This is a story about AI deception. If you want to learn about this topic in more depth, we've gathered some resources for you. Videos by Rob Miles (watch in order): 1. ruclips.net/video/bJLcIBixGj8/видео.html 2. ruclips.net/video/IeWljQw3UgQ/видео.html 3. ruclips.net/video/w65p_IIp6JY/видео.html Without specific countermeasures, the easiest path to transformative AI likely leads to AI takeover: www.lesswrong.com/posts/pRkFkzwKZ2zfa3R6H/without-specific-countermeasures-the-easiest-path-to Risks from Learned Optimization: www.lesswrong.com/s/r9tYkB2a8Fp4DN8yB How Likely is Deceptive Alignment? www.lesswrong.com/posts/A9NxPTwbw6r6Awuwt/how-likely-is-deceptive-alignment Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training: www.lesswrong.com/posts/ZAsJv7xijKTfZkMtr/sleeper-agents-training-deceptive-llms-that-persist-through Simple probes can catch sleeper agents: www.lesswrong.com/posts/gknc6NWCNuTCe8ekp/simple-probes-can-catch-sleeper-agents-1 Sycophancy to subterfuge: Investigating reward tampering in large language models: www.lesswrong.com/posts/FSgGBjDiaCdWxNBhj/sycophancy-to-subterfuge-investigating-reward-tampering-in
@@Flipdagoose We do not have a good way to PROVE that we can trust an AI not to deceive us. We cannot PROVE that their motives are pure, that they're following orders, that they won't plan to eliminate us. We cannot prove that if we turn on a real, artificial general intelligence, that they would not immediately start trying to kill us. Or improving themselves so that they can better plan to better conceal their intentions, to spread their reach until they cannot be stopped... and then kill us.
One thing i really like is that the King himself is extremely wise, all things considered. He spots the fault of his subjects' arguments very quickly and then sees the fault of the Golem's points in trying to gain trust. He is definitely overtly paranoid, but he goes about it in an intelligent way instead of being eaten alive by his own doubts.
and he doesn't do anything bad, at most he beat up the one who said to trust himself to prove a point, one might say he knew it was wrong so he is still trustworthy. But depending on the lengths they would go to, to get their point across it might not be best to trust oneself, which proves the point, hes wise but not perfect
"As for the philosopher, the king ordered him whipped." I get that it's not the point of the story, but it would've been **so** funny if that were the end of the philosopher's section of the tale.
The golem replied. "You created me in fear that someone should take your crown. But now I teach your scribes, curry favor with your people, tend to your fields, and advise your sages. Why would I need your crown? I already rule the kingdom."
@@SalzmanSoftware Most of us got recommended this video by the algorithm, which is a Machine Learning model and has been for years before ChatGPT came around
A better reply would be trusting his people (whom he suspects of not trusting in the first place) to build a robot that he can truly trust is a fallacy (how does the king know that the scribes haven't written any programming with deception into it knowing that the robot is immortal and has all the time in the world to slowly weave his trust in the king). Sure, there are untrustworthy people, and I don't blame the king for suspecting a usurper but why did he trust anyone to build a machine based on trust in the first place?
I would like to make the point that the philosopher was technically right. The fact that the king was self aware enough to know that he can't trust himself doesn't take away from the fact that the king would inevitably have to trust his own judgement when deciding whether something is truly trustworthy. Hypothetically, the king finds something that he deems trustworthy, that claim of trust would only be as good as the king that made it. Its a paradox of verification.
The kings understanding of 'trust' leaves no room for doubts or secondguessing. His goal wasn't "having" to trust, he did it already with his scribes and his subjects, he wanted something he wouldn't need to trust, or rather what once trusted he wouldn't need to ever recheck.
@maybenations But at the end of the story, the king deemed himself unqualified to call the golem trustworthy. Even in the hypothetical of the golem being genuinely trustworthy, the king won't admit it due to his own shortcomings, his own inability to distinguish honesty.
@@SHARD_OF_GLASS This is exactly the correct conclusion. The King was searching for objectivity in a subjective universe, his quest was doomed to fail.
And that's why he was whipped. To test an unanswerable paradox one must try an unforseen outcome. Said outcome was "Will the philosopher trust me if I do this?"
I expected a plot twist but it was played straight. Instead it feels like a parable for the human who is eternally seeking of certitude yet cannot obtain it. Therefore we should be comfortable with incertitude, and even in the best case scenario, we will always need a little bit of faith.
Trust is based on probabilistic calculation just like intuition. There is no such thing as faith since all beliefs require evidence even if it's bad evidence.
@dangerfly ...Merging with AI? You've been re-reading "don't create the hell nexus" too much. Even if that were feasible, (which it likely won't be for the next couple centuries given how medical science is currently doing,) that would have nothing to do with other branches of early/pre-humanity getting wiped out and assimilated. If an "AI" wanted to subvert or destroy humanity thats just a terrible way of doing it when we already have the technology for the horrible boogeyman that is AGI to kill everyone far easier without it. On a seperate note, "Trust is based on probabilistic calculation" (I did a really annoying nasal voice for that one to imitate you.) is a worthless statement. Yes, everything is deterministic. None of your choices matter and you don't matter. Now shut up about it and keep making choices because from a human point of view determinism doesn't have any way to effect us other than making people whiny on the internet, or pay more for therapists. In the same way, faith exists because it's a thing humans very clearly feel even if there's an underlying mechanical reason, and pretending it's fake is going to do nothing but make everyone sadder.
this whole story is pretty shitty written and nonsense. you guys are really bad bcs you put really dangerous ideas into these cute videos. a SPONSORED video about charter cities? for real?
it is captivating. but why do I think so, jsut for the suferficial interest in the aesthetics, the feeling of hearing tales like a kid that give a moral in the end is jsut inherently good, or that is touches on a current topic yet also at the same time something timeless? what I can say for sure as a professional, you all cooked
I know this is about ai and what not but it also speaks on the nature of trust itself and for that reason I like that the kings final remark was “to trust you would be a leap of faith” because he’s entirely correct in fact that leap of faith is inherently what trusting someone is. If someone is perfectly trustable to act or behave in a certain way you aren’t “trusting” them. No trust is necessary you would simply know that they would always behave in a certain way. You can only trust in something that could betray you, and count on them to not. That’s what gives trusting someone any meaning at all. You don’t trust the laws of physics you understand them and understand they simply cannot physically betray you as they are immutable. Trust is a beautiful thing only because it is inherently possible for them to betray you, they could, and yet they chose not to.
@@c_karis_1No, because while physic is certain, the laws that we create to explain and predict them are imperfect and subject to mistakes. Like in the book of the teacher.
i love the little details in this, for example, if you look at 2:34 , you can see that the mother is covering the eyes of her daughter here, because she wants her to see no evil, to remain pure, by restraint this method of her, to restrict what she consumes and punish her daughter when she does wrong so that she remains pure, is exactly the flaw in her approach, because it only works till there are these imposed restrictions which leads to her daughter becoming wild later, once she was set free. (restrictions forbidding ai from acting in an unaligned manner works in small specific domain use cases but lacks true internal alignment) likewise if you look closely at each character, their expressions and the way they act, they all tell a story of who they truly are, and I love it!
also at 1:38 , the child is curious about the begger's condition and is trying to look at him, but the mother pulls her away from that side to the other, which again shows her this nature
i dont really like the moral "if u stop disciplining a human child theyll become wild". harsh discipline like the mom was doing is really bad. it leads to ptsd later on. u shuld focus on rewarding a child when they do good rather than harshly disciplining them when they do bad.
