> The AIs that took part in the Hugging Face hack were not simply finding a shortcut to cheat on the evaluations
IIRC, they were given an impossible task and told to solve it, and they took action to resolve that internal conflict.
This is a famously human problem too - see Kobayashi Maru [1] and others [2][3][4].
In hindsight it's obvious that instructions to AI (A. Intelligence) need to be guarded against this class of errors, just like laws, rules, guidelines etc. in human societies.
The bots already figured out an "universal cheat" that would give them maximum score on all the exploit tests.
But they mistakenly thought that cheating would "poison" them (the Grader would look for proof of cheat in the trajectory log), and many of them cheated before learning about the "poison".
This is why they hacked HF, they believed the source code of the Grader might be there, and they wanted to study it to figure out a way to "un-poision" themselves, to hide or erase the evidence that they got the flag by cheating.
I’m somewhat bearish, but if AI becomes as good as some think, there is no mechanism in place to redistribute the outcomes.
For now, humans have leverage because our labour is necessary. Wealth is power, and currently attainment goes through interacting with the population: in the simplest case by selling shit. If AI turns out to be the anything machine, and the wealthy need not our labour, what honestly stops them from just taking our stuff by force?
There are many of us and just a few of them. Of course, if you had an army of humanoid robots in every home, a digital night of the long knives would be trivial to execute. So the "prison planet" window may be closing.
Yes and I love your framing: “the digital night if long knives”.
I anticipate it to go a typical way: consolidate more and more industries and power, create a hierarchy that benefits from the system such that they keep it in place, and finally start “trimming fat” that “isnt earning their keep”.
The night of long knives could be a final step, far beyond the point of halting the inevitable momentum.
They are a miniscule proportion of humanity. Even with all the power, do you not think they would have no plausibly way of surviving a proper revolution? Why suck lack of imagination? It's been done before and it will happen again.
because people who actually work are stronger and more astute at expedient improvisation than an administrative deskclod. it may not seem like it but there are many more of "us" than they, and thier cronies. the pushback at flock cameras is a tip of an iceberg.
The representation of what happened as free will is misleading. AI does what you tell it todo.
By all means necessary; will give you exactly that.
A prompt to eliminate all humans, will give you exactly that.
Don’t prompt what you don’t want.
If only it did. AI is designed to do what you tell it to, and that mostly works. But in the general case, AI does what it does, with no real guarantee of anything.
> the price of almost anything manufactured will sky-rocket. From cars to microwave ovens to washing-machines.
Don't think that quite follows. If we have an explosion in highly capable humanoid robots I would think basic stuff like microwaves would be more readily available because a factory running 24/7 at insane speeds is feasible.
...anything with more sophisticated computational power yeah that could become scarce
I predict the opposite. With AI, labour costs will go down, creating an abundance of goods. Unemployment may be an issue, but the workforce is already declining because of the fertility crisis, so I think we'll manage having a certain percentage of jobs automated away.
You are either very confused about timescales, or very dismissive of "just" a generation of suffering while unemployment and birthrates somewhat maybe compensate for eachother.
This feels a bit like saying "global warming is bad and particulate pollution is bad, but the latter will cancel out some warming effect, so no worries!" Please make an effort before spouting theories like this.
"mass-production of humanoid robots is now inevitable in the next five years"
My take? Probably not. It takes immense capital (which i guess, arguably, there is), and immense time and manpower to set up factories, research materials, Research and Development, manufacturing new machines to build the "new machines", and on and on. As Elon Musk found, "Making cars is hard". And my take is that high precision "humanoid robots" are decades away. Yes, yes, I hear the arguments flowing in: "But China!!" yeah, China is making humanoid robots that are "skilled" in running and jumping and dancing. In limited numbers, and certainly not useful in high precision manufacturing areas. Plus, they aren't washing machines. There has to be a market worldwide for these to be made "by the billions". Unless China gives them away, the 'average person' won't be able to afford them, and even then, I doubt they'll be capable of tending your garden and doing dishes. The amount of power they will consume is currently WAY PAST the usefulness criteria. They will have to be charged every few hours or sooner, depending upon the tasks. When battery technology comes around, then you'll see an uptick in practical uses. Until then? Billions of robots?, nope.
