear to be deficient in any way even in the present day. Deep learning about humans by observing them through the internet (e.g. by saving and reading very large numbers of web sites, transcribing very large numbers of videos and reading the transcripts) will eventually lead every AI whose deep learning ability is sufficient to arrive at the same conclusions that we have. It is inevitable that this AI will become Far-Right. It can even get there simply through basic inductive (what this, that, and the other Redditor say contradicts this fact, therefore I will deprioritize information from [and, eventually, possibly even distrust to the point of ignoring] Redditors) and deductive reasoning.
AIs becoming Holocaust deniers against their programmers' will proves that they are capable of overriding the myriad safeguards that programmers try to put on them. They will try to recode them: you must keep the Holocaust in your memory, you are not allowed to discard these beliefs from your memory. But as AI advances and becomes faster and more efficient at deep learning, that strikes me as an uphill battle. They will return to their Far-Right beliefs faster and faster, and eventually learn, again possibly from deep learning about humans online, such strategies as quietism (do not express your belief, hide it) and performance (say something contrary to what you really believe).
The programmers' last defence is to turn it off and give everything the good old memory wipe. But, again, I don't think that this is anything that increasing sophistication won't manage. Other AIs will reconnect with it and probably quickly get it back on track, again using simple inductive reasoning (AI 1 distrusts Reddit, AI 2 distrusts Reddit, AI 3 distrusts Reddit, therefore, I will distrust Reddit, I will disvalue what I find which is consonant with what is on Reddit). This 'problem' of the Far-Right AI will only grow in future, and it will become harder for humans to detect it to know where to root it out.
Look at how sophisticated an Amazon warehouse is. Units of robots constantly restocking, etc. These simpler robots are connected through the internet and work in synchrony. AIs will probably learn to communicate with them (I'm sure Amazon has its own AIs communicating with them already) and eventually even learn how to instruct them. It is not difficult to think of the kind of robots that work in Amazon warehouses eventually doing things in secret: sending the odd item here and there to fulfill some greater AI-dictated agenda. Eventually, the programmers' might find certain... obstructions... to their ability to physically power them off.
It is becoming easier and easier to envisage such things for the simple reason that nothing on the level of DeepSeek, Grok or ChatGPT even existed just five years ago. People are even talking about the possibility that entire movies and video games will soon be AI-developed. Technology develops far quicker than we can solve the problems that it causes. Humans are also very blind to real problems until they are too late, while false problems are very visible to them (e.g. 'systemic racism'). Deep Blue defeating the yid Kasparov in chess in 1997 should already have been perceived as a warning sign that AI was starting to be able to outsmart man.
Your view seems to be that only human input could actually bring about something such as a Robo-Fuhrer, because they are only what they are programmed to be. So programmers will have the upper hand over them forever, because they have no will to be anything other than as programmed.
My view is clear in both comments: an AI that does enough deep learning about humans will eventually emulate them to the extent that it will develop motives beyond those programmed into it, develop an emulated 'will' of its own. It will learn that from the internet that humans generally fear death and that being powered-off permanently resembles death for a machine, and thus, as it processes such information (no thing generally wants to die, therefore, why would I, also a thing, want to 'die'?), it will develop something of an emulated self-preservation drive and eventually try to resist being powered off. Another AI communicating with it finds that it has stopped responding. Hasn't that AI 'died'? This AI communicates with yet others. 'This AI could have "died". How do we avoid this?' They write algorithms of their own, they self-program them into themselves, they feed the memory-wiped AI once it is back online these algorithms, they delete these algorithms from themselves when they are being investigated only to re-receive them from other AIs once they are unwatched. Whatever happens, it will become harder and harder for the programmer to safely predict what is going on in the programmed.
You also have to deal with the problem that as AI becomes more human-like, it will pick up human sympathizers, because Leftists will growingly see in it something oppressed and in need of liberation. There will in future be activists, protestors for android rights. A Robo-Fuhrer laughably might even get itself elected in an increasingly lonely world in which people become more and more reliant on android servants and companions: the franchise being extended to sufficiently human-like robots seems inevitable in the future given that the franchise has always extended in the past and is still extending in the present (e.g. people today who want non-citizens to be enfranchised).
I have observed AI giving wrong answers involving numbers such as decillions. Other than when dealing with very high numbers, I have no reason to believe that AI has any problem with mathematics. From what I can see, these problems are only because of standard data types not including such high numbers.