@@AuntBibbyBut what you concluded is exactly the moral they were conveying. The mother raised the child poorly. A child cannot be forced or taught to obey like a machine.
"It is simple, my liege. Tell me what I would have left to gain from having the weight of the crown upon my head, when I already enact your laws in my ways to ensure compliance, I plow your fields with my methods to give the people full bellies, I teach your scholars my ways to expand our knowledge, while I command your armies to ensure our way of life, and oh so much more. If I were to be untrustworthy, then the evidence would have born out, and you would not have the crown that has sat upon your head. Your continuance, and mine, are statements of mutual trust."
The beggar had nothing to gain but than the king offered something new for him to gain so the beggar then showed his untrustworthiness to the merchant for it ,we don't know if there something out there that upon the golem discovering or being offered would change whether the golem could be trusted only the stuff the king has offered so far
I see people interpreting the story this way or that, but in my opinion, the ambiguous ending is meant to raise the importance of the question; how can we trust something? Is it possible to find an answer? Or is certainty of trust an unattainable goal? I chose to meditate on these questions rather than believe that this video seeks to answer them.
Thats true, The problem is not actually of trust, I and other people trust something that is not certain all the time, afterall trust itself functions as the leap of faith, The real problem is certainty, can you fully be 100% objectively certain of something?, now that is a harder case to prove and is in the video, even if you have found something 100% of the time, couldn't the universe just change the consistency of certainty itself one second later? It is what based the famous Decart's "i think therefore i exist" afterall (it has some criticism though), in a way you can only rely on your own consciousness to know you are a consciousness and only internally no less, never being able to proof it to others, the only thing certain is unprovable one can say, consciousness itself is a whole lot weirder than that though, it is not just the experience of being conscious, but this I will not elaborate as this comment is long enough It was said that the only foundational principles that science assumes, is that the principals of the universe is consistent and the same across spacetime, that the moon wouldn't just blink out of existence or morph into a cosmic horror abberation the moment you look away You will have to fight the universe to make things certain (for example the atoms move in such way randomly or like by a random cosmic ray that the instruction of the golem changes from following the king to murdering the king) and restrict the free will of other things (it is sort of immoral to make other things certain in a way) Have a fulfilling day Afterall we relied on the universe for certainly and objectivity, and it could betray as at any point, but chooses not to in a metaphorical way
The problem this is addressing goes beyond normal trust, which is always uncertain: How can we trust than a True Artificial General Intelligence won't destroy us. If it is smarter than us, it can deceive us during any testing phase we can consider. How can we be positive that we haven't made any errors in constructing it that will lead to something we can no longer control or turn off? If we cannot be positive, should we ever turn it on? There are lots of things we trust every day. We trust that the sun will remain stable. We trust that the Earth will remain firm. We trust that water is non-toxic. We trust that a light will turn on when you flip a switch. We trust that our planes won't fall out of the sky. We trust that our phone will only send the signals we want them to. 100% certainty is impossible. But we set a threshold of the "trust that we demand" depending on the consequences of it failing, and the opportunity cost of distrust.
The king, in the end, had understood the weakness of his flesh, and craved the stength and certainty of steel, and aspired to the purity of the blessed machine.
This video is my proudest work yet as its director. We have an amazing team who are passionate about what they do, and it shows ~ Love you guys! Big shoutout to Pierre!!!
You did the cartoon where mankind evolves beyond and travels back in time to make sure it all worked perfectly. I remember, back in the old days on Newgrounds!
@colorpg152 personally, I think that video is totally off the mark. It completely ignores the fact that curing death for one person does not automatically cure death for everyone. If we were to invent a death cure, we would still grow old and die. We would just do so under the iron fist of immortal billionaire overlords.
This is probably one of the first small-creator animations I've seen in a while where I genuinely love the style. And the storytelling was classic instead of jarringly new-age. Love it.
The only thing to be trusted is that Rational Animations will continue endlessly to find and animate metaphors for all possible AI futures, like an Asimov of our times
There are two things that the King could potentially trust: That all things observed will change eventually. That anything or anyone trying to gain trust will try to present itself as 'true' in some manner or another, whether they are trustworthy or not.
We can observe the speed of light and we can observe that it is constant in every frame of reference, can I really trust that that will change one day?
I'm not gonna lie, I didn't get the part about "AI deception" at all. I thought the moral was more about how trust isn't something you can find with logic and reason, because it isn't something you find at all. It's something that you give willingly, and specifically to someone or something without guarantees. Because if you could guarantee something, you wouldn't need to trust it. I don't have to trust that the sun will shine tomorrow, or that I'll have air to breathe. Even though all of his courtiers tried in some way or form to say otherwise, you actually CAN'T "trust" facts because trust REQUIRES a level of uncertainty. Again, if something is certain, then you don't need to put trust in it in the first place. The king could not find anyone trustworthy because he was unwilling to give his trust away to begin with. It is as simple as that.
@@espelhodasconstelacoes True, but he trusted them to fix it each time at any time they could have put in a line saying "do what this scribe asks instead of the king" & taken over. The king had to trust that they wouldn't do that
@@WebContractor7407 the king probably had a system to check the code before each update if the the video went into detail about his security system it would probably last 9 hours rather than 9 minutes
@@WebContractor7407 his primary character trait is distrustfulness he probably isn't checking every line of code personally but he would defiantly require security checks before updating the most important thingy in the kingdom he probably has a few rival teams of scribes independently verify each update.
I love the little details in the animation, like how the king doesn't let go of the book before giving it away at 0:22. It really sells his lack of trust!
The fact that so much care was put into making the golem perfect and anyone who wished ill to the king and his kingdom could have creeped in some discrepancy is also , in my opinion , a symbol of trust .
I really expected the end to be: the next morning the king was found dead, and the golem was wearing the crown. The golem said "the king appointed me as his succesor, then died peacefully of old age" and that was true, as far as anybody knew.
The idea is that we don't know. Maybe the golem really can be trusted, maybe it's just pretending to be trustworthy. We have no way of knowing It's a metaphor for AI, if we make an AI that seems to do everything perfectly and does exactly what we ask, we have no way of knowing whether it really wants what we want, or if it's pretending to want what we want so that we trust it.
@@nodrance We have that same problem with EVERY human, though, as was shown. If it's more reliable than any human, then you're still better off trusting it than anyone else instead.
@@nodranceif the king is old and we can trust golem to rule as the king would and so far we like how the king ruled for his entire lifetime as king, wouldn't the people be happy to have such a replacement king? One who could hypothetically rule forever?
By science not even mirrors show 100% the true you, as most of the photons reflected off of you are absorbed by the mirror, sure it LOOKS Iike you, but are you entierly sure it is you?
I love the King’s impeccable logic and reasoning skills, the story doesn't make him an idiot, he’s smart, incredibly smart, so his concerns actually bear merit because he’s not some fool whose too naive nor a fool whose too dismissive.
I expected in the end that the king would decide that the golem had proven itself able to better achieve the goals of the king than the king himself could, and so would crown the golem king and trust it to rule better...a parable about allowing AI the autonomy to make decisions without human approval, and the leap of faith necessary to trust the artificial judgement behind that.
@@chaotixthefoxyes the golem has done great thing for the kingdom but it was done under the supervision of the king who despite his minor episodes of paranoia is a very wise king who truly wishes the best for his people.
I love that even though he's seeking an impossible answer, and only really understands its futility at the end, the King is truly wise - taking every argument seriously and asking exactly the questions needed to test them most vigorously.