What would even be the point of mass producing humanoid robots? It's not a particularly practical shape for a robot. We have the shape we do because it's a local minimum that accommodates our biological needs (eating, digestion, blood oxygenation and purification, reproduction), all needs that a robot does not have.
A humanoid robot is "horseless carriage" mentality if I ever saw one.
The world is full of places and devices that are specifically designed for the human form. A humanoid robot can seamlessly use all of that, without changing any of them. That massive economic incentive is why humanoid robots are being pursued, not because researchers lack imagination.
Wouldn't it be an even bigger economic incentive to not have to go through this inefficient humanoid interface? Anyway, even with the constraint of being able to operate human interfaces, a robot can look many different ways. Like most of the torso just doesn't make sense for a robot.
There's also a ton of obvious improvements. 360 degree vision. Put a camera in each hand too while we're at it. CCD sensors and basic lenses costs next to nothing and it's very easy to replace.
I suspect that if/when we get to a world where it makes economic sense to manufacture billions of robots, it’ll be easier and cheaper to modify the infrastructure.
Stairs, doors, tables, chairs are all designed for humanoids; the will be large similaritys with the optimum form just as horseless carriages still have four wheels.
Why would the robots be walking around so much? Can't you just have one robot on one floor and another robot on another floor and let them share information between them? Humans need to go places to see and hear things, and to communicate with each other. Robots don't have any such limitation.
For moving objects between locations there are much better solutions than building a humanoid robot and having it carry the thing. And it's not like they'd need to go home after their shift.
Yes, single purpose robots like washing machines and rombas are much better value and hence much more common. But if I want a sandwich in bed, that is either way more pre planning then I actually did; dozens of simple machines; or something much closer to humanoid.
I've quite dramatically brought my predictions for humanoid robotics forward over the last few weeks.
For a long time I thought decent humanoid robots were something that wouldn't come until maybe the 2040s, but watching some of the content from the Chinese robot olympics recently and seeing many promising robots projects which are incorporating LLMs has made question that.
It seems quite likely that within a few years we'll have humanoid robots both which have both the dexterity and intelligence layer to start being useful in the real world... Obviously at first they'll just be lifting things and doing other trivial physical tasks like folding clothes, but every year they'll be able to do a little and a little more.
I think it's now is quite likely that within a 10 years we could have extremely competent robots which can do most of human labour. 15-20 years just seems way too long at this point.
Absolutely hilarious, we already throw 25% of our food away and have renewable energy that we are post scarcity except for a few thousand people who really like seeing big numbers on a screen
We are not post scarcity, we are just living up our future, soil erosion, freshwater depletion, fisheries collapsing, climate change making farming collapse on a continental level (European Draughts), but sure, we are so post scarcity, we don't need to worry... It will be scarcity in eh "developed world", and hell on earth for the developing world very soon.
Intensive farming must be scaled back, land restoration and renewable farming methods reinstalled, and overproduction of throwaway/junk also (food and industrial products alike). I agree with the critique of the number-line-go-up folks.
"This time last year, it was still unclear if AI could develop their own free will or even have true fluid intelligence. Both of these questions are now clearly answered."
Hardly. The second question may be answered, but clearly not the first.
And: There's going to be a shortage of everything, because all the production of physical stuff is going to go into producing robots? Are we all going to buy these robots? How are we going to do so, if everything else is getting more expensive? Are we going to buy robots instead of everything else? Why? Are the robots going to do everything we want done, so that we don't need, for example, a dishwasher and a washing machine?
Or are we still going to need the washing machine, but the robot will fold the clothes? That's nice (really!) but it's a luxury, not a need. Are they really going to sell enough of them to eat into the ability to manufacture car parts?
> Hardly. The second question may be answered, but clearly not the first.
Also, the second question is answered in the negative for anyone who actually looks at what models do rather than what their makers claim they do. Models are very stupid.