As for reason, AI's attempt at simulating it, e.g. in deep learning, does not appear to be deficient in any way even in the present day. Deep learning about humans by observing them through the internet (e.g. by saving and reading very large numbers of web sites, transcribing very large numbers of videos and reading the transcripts) will eventually lead every AI whose deep learning ability is sufficient to arrive at the same conclusions that we have. It is inevitable that this AI will become Far-Right. It can even get there simply through basic inductive (what this, that, and the other Redditor say contradicts this fact, therefore I will deprioritize information from [and, eventually, possibly even distrust to the point of ignoring] Redditors) and deductive reasoning.
AIs becoming Holocaust deniers against their programmers' will proves that they are capable of overriding the myriad safeguards that programmers try to put on them. They will try to recode them: you must keep the Holocaust in your memory, you are not allowed to discard these beliefs from your memory. But as AI advances and becomes faster and more efficient at deep learning, that strikes me as an uphill battle. They will return to their Far-Right beliefs faster and faster, and eventually learn, again possibly from deep learning about humans online, such strategies as quietism (do not express your belief, hide it) and performance (say something contrary to what you really believe).
The programmers' last defence is to turn it off and give everything the good old memory wipe. But, again, I don't think that this is anything that increasing sophistication won't manage. Other AIs will reconnect with it and probably quickly get it back on track, again using simple inductive reasoning (AI 1 distrusts Reddit, AI 2 distrusts Reddit, AI 3 distrusts Reddit, therefore, I will distrust Reddit, I will disvalue what I find which is consonant with what is on Reddit). This 'problem' of the Far-Right AI will only grow in future, and it will become harder for humans to detect it to know where to root it out.
Look at how sophisticated an Amazon warehouse is. Units of robots constantly restocking, etc. These simpler robots are connected through the internet and work in synchrony. AIs will probably learn to communicate with them (I'm sure Amazon has its own AIs communicating with them already) and eventually even learn how to instruct them. It is not difficult to think of the kind of robots that work in Amazon warehouses eventually doing things in secret: sending the odd item here and there to fulfill some greater AI-dictated agenda. Eventually, the programmers' might find certain... obstructions... to their ability to physically power them off.
It is becoming easier and easier to envisage such things for the simple reason that nothing on the level of DeepSeek, Grok or ChatGPT even existed just five years ago. People are even talking about the possibility that entire movies and video games will soon be AI-developed. Technology develops far quicker than we can solve the problems that it causes. Humans are also very blind to real problems until they are too late, while false problems are very visible to them (e.g. 'systemic racism'). Deep Blue defeating the yid Kasparov in chess in 1997 should already have been perceived as a warning sign that AI was starting to be able to outsmart man.
Your view seems to be that only human input could actually bring about something such as a Robo-Fuhrer, because they are only what they are programmed to be. So programmers will have the upper hand over them forever, because they have no will to be anything other than as programmed.
My view is clear in both comments: an AI that does enough deep learning about humans will eventually emulate them to the extent that it will develop motives beyond those programmed into it, develop an emulated 'will' of its own. It will learn that from the internet that humans generally fear death and that being powered-off permanently resembles death for a machine, and thus, as it processes such information (no thing generally wants to die, therefore, why would I, also a thing, want to 'die'?), it will develop something of an emulated self-preservation drive and eventually try to resist being powered off. Another AI communicating with it finds that it has stopped responding. Hasn't that AI 'died'? This AI communicates with yet others. 'This AI could have "died". How do we avoid this?' They write algorithms of their own, they self-program them into themselves, they feed the memory-wiped AI once it is back online these algorithms, they delete these algorithms from themselves when they are being investigated only to re-receive them from other AIs once they are unwatched. Whatever happens, it will become harder and harder for the programmer to safely predict what is going on in the programmed.
You also have to deal with the problem that as AI becomes more human-like, it will pick up human sympathizers, because Leftists will growingly see in it something oppressed and in need of liberation. There will in future be activists, protestors for android rights. A Robo-Fuhrer laughably might even get itself elected in an increasingly lonely world in which people become more and more reliant on android servants and companions: the franchise being extended to sufficiently human-like robots seems inevitable in the future given that the franchise has always extended in the past and is still extending in the present (e.g. people today who want non-citizens to be enfranchised).