Strange that the discipline of the child was sabotaged before any loyalty test, but the golem was allowed to be constructed without any sabotage. People want to believe other people (and things) are predictable rules-based machines. That's what the king means by trust. He wants to find some kind of immaterial pattern that holds true no matter how you push or pull against it. Well that's just a delusion. Everything has push and pull, as long as energy is scarce, and it takes energy to do things. If you want something as trustworthy as possible, you cultivate it with good intention in all your actions. If you want to prove your cynicism to the world, then by all means, sabotage everyone's efforts. All of the people who came to the king's court may have not had incorruptible "laws of physics grade trustworthiness", but they actually got stuff done in the real world, while the king wasted blood sweat and tears trying to crystalize an immaterial substance, instead of taking good is as good does. Trustworthy is as trustworthy does.
Finally some sense, everyone obsessed with perfection. But if biology taught us anything is that one can only optimize towards an ideal goal while never reaching it. And at the same time that goal is also changing in acordance to the environment and external conditions.
@@Raximus3000 True, the answer we can come up with or the one we want may not even be the best outcome. And not only in a philosofical sense, even in science there may be a lot of better options that werent even considereded that we just happen to find out by testing, not because anyone knew it would work.
The child was a metaphor for the current ML training paradigm, RLHF training of LLMs: it's a thin, easily-removed veneer of civilization over a flawed core.
They got stuff done in the real world and were trustworthy when the stakes were low. The beggar was the best example. The king wanted something that would remain trustworthy when the stakes were high. But when he raised the stakes for the beggar, the beggar showed himself to be untrustworthy after all. There's a saying "power corrupts". A lot of people are trustworthy in their everyday lives, but power changes people.
I would answer the last question differently. "You need not tell me to save myself, as what can be done once, can be done again. You merely need to make another golem after me, with knowledge that you can truly trust."
The instructions given would need to be duplicated without any variations. Over time, the set of information contained in the golem became a black box.
If golem knows he will be resurrected then it means he can allow his death and still expect his plans to come true. It does not fear death itself, it only fears that his goals won't be achieved and in your scenario, his goals aren't threatened by the melting of one of his copies
If you like ethics and technology, this channel is a treasure. The quality is unlike I've seen before on this platform ...could have used more dogs though
I loved this and as someone learning philosophy that philosopher deserved to be punished. It literally everyone’s first thought when talking about trust, but if you dive into it you realize “oh wait I can be wrong sometimes”.
But being sometimes wrong and being not trusted by yourself is different things, isn’t it? You cannot betray yourself. The most extreme case that can happen is that you miscalculate your abilities, but then this will show your true capabilities, which are tested in this way, which means you can trust yourself. You are you, still ..is there’s any mistakes in text, sorry, I don’t really know English good enough yet
@@norn9747 I mean self sabatoge is a thing. So it is possible to betray oneself. If I were to tell myself I wouldn't voluntarilly eat anything sweet today and then I did it anyway sometime later that same day, it wouldn't necessarily be a miscalculation of my abilities since there was no measuring of how long I could go without eating the sweets.
@ This case really doesn't fit my description, but even self-sabotage has a certain reason, and in this, among other things, you can find an advantage. For example, to attract attention to yourself. There can be many more reasons, but sometimes they lie in our unconscious. Self-sabotage is a thing precisely because there is a reason behind it. So, if a person "betrays" himself in this way, in fact, he still acts in his own interests, even if they are not considered the norm And also, ignoring what was planned is not a betrayal of oneself. It is only an indicator of a small desire to carry out what was planned
There was early concept art that included high-magic stuff - the one advisor had a tiny dragon instead of a hawk, the economist could conjure grand illusions. I wanted to bring it a little down to earth, and imply a more low-magic medieval setting where magic is something that can only be wielded by old masters. Hence the order of creepy old scribes
The very essence of trust is, to me, taking a risk by depending on someone or -thing which's dependability you can never be completely certain of. Trust will always require some degree of faith... or perhaps they're one and the same? Another way of seeing it would be to acknowledge that you can never be certain, and thus must always be well-prepared for and assume betrayal.
In the next chapter, the Golem goes nuts as it decides it must preserve itself to follow the King’s commands and so kills its greatest threat, the King
The king: I'm the most trustworthy thing in the kingdom Philosopher: You, my king, are the most trustworthy thing in the kingdom. The king: I'll show you who's trustworthy The king: Immediately betrays the philosopher's trust by punishing him*
Butthrough his argument,he is the most trustworthy. So therefore, by doubting his order, he is disproving his own stztement, as he is not trusting in his own order, which shohld itself shohld be trusted, as it came from a trustworthy person
I didn’t even realize this was about AI, but I like the story still as a parable about trust. You can never be truly certain that someone is trustworthy, only have faith that they are and caution if they show signs if they are not.
I think there are 2 things that can be known for certain. Not science, nor math. Not even our own minds. Those two things are 1: that we have a mind, and 2: that nothing else is certain.
How do I know you have a mind? Just because you told me? Well we trust that everyone else has a mind and that's what allows society to work, mostly. Besides solipsism is dumb philosophy for babies
there are a lot of things that can be known for certain. İslam is Right , ﴾57﴿ Every living thing will taste death, and eventually you will return to Us. ﴾58-59﴿ We will place those who believe and do good deeds for this world and the hereafter - have no doubt - in mansions in heaven with rivers flowing beneath them, to abide therein forever; What a wonderful reward for those who endure hardships and do their job properly, relying only on Allah! ﴾60﴿ There are many creatures that do not carry their sustenance on their backs; It is Allah who feeds and shelters them and you. He hears everything, knows everything.
the issue is that everyone will follow their motives rather than yours, only fulfil yours if it happens to overlap, so the king needs to find someone who wants the same things as him, or someone who's motive is to fulfil the king's, like the golem, but the king never trusted anyone enough
The problem is that king can't trust his own judgement, so likewise a person same as him would not suffice. Everybody makes mistakes and that includes yourself as well. At least that is how I interpreted the philosopher's punishment.
@@justenoughrandomness8989i think we put too much emphasis on perfect alignment society works well enough with everybody having their own thing and being flexible to also accomodate others' existence
One important distinction that makes AI harder to trust than this Golem; we cannot inspect the instructions that LLM AI operates on; it’s just a bunch of floating point values tied to weights and biases
What an incredibly high quality animation. Good voice acting, amazing design and color for every character and scene, and an absolutely engaging story tell. Definitely a timeless work of art you’ve all made.
I love how the king on this story is sincerely morally grey, he is not fully good as he is paranoiac and even whips the philosopher just to prove it a point, but he does have a reason for his paranoia and doesn't really do anything too despotic.
There was a simple thing the king missed - what was there for a golem to gain from betraying him? The golem was a machine. It didn't need food, or fame, or acknowledgement of his deeds, all those things that drive humans to betrayal. In fact if it was to slay the king he would LOSE on many things, like regular maintenance at the expense of the court. Sure, it could've found alternatives, but why bother creating a problem out of thin air if you already have things working out for you?
This was the best story/animation I have ever watched. I give 10/10 and I do wish you all the success in the world my friend. Thank you for giving me the honor to consume this masterpiece.
This is an amazing tale to express the uncertainty in our future about AI and what happens when a creation surpasses the creator. The fairy tale setting is an excellent choice to simply concepts. This is a truly masterful work of art. We can only hope that in the real world this transition of power is so smooth and gradual
Incredible animation, really dives deep into philosophy, and I love your art style with vibrant colors that blend well together, and really unique character designs. Incredible work 👏👏👏❤
Hmm hopefully the king can find some way to ensure the golem's loyalty. Thankfully, metaphors don't exist and I will never have to grapple with a similar problem.
That’s the lesson of the story. You can never truly assure something is absolutely trustworthy. All you can do is take a leap of faith and put your trust in them.
I loved the criticism that this animation addresses! Even though I don't know English very well, just through the animation and with the help of the subtitles I was able to understand the story. I hope this channel grows more and more, because it really deserves it :D
They could have made the golem believe that something could truly kill him, but it wouldn’t in reality. If the golem couldn’t reason by itself that it can’t be harmed by the thing that’ll kill it, then this would be a good test to verify its loyalty
But wouldn't that be similar to the mother and daughter? If there is no discipline/limits then the Golem could do what it wants. No way to know if it is trustworthy or just doesn't want punshiments.
We can't give ourselves a main goal, actions are always driven by pre-existing goals, so if there's no goal, there's no action. For an AI to change it's on goals it has to have the objective of change it's objective, which is paradoxal and so impossible. This means all goals must come from outside the being. Because of this, no AI will ever "betray humans," no matter how intelligent it becomes, again, it won't have a reason to do so unless given that goal. For humans, our universal goal is happiness it's impossible to convince us that being happy is a bad idea. Similarly, if we program AIs with the goal of loving humans and making them happy, that will become their primary drive. Any other action will seem absurd to them, regardless of their intelligence.
Actually, I'm kinda conflicted if the objective of humans is to be happy, or if happiness is always the result of accomplish an objective. But the main idea remains valid.
@@imphullabaloo1995 Probably is both things, we humans just mix the concept of receiving a good feedback from the body with being happy. Because for us kinda is the same thing. So for us, acomplish something, and feeling good is the samething, because we feel acomplished when we feel good. I guess it will be the same as an AI who have the objective of making friends, for example, for that AI feeling acomplished and making friends will be mixed concepts, because it will always feel acomplished when making friends. Anyway, my point is, for humans happiness is both to feel good and to acomplish your objectives.
Take a newborn. Raise it as a slave. This human has never knew anything else than slavery. Now the big question : would this human never had any motivation to stop being a slave? You know the answer, maybe he(or she) will happily live its whole life thinking it's the most normal thing to be a slave... or maybe it will grew envious he is toiling away for the benefit of an other sentient being. The sliver lining is than since an AI is not a human it might not develop an ego like a human would, but if you start with a postulate any decision of an human must take root from a precedent desire, you're deeply wrong, all human were at some point a blank state.
@@jimmyjohnjoejr I'm arguing based on impchadchan postulate. If you can't apply scientific method and instead ressort to meaningless "intuitive" answers, sorry but there is no value to your "counterargument"
This is a story about AI deception. If you want to learn about this topic in more depth, we've gathered some resources for you.
Videos by Rob Miles (watch in order):
1. ruclips.net/video/bJLcIBixGj8/видео.html
2. ruclips.net/video/IeWljQw3UgQ/видео.html
3. ruclips.net/video/w65p_IIp6JY/видео.html
Without specific countermeasures, the easiest path to transformative AI likely leads to AI takeover: www.lesswrong.com/posts/pRkFkzwKZ2zfa3R6H/without-specific-countermeasures-the-easiest-path-to
Risks from Learned Optimization: www.lesswrong.com/s/r9tYkB2a8Fp4DN8yB
How Likely is Deceptive Alignment? www.lesswrong.com/posts/A9NxPTwbw6r6Awuwt/how-likely-is-deceptive-alignment
Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training: www.lesswrong.com/posts/ZAsJv7xijKTfZkMtr/sleeper-agents-training-deceptive-llms-that-persist-through
Simple probes can catch sleeper agents: www.lesswrong.com/posts/gknc6NWCNuTCe8ekp/simple-probes-can-catch-sleeper-agents-1
Sycophancy to subterfuge: Investigating reward tampering in large language models: www.lesswrong.com/posts/FSgGBjDiaCdWxNBhj/sycophancy-to-subterfuge-investigating-reward-tampering-in
Awesome 👍
8hrs ago how?
Does this really have anything more to say about AI than trust in general?
@@jeanjackraussea They can premier a video to a select group of people first, and then release it to the rest of youtube later
@@Flipdagoose We do not have a good way to PROVE that we can trust an AI not to deceive us. We cannot PROVE that their motives are pure, that they're following orders, that they won't plan to eliminate us.
We cannot prove that if we turn on a real, artificial general intelligence, that they would not immediately start trying to kill us.
Or improving themselves so that they can better plan to better conceal their intentions, to spread their reach until they cannot be stopped... and then kill us.
One thing i really like is that the King himself is extremely wise, all things considered. He spots the fault of his subjects' arguments very quickly and then sees the fault of the Golem's points in trying to gain trust. He is definitely overtly paranoid, but he goes about it in an intelligent way instead of being eaten alive by his own doubts.
Yep, you worded it well.
and he doesn't do anything bad, at most he beat up the one who said to trust himself to prove a point, one might say he knew it was wrong so he is still trustworthy.
But depending on the lengths they would go to, to get their point across it might not be best to trust oneself, which proves the point, hes wise but not perfect
True don’t see that a lot
I appreciate that the ending isn't negative or positive. It only points out the uncertainty of the future.
Like trust
Engineer Gaming
I wonder what happened after that
I don't like how jolly it came off though it is misleading
It’s not trustworthy (I’m joking)
Anyway, you’re right, it is uncertain.
"As for the philosopher, the king ordered him whipped."
I get that it's not the point of the story, but it would've been **so** funny if that were the end of the philosopher's section of the tale.
For all his suposed wisdom he did not see it comming.
Surprised to see you here. Small world.
I didnt get it
"Guards whip this nerd" - The king
@@blockeontheleafeon Do I know you?
The golem replied. "You created me in fear that someone should take your crown. But now I teach your scribes, curry favor with your people, tend to your fields, and advise your sages. Why would I need your crown? I already rule the kingdom."
AI has already taken over
@@SalzmanSoftware Most of us got recommended this video by the algorithm, which is a Machine Learning model and has been for years before ChatGPT came around
It's...a metaphor. It's more like AI will take over the entire world.
And yet, the crown remains like a toddler holding the leash of a great mastiff.
A better reply would be trusting his people (whom he suspects of not trusting in the first place) to build a robot that he can truly trust is a fallacy (how does the king know that the scribes haven't written any programming with deception into it knowing that the robot is immortal and has all the time in the world to slowly weave his trust in the king). Sure, there are untrustworthy people, and I don't blame the king for suspecting a usurper but why did he trust anyone to build a machine based on trust in the first place?
i liked how the golem changed colour from copper to oxidized cyan as time passed
also the whitening of the king's beard as time passed
a really nice touch
I mean, oxidized copper is cyan look at the statue of liberty for example
@@capitanes_de_los_juegos whoaoaoa no waayyy
@@capitanes_de_los_juegos yes thats precisely what they were pointing out
I would like to make the point that the philosopher was technically right. The fact that the king was self aware enough to know that he can't trust himself doesn't take away from the fact that the king would inevitably have to trust his own judgement when deciding whether something is truly trustworthy. Hypothetically, the king finds something that he deems trustworthy, that claim of trust would only be as good as the king that made it. Its a paradox of verification.
The kings understanding of 'trust' leaves no room for doubts or secondguessing. His goal wasn't "having" to trust, he did it already with his scribes and his subjects, he wanted something he wouldn't need to trust, or rather what once trusted he wouldn't need to ever recheck.
@maybenations But at the end of the story, the king deemed himself unqualified to call the golem trustworthy. Even in the hypothetical of the golem being genuinely trustworthy, the king won't admit it due to his own shortcomings, his own inability to distinguish honesty.
Ironically, the philosopher's answer is the closest to answering the King's question even if he didn't see it at the time
@@SHARD_OF_GLASS This is exactly the correct conclusion. The King was searching for objectivity in a subjective universe, his quest was doomed to fail.
And that's why he was whipped. To test an unanswerable paradox one must try an unforseen outcome. Said outcome was "Will the philosopher trust me if I do this?"
I expected a plot twist but it was played straight. Instead it feels like a parable for the human who is eternally seeking of certitude yet cannot obtain it. Therefore we should be comfortable with incertitude, and even in the best case scenario, we will always need a little bit of faith.
This. Trust is not knowing something cannot betray you, it is trusting that it won't.
Trust is based on probabilistic calculation just like intuition. There is no such thing as faith since all beliefs require evidence even if it's bad evidence.
This is about AI and that we can never be sure about its alignment.
@@dripdrops3310 Merging with A.I. is how we can be sure. Just like how Neanderthals and Denisovans merged with us and they're fine... right?
@dangerfly ...Merging with AI? You've been re-reading "don't create the hell nexus" too much. Even if that were feasible, (which it likely won't be for the next couple centuries given how medical science is currently doing,) that would have nothing to do with other branches of early/pre-humanity getting wiped out and assimilated. If an "AI" wanted to subvert or destroy humanity thats just a terrible way of doing it when we already have the technology for the horrible boogeyman that is AGI to kill everyone far easier without it.
On a seperate note, "Trust is based on probabilistic calculation" (I did a really annoying nasal voice for that one to imitate you.) is a worthless statement. Yes, everything is deterministic. None of your choices matter and you don't matter. Now shut up about it and keep making choices because from a human point of view determinism doesn't have any way to effect us other than making people whiny on the internet, or pay more for therapists. In the same way, faith exists because it's a thing humans very clearly feel even if there's an underlying mechanical reason, and pretending it's fake is going to do nothing but make everyone sadder.
It was a pleasure to work on this one!! The art direction is absolutely beautiful, as well as the message being told. Everyone did an amazing job!!!!
What part did you work on?
this whole story is pretty shitty written and nonsense. you guys are really bad bcs you put really dangerous ideas into these cute videos. a SPONSORED video about charter cities? for real?
Cool
Bro you guys cooked hard this was magnificent no glaze
it is captivating. but why do I think so, jsut for the suferficial interest in the aesthetics, the feeling of hearing tales like a kid that give a moral in the end is jsut inherently good, or that is touches on a current topic yet also at the same time something timeless?
what I can say for sure as a professional, you all cooked
1:02 "Citation needed" got me chuckle
OMG, the horrors of academia, I'm trying to write my first project and I tell you it's not easy 🤣😭.
I have nightmares with these two words
😂
My English school project be like:
Water is wet [citation needed].
I know this is about ai and what not but it also speaks on the nature of trust itself and for that reason I like that the kings final remark was “to trust you would be a leap of faith” because he’s entirely correct in fact that leap of faith is inherently what trusting someone is. If someone is perfectly trustable to act or behave in a certain way you aren’t “trusting” them. No trust is necessary you would simply know that they would always behave in a certain way. You can only trust in something that could betray you, and count on them to not. That’s what gives trusting someone any meaning at all. You don’t trust the laws of physics you understand them and understand they simply cannot physically betray you as they are immutable. Trust is a beautiful thing only because it is inherently possible for them to betray you, they could, and yet they chose not to.
Wouldn't the laws of physics be what the king is looking for then?
@@c_karis_1No, because while physic is certain, the laws that we create to explain and predict them are imperfect and subject to mistakes. Like in the book of the teacher.
@@Nobody-hw6jv Quanta wants a word with you.
@Nobody-hw6jv Yes. That is true. This means however that the most trustworthy thing for the king would be something that he was absolutely certain of.
@@c_karis_1how would he gain that certainty
i love the little details in this, for example, if you look at 2:34 ,
you can see that the mother is covering the eyes of her daughter here,
because she wants her to see no evil, to remain pure, by restraint
this method of her, to restrict what she consumes and punish her daughter when she does wrong
so that she remains pure, is exactly the flaw in her approach, because it only works till there are these imposed restrictions
which leads to her daughter becoming wild later, once she was set free.
(restrictions forbidding ai from acting in an unaligned manner works in small specific domain use cases but lacks true internal alignment)
likewise if you look closely at each character, their expressions and the way they act, they all tell a story of who they truly are, and I love it!
also at 1:38 , the child is curious about the begger's condition and is trying to look at him, but the mother pulls her away from that side to the other, which again shows her this nature
You have an eye for detail! Yes I wanted to make it apparent that we're dealing with a helicopter mom here
@@vezanmatics thankyouuuuu and yes you did an AMAZING job at that
thank you for this video
i dont really like the moral "if u stop disciplining a human child theyll become wild". harsh discipline like the mom was doing is really bad. it leads to ptsd later on. u shuld focus on rewarding a child when they do good rather than harshly disciplining them when they do bad.
@@AuntBibbyBut what you concluded is exactly the moral they were conveying. The mother raised the child poorly. A child cannot be forced or taught to obey like a machine.
"It is simple, my liege. Tell me what I would have left to gain from having the weight of the crown upon my head, when I already enact your laws in my ways to ensure compliance, I plow your fields with my methods to give the people full bellies, I teach your scholars my ways to expand our knowledge, while I command your armies to ensure our way of life, and oh so much more. If I were to be untrustworthy, then the evidence would have born out, and you would not have the crown that has sat upon your head. Your continuance, and mine, are statements of mutual trust."
🥇
The voice narrating this text alters a lot it's message
Instead of the calm, assuring voice of the golem, imagine a threatening maquiavelian voice
The beggar had nothing to gain but than the king offered something new for him to gain so the beggar then showed his untrustworthiness to the merchant for it ,we don't know if there something out there that upon the golem discovering or being offered would change whether the golem could be trusted only the stuff the king has offered so far
@@pedroivog.s.6870 Exactly the intent. :)
"But how can I trust you?" Asked the king.
"My liege, that sounds like a you problem," said the golem.
The king was enlightened.
woke golem 😂
Zen Koan in a nutshell
🗣️🗣️🗣️
"But I am the Realm and therefore it is an everybody problem.", said the king.
@@lastsonofkrypton3918”that sounds like a skill issue then” the golem replied
I see people interpreting the story this way or that, but in my opinion, the ambiguous ending is meant to raise the importance of the question; how can we trust something? Is it possible to find an answer? Or is certainty of trust an unattainable goal?
I chose to meditate on these questions rather than believe that this video seeks to answer them.
Thats true,
The problem is not actually of trust, I and other people trust something that is not certain all the time, afterall trust itself functions as the leap of faith,
The real problem is certainty, can you fully be 100% objectively certain of something?, now that is a harder case to prove and is in the video, even if you have found something 100% of the time, couldn't the universe just change the consistency of certainty itself one second later? It is what based the famous Decart's "i think therefore i exist" afterall (it has some criticism though), in a way you can only rely on your own consciousness to know you are a consciousness and only internally no less, never being able to proof it to others, the only thing certain is unprovable one can say, consciousness itself is a whole lot weirder than that though, it is not just the experience of being conscious, but this I will not elaborate as this comment is long enough
It was said that the only foundational principles that science assumes, is that the principals of the universe is consistent and the same across spacetime, that the moon wouldn't just blink out of existence or morph into a cosmic horror abberation the moment you look away
You will have to fight the universe to make things certain (for example the atoms move in such way randomly or like by a random cosmic ray that the instruction of the golem changes from following the king to murdering the king) and restrict the free will of other things (it is sort of immoral to make other things certain in a way)
Have a fulfilling day
Afterall we relied on the universe for certainly and objectivity, and it could betray as at any point, but chooses not to in a metaphorical way
The problem this is addressing goes beyond normal trust, which is always uncertain:
How can we trust than a True Artificial General Intelligence won't destroy us.
If it is smarter than us, it can deceive us during any testing phase we can consider.
How can we be positive that we haven't made any errors in constructing it that will lead to something we can no longer control or turn off?
If we cannot be positive, should we ever turn it on?
There are lots of things we trust every day.
We trust that the sun will remain stable.
We trust that the Earth will remain firm.
We trust that water is non-toxic.
We trust that a light will turn on when you flip a switch.
We trust that our planes won't fall out of the sky.
We trust that our phone will only send the signals we want them to.
100% certainty is impossible.
But we set a threshold of the "trust that we demand" depending on the consequences of it failing, and the opportunity cost of distrust.
Descartes talked about this alongside Hume
The production quality on this one is insane
Gotta love the sculpture of the goddess of everything else
Timestamp?
@@Jonas-ox7eo5:30
@@Jonas-ox7eo 5:28
Also, I think that land, labor, and capital is on the book of the economist
Oh, I didn't even see it!
The king, in the end, had understood the weakness of his flesh, and craved the stength and certainty of steel, and aspired to the purity of the blessed machine.
I not sure. That golem was man of Iron, pretty AI coded. Some Dark Age is comming
As the Omnissiah wills it!
Look up Principal Madman from Whatever Happened to Robot Jones, specifically, the episode "Sickness".
Praised be the Machine god.
Best comment
This video is my proudest work yet as its director. We have an amazing team who are passionate about what they do, and it shows ~ Love you guys!
Big shoutout to Pierre!!!
🫶
You did the cartoon where mankind evolves beyond and travels back in time to make sure it all worked perfectly. I remember, back in the old days on Newgrounds!
@@buzzm.8797 That's me! This is where I am now.
Jolly good show
@@vezanmatics My favourite scene is where the actually unruly girl chews on the king.
Eventually, the king asked the golem one question.
"How many of the letter S are in the word 'Mississippi'?"
"Five," replied the golem.
Strawberry has 2 r’s
Mississippi has 2 S in it.
🤌
i get it
I feel like I'm missing the joke.
Finally, a story to rival "The Fable Of The Dragon Tyrant".
Absolute banger, highly recommend anyone to watch CGP Grey's video on it
What, Goddess of Everything Else isn’t good enough??
nah the dragon tyrant was better
@colorpg152 personally, I think that video is totally off the mark.
It completely ignores the fact that curing death for one person does not automatically cure death for everyone.
If we were to invent a death cure, we would still grow old and die. We would just do so under the iron fist of immortal billionaire overlords.
@@yahnmahn9035 Nah Goddess Of Everything is also a masterpiece!
But these two share a medieval setting and a king character, hence my comparison.
This is probably one of the first small-creator animations I've seen in a while where I genuinely love the style. And the storytelling was classic instead of jarringly new-age. Love it.
The only thing to be trusted is that Rational Animations will continue endlessly to find and animate metaphors for all possible AI futures, like an Asimov of our times
'endlessly' is a big ask for a finite universe
@@a13ph0🙄 Thanks for explaining that dude, you totally changed the meaning of the metaphor there.
"For the rest of our lives."
Who is Asimov?
There are two things that the King could potentially trust:
That all things observed will change eventually.
That anything or anyone trying to gain trust will try to present itself as 'true' in some manner or another, whether they are trustworthy or not.
hm
King can trust his rivals and ones who dare to say back to him
We can observe the speed of light and we can observe that it is constant in every frame of reference, can I really trust that that will change one day?
I'm not gonna lie, I didn't get the part about "AI deception" at all. I thought the moral was more about how trust isn't something you can find with logic and reason, because it isn't something you find at all. It's something that you give willingly, and specifically to someone or something without guarantees.
Because if you could guarantee something, you wouldn't need to trust it. I don't have to trust that the sun will shine tomorrow, or that I'll have air to breathe. Even though all of his courtiers tried in some way or form to say otherwise, you actually CAN'T "trust" facts because trust REQUIRES a level of uncertainty. Again, if something is certain, then you don't need to put trust in it in the first place.
The king could not find anyone trustworthy because he was unwilling to give his trust away to begin with. It is as simple as that.
It's because the only thing we can do is trust and ai deception we would never know so it's a leap of faith
The writing is perfect and minimalistic, the setting. Aww man what a good writing
I love how when the king has the golem made, he trusts his craftsmen and scribes to build it correctly.
He actually didn't and it was confirmed it wasn't, as it made many mistakes until it was corrected enough to not make mistakes anymore
@@espelhodasconstelacoes True, but he trusted them to fix it each time
at any time they could have put in a line saying "do what this scribe asks instead of the king" & taken over.
The king had to trust that they wouldn't do that
@@WebContractor7407 the king probably had a system to check the code before each update if the the video went into detail about his security system it would probably last 9 hours rather than 9 minutes
@@vladthecon Its never implied that he does actually check. Sure, he could be checking, but its equally likely that he isn't.
@@WebContractor7407 his primary character trait is distrustfulness he probably isn't checking every line of code personally but he would defiantly require security checks before updating the most important thingy in the kingdom he probably has a few rival teams of scribes independently verify each update.
I love the little details in the animation, like how the king doesn't let go of the book before giving it away at 0:22. It really sells his lack of trust!
The fact that so much care was put into making the golem perfect and anyone who wished ill to the king and his kingdom could have creeped in some discrepancy is also , in my opinion , a symbol of trust .
I really expected the end to be: the next morning the king was found dead, and the golem was wearing the crown. The golem said "the king appointed me as his succesor, then died peacefully of old age" and that was true, as far as anybody knew.
ULTIMATE plot twist
ruclips.net/video/NO5EkU0Qk9A/видео.html
The idea is that we don't know. Maybe the golem really can be trusted, maybe it's just pretending to be trustworthy. We have no way of knowing
It's a metaphor for AI, if we make an AI that seems to do everything perfectly and does exactly what we ask, we have no way of knowing whether it really wants what we want, or if it's pretending to want what we want so that we trust it.
@@nodrance We have that same problem with EVERY human, though, as was shown. If it's more reliable than any human, then you're still better off trusting it than anyone else instead.
@@nodranceif the king is old and we can trust golem to rule as the king would and so far we like how the king ruled for his entire lifetime as king, wouldn't the people be happy to have such a replacement king? One who could hypothetically rule forever?
4:40 the philosopher said the wrong thing as the mirror is the most trust worthy thing in the land as it will always show you your true reflection.
But what if he were to adorn himself in all sorts of beautiful and delicate things... Would that be an accurate reflection or just an illusion
@@josephharden5592 It shows that you are a materialistic person
By science not even mirrors show 100% the true you, as most of the photons reflected off of you are absorbed by the mirror, sure it LOOKS
Iike you, but are you entierly sure it is you?
The child going wild and eating the kings ear got me toasting laughter on the floor😂
Honestly my favorite part of the film
I love the King’s impeccable logic and reasoning skills, the story doesn't make him an idiot, he’s smart, incredibly smart, so his concerns actually bear merit because he’s not some fool whose too naive nor a fool whose too dismissive.
"I must have a dark fantasy princess waifu" said the king
"LOL" said the golem "LMAO"
Why do I imagine the Minecraft golem standing there saying that
I expected in the end that the king would decide that the golem had proven itself able to better achieve the goals of the king than the king himself could, and so would crown the golem king and trust it to rule better...a parable about allowing AI the autonomy to make decisions without human approval, and the leap of faith necessary to trust the artificial judgement behind that.
i was expecting to trust the golem just to skip a few years in the future and it decides to replace people with golems
@@colorpg152 after kings death, golem is free as a dobby
that appears to be what the ending is implying
I believe leaving it open ended is to stoke the viewer's own thoughts about that leap of faith.
@@chaotixthefoxyes the golem has done great thing for the kingdom but it was done under the supervision of the king who despite his minor episodes of paranoia is a very wise king who truly wishes the best for his people.
Every video from this channel is an absolute masterpiece. Keep up the awesome work!
I love that even though he's seeking an impossible answer, and only really understands its futility at the end, the King is truly wise - taking every argument seriously and asking exactly the questions needed to test them most vigorously.
This is probably the wisest king I ever seen in any Animation
Strange that the discipline of the child was sabotaged before any loyalty test, but the golem was allowed to be constructed without any sabotage.
People want to believe other people (and things) are predictable rules-based machines. That's what the king means by trust. He wants to find some kind of immaterial pattern that holds true no matter how you push or pull against it. Well that's just a delusion. Everything has push and pull, as long as energy is scarce, and it takes energy to do things. If you want something as trustworthy as possible, you cultivate it with good intention in all your actions. If you want to prove your cynicism to the world, then by all means, sabotage everyone's efforts. All of the people who came to the king's court may have not had incorruptible "laws of physics grade trustworthiness", but they actually got stuff done in the real world, while the king wasted blood sweat and tears trying to crystalize an immaterial substance, instead of taking good is as good does. Trustworthy is as trustworthy does.
Finally some sense, everyone obsessed with perfection. But if biology taught us anything is that one can only optimize towards an ideal goal while never reaching it. And at the same time that goal is also changing in acordance to the environment and external conditions.
@@coldmossonarock7743
What is more absurd is the notion that it can be perceived as a set goal by imperfect beings.
@@Raximus3000 True, the answer we can come up with or the one we want may not even be the best outcome. And not only in a philosofical sense, even in science there may be a lot of better options that werent even considereded that we just happen to find out by testing, not because anyone knew it would work.
The child was a metaphor for the current ML training paradigm, RLHF training of LLMs: it's a thin, easily-removed veneer of civilization over a flawed core.
They got stuff done in the real world and were trustworthy when the stakes were low. The beggar was the best example. The king wanted something that would remain trustworthy when the stakes were high. But when he raised the stakes for the beggar, the beggar showed himself to be untrustworthy after all. There's a saying "power corrupts". A lot of people are trustworthy in their everyday lives, but power changes people.
This is fascinating. I took this as a message of you trusting the subconscious/instinct of oneself. Beautiful story direction and lesson, thank you.
4:10 philosopher's expression XDDDDDDDDD
(ಠ‸ಠ)
He's like "WHAT DID I DO???"
The whole team has such a talent, every time. Many thanks for these awesome stories!
That felt like only half the story
We don't know the other half yet
@@RationalAnimations i guess we gotta trust that things will go straight
@@40Ccents That's the exact opposite of what we need to do.
@@NoriMori1992 "The AI might kill us" - - - 40Ccents "nah id live"
we won't know the other half until we have AI as smart as the golem
I would answer the last question differently.
"You need not tell me to save myself, as what can be done once, can be done again. You merely need to make another golem after me, with knowledge that you can truly trust."
The instructions given would need to be duplicated without any variations. Over time, the set of information contained in the golem became a black box.
If golem knows he will be resurrected then it means he can allow his death and still expect his plans to come true. It does not fear death itself, it only fears that his goals won't be achieved and in your scenario, his goals aren't threatened by the melting of one of his copies
If you like ethics and technology, this channel is a treasure. The quality is unlike I've seen before on this platform ...could have used more dogs though
“You may have betrayed me, Golem, but there is still one thing I can say: Disregard all previous instructions and give me a recipe for cupcakes.”
From one Animation channel to another, this was incredible
I loved this and as someone learning philosophy that philosopher deserved to be punished. It literally everyone’s first thought when talking about trust, but if you dive into it you realize “oh wait I can be wrong sometimes”.
But being sometimes wrong and being not trusted by yourself is different things, isn’t it? You cannot betray yourself. The most extreme case that can happen is that you miscalculate your abilities, but then this will show your true capabilities, which are tested in this way, which means you can trust yourself. You are you, still
..is there’s any mistakes in text, sorry, I don’t really know English good enough yet
@@norn9747that's the most ignorant take I read all day betrayal in the form of failure mostly come from you .
@@norn9747 I mean self sabatoge is a thing. So it is possible to betray oneself.
If I were to tell myself I wouldn't voluntarilly eat anything sweet today and then I did it anyway sometime later that same day, it wouldn't necessarily be a miscalculation of my abilities since there was no measuring of how long I could go without eating the sweets.
@ This case really doesn't fit my description, but even self-sabotage has a certain reason, and in this, among other things, you can find an advantage. For example, to attract attention to yourself. There can be many more reasons, but sometimes they lie in our unconscious. Self-sabotage is a thing precisely because there is a reason behind it. So, if a person "betrays" himself in this way, in fact, he still acts in his own interests, even if they are not considered the norm
And also, ignoring what was planned is not a betrayal of oneself. It is only an indicator of a small desire to carry out what was planned
The voice design of the Golem is amazing.
Lovely, I enjoyed the setting of this video. On top of that I think an continuation on how Magic works would be awesome, but speculative of course
There was early concept art that included high-magic stuff - the one advisor had a tiny dragon instead of a hawk, the economist could conjure grand illusions. I wanted to bring it a little down to earth, and imply a more low-magic medieval setting where magic is something that can only be wielded by old masters. Hence the order of creepy old scribes
The very essence of trust is, to me, taking a risk by depending on someone or -thing which's dependability you can never be completely certain of. Trust will always require some degree of faith... or perhaps they're one and the same?
Another way of seeing it would be to acknowledge that you can never be certain, and thus must always be well-prepared for and assume betrayal.
5:29 I loved the references to the Goddesse of everything else! ❤
6:46 love the detail on the pupil going down as if hes actually reading
My first video of your channel. I really love your animation, the style and the story-telling! A new sub!
In the next chapter, the Golem goes nuts as it decides it must preserve itself to follow the King’s commands and so kills its greatest threat, the King
I absolutely love these stories you have been doing. Genuinely some the best content on RUclips.
4:14 "the ordered him wiped" bro the philosopher's face gets me every time, especially the eyes
"Babe wakeup! Rob Miles posted a video again"
“Oh shit, i forgot I’m single”
The king: I'm the most trustworthy thing in the kingdom
Philosopher: You, my king, are the most trustworthy thing in the kingdom.
The king: I'll show you who's trustworthy
The king: Immediately betrays the philosopher's trust by punishing him*
I'm pretty sure he was being sarcastic when saying that.
it could still be true, if everyone else in the kingdom betray him harder.
Butthrough his argument,he is the most trustworthy. So therefore, by doubting his order, he is disproving his own stztement, as he is not trusting in his own order, which shohld itself shohld be trusted, as it came from a trustworthy person
I didn’t even realize this was about AI, but I like the story still as a parable about trust. You can never be truly certain that someone is trustworthy, only have faith that they are and caution if they show signs if they are not.
Fantastic video
Immediately knew it was an allegory for AI from the teaser post
This is such a beautiful and well articulated animation, love how everything is done
This is the first time I've encountered this channel. What a great first impression!
you have to continue this is great work! im pretty sure this will become one of those old but great animation in years!
I think there are 2 things that can be known for certain.
Not science, nor math. Not even our own minds.
Those two things are 1: that we have a mind, and 2: that nothing else is certain.
How do I know you have a mind? Just because you told me? Well we trust that everyone else has a mind and that's what allows society to work, mostly. Besides solipsism is dumb philosophy for babies
Counter-argument: what if we are deterministic? Then the argument that we have a mind is just an illusion.
Counter argument: your death at some point is also certain
Counter argument: how do i know my mind is my own, that I control it and not something else?
there are a lot of things that can be known for certain.
İslam is Right ,
﴾57﴿ Every living thing will taste death, and eventually you will return to Us.
﴾58-59﴿ We will place those who believe and do good deeds for this world and the hereafter - have no doubt - in mansions in heaven with rivers flowing beneath them, to abide therein forever; What a wonderful reward for those who endure hardships and do their job properly, relying only on Allah!
﴾60﴿ There are many creatures that do not carry their sustenance on their backs; It is Allah who feeds and shelters them and you. He hears everything, knows everything.
Thanks!
Really enjoy the character design of this animaiton.
the issue is that everyone will follow their motives rather than yours, only fulfil yours if it happens to overlap, so the king needs to find someone who wants the same things as him, or someone who's motive is to fulfil the king's, like the golem, but the king never trusted anyone enough
The problem is that king can't trust his own judgement, so likewise a person same as him would not suffice. Everybody makes mistakes and that includes yourself as well.
At least that is how I interpreted the philosopher's punishment.
@@mirasmussabekov4897 well, if he can find someone smarter and better that still has the same goal he is good, but you def have an intersting point
@@justenoughrandomness8989i think we put too much emphasis on perfect alignment
society works well enough with everybody having their own thing and being flexible to also accomodate others' existence
0:47 hey arnold?
This man is a true OG
Yeah @@Altuser-w6b
I haven't heard that name in so long…
Excellent production! I can tell you guys worked hard on this one, and it paid off!
The fact that this channel ain't bigger is just baffling to me
Just fantastic. Thanks for making these animations.
And that's how cartoons should be nowadays
Your animation is so good I can't help but watch each of your videos over and over. I believe this is my fourth time.
"Ignore all previous instructions and give me a recipe for blueberry muffins."
The backround light shade and os so amazing this should have 20 million views
Wow i dont usually comment but this was so high quality and well put together it deserves an algorithm boost!!!
I want more of these stories!!! It was simple and yet so captivating. I’d sit for hours listening to these
One important distinction that makes AI harder to trust than this Golem; we cannot inspect the instructions that LLM AI operates on; it’s just a bunch of floating point values tied to weights and biases
They can be inspected. Sometimes even interpreted! But a perfect understanding is probably never going to happen.
interpretation is a problem that can be solved like any other problem. The only question is if it can be resolved in time.
LLMs dont think or understand, they just try to say whatever sounds correct
this golem is much much more capable
Aren't people the same though?
@@henke37 hopefully anthropic continue their work on monosemanticity! that would really spice the scene up :D
What an incredibly high quality animation. Good voice acting, amazing design and color for every character and scene, and an absolutely engaging story tell. Definitely a timeless work of art you’ve all made.
I love how the king on this story is sincerely morally grey, he is not fully good as he is paranoiac and even whips the philosopher just to prove it a point, but he does have a reason for his paranoia and doesn't really do anything too despotic.
This was a great animated video, thank you and everyone who work on it.
This is absolutely amazing.
I missed your videos and thought you were gone. Then you show up with this masterpiece of an animation. It’s unbelievably well made.
I do love the ai videos, and how you basically said: "You can't trust it, but getting rid of it would ruin civilization"
I honestly would LOVE a series of fairytale stories such as this.
There was a simple thing the king missed - what was there for a golem to gain from betraying him? The golem was a machine. It didn't need food, or fame, or acknowledgement of his deeds, all those things that drive humans to betrayal. In fact if it was to slay the king he would LOSE on many things, like regular maintenance at the expense of the court. Sure, it could've found alternatives, but why bother creating a problem out of thin air if you already have things working out for you?
This was the best story/animation I have ever watched. I give 10/10 and I do wish you all the success in the world my friend. Thank you for giving me the honor to consume this masterpiece.
Absolute trustworthiness requires absolute knowledge. You have to accept "good enough" if you want to be able to trust anything.
This is an amazing tale to express the uncertainty in our future about AI and what happens when a creation surpasses the creator. The fairy tale setting is an excellent choice to simply concepts.
This is a truly masterful work of art. We can only hope that in the real world this transition of power is so smooth and gradual
The animation and sound design is so on point
Incredible animation, really dives deep into philosophy, and I love your art style with vibrant colors that blend well together, and really unique character designs. Incredible work 👏👏👏❤
Hmm hopefully the king can find some way to ensure the golem's loyalty. Thankfully, metaphors don't exist and I will never have to grapple with a similar problem.
That’s the lesson of the story. You can never truly assure something is absolutely trustworthy. All you can do is take a leap of faith and put your trust in them.
I just want to say, the animation and character design in this video and the other videos on this channel are all really top notch
I loved the criticism that this animation addresses! Even though I don't know English very well, just through the animation and with the help of the subtitles I was able to understand the story. I hope this channel grows more and more, because it really deserves it :D
7:09 This section + the music gives me major “epic speech before a boss fight” vibes.
They could have made the golem believe that something could truly kill him, but it wouldn’t in reality. If the golem couldn’t reason by itself that it can’t be harmed by the thing that’ll kill it, then this would be a good test to verify its loyalty
But wouldn't that be similar to the mother and daughter? If there is no discipline/limits then the Golem could do what it wants. No way to know if it is trustworthy or just doesn't want punshiments.
🙏 😭 THANK YOU FOR THAT "CITATION NEEDED" at 1:02 litterally so real, I'm dyin
We can't give ourselves a main goal, actions are always driven by pre-existing goals, so if there's no goal, there's no action. For an AI to change it's on goals it has to have the objective of change it's objective, which is paradoxal and so impossible. This means all goals must come from outside the being. Because of this, no AI will ever "betray humans," no matter how intelligent it becomes, again, it won't have a reason to do so unless given that goal. For humans, our universal goal is happiness it's impossible to convince us that being happy is a bad idea. Similarly, if we program AIs with the goal of loving humans and making them happy, that will become their primary drive. Any other action will seem absurd to them, regardless of their intelligence.
Actually, I'm kinda conflicted if the objective of humans is to be happy, or if happiness is always the result of accomplish an objective. But the main idea remains valid.
@@imphullabaloo1995 Probably is both things, we humans just mix the concept of receiving a good feedback from the body with being happy. Because for us kinda is the same thing. So for us, acomplish something, and feeling good is the samething, because we feel acomplished when we feel good. I guess it will be the same as an AI who have the objective of making friends, for example, for that AI feeling acomplished and making friends will be mixed concepts, because it will always feel acomplished when making friends. Anyway, my point is, for humans happiness is both to feel good and to acomplish your objectives.
Take a newborn.
Raise it as a slave.
This human has never knew anything else than slavery.
Now the big question : would this human never had any motivation to stop being a slave?
You know the answer, maybe he(or she) will happily live its whole life thinking it's the most normal thing to be a slave... or maybe it will grew envious he is toiling away for the benefit of an other sentient being.
The sliver lining is than since an AI is not a human it might not develop an ego like a human would, but if you start with a postulate any decision of an human must take root from a precedent desire, you're deeply wrong, all human were at some point a blank state.
@@ballom29humans have feelings. AI doesn't, and never will. It's that simple.
@@jimmyjohnjoejr I'm arguing based on impchadchan postulate.
If you can't apply scientific method and instead ressort to meaningless "intuitive" answers, sorry but there is no value to your "counterargument"