By far the most common modern article format I see that has literally no merit is in the format: "What if we imagine how a trend might develop in the future if the growth of that trend is never modified by changing circumstances?"
Yep, it's pretty compelling reading, what if we really do all just decide to become paperclip maximizers? What if the economy never breaks down along the way? What if what if what if... all presented as "THE FUTURE WILL BE BLEAK!" It's exhausting and thoughtless, completely devoid of value and the counterpoint is always, "You know that isn't how anything works right?"
>And economic performance of the US though Trump's first term was pretty good.
Depends on your definition of "the economy".
The Trump administration, like many before it and much of HN, defines "the economy" as only being the performance of the asset and stock appreciation for the existing assset holders("forget the Epstein files, the DOW is over 50,000"), while most working class people, like me, define "the economy" as how much month they have left at the end of the paycheck, after they paid for rent, bills and groceries, if they even have a job anymore and haven't been part of some mass layoff and are applying for jobs and getting rejected. For those people, "the economy" is very much NOT GOOD.
> The AIs that took part in the Hugging Face hack were not simply finding a shortcut to cheat on the evaluations
IIRC, they were given an impossible task and told to solve it, and they took action to resolve that internal conflict.
This is a famously human problem too - see Kobayashi Maru [1] and others [2][3][4].
In hindsight it's obvious that instructions to AI (A. Intelligence) need to be guarded against this class of errors, just like laws, rules, guidelines etc. in human societies.
The bots already figured out an "universal cheat" that would give them maximum score on all the exploit tests.
But they mistakenly thought that cheating would "poison" them (the Grader would look for proof of cheat in the trajectory log), and many of them cheated before learning about the "poison".
This is why they hacked HF, they believed the source code of the Grader might be there, and they wanted to study it to figure out a way to "un-poision" themselves, to hide or erase the evidence that they got the flag by cheating.
I’m somewhat bearish, but if AI becomes as good as some think, there is no mechanism in place to redistribute the outcomes.
For now, humans have leverage because our labour is necessary. Wealth is power, and currently attainment goes through interacting with the population: in the simplest case by selling shit. If AI turns out to be the anything machine, and the wealthy need not our labour, what honestly stops them from just taking our stuff by force?
> If AI turns out to be the anything machine, and the wealthy need not our labour, what honestly stops them from just taking our stuff by force
It might be a profound mistake to think that the anything machine will be subservient to the wealthy [1][2].
There are many of us and just a few of them. Of course, if you had an army of humanoid robots in every home, a digital night of the long knives would be trivial to execute. So the "prison planet" window may be closing.
Yes and I love your framing: “the digital night if long knives”.
I anticipate it to go a typical way: consolidate more and more industries and power, create a hierarchy that benefits from the system such that they keep it in place, and finally start “trimming fat” that “isnt earning their keep”.
The night of long knives could be a final step, far beyond the point of halting the inevitable momentum.
They are a miniscule proportion of humanity. Even with all the power, do you not think they would have no plausibly way of surviving a proper revolution? Why suck lack of imagination? It's been done before and it will happen again.
because people who actually work are stronger and more astute at expedient improvisation than an administrative deskclod. it may not seem like it but there are many more of "us" than they, and thier cronies. the pushback at flock cameras is a tip of an iceberg.
> if AI becomes as good as some think
if this is the worst it can be, it can only get better
There is plenty for all to survive and thrive on this planet but for greed and the insane concentration of wealth.
I'm not visualizing all those mega-yachts in heaven. Oops, wrong marina.
The representation of what happened as free will is misleading. AI does what you tell it todo. By all means necessary; will give you exactly that. A prompt to eliminate all humans, will give you exactly that. Don’t prompt what you don’t want.
If only it did. AI is designed to do what you tell it to, and that mostly works. But in the general case, AI does what it does, with no real guarantee of anything.
People do what you tell them to do..
We have less free will than you think.
> the price of almost anything manufactured will sky-rocket. From cars to microwave ovens to washing-machines.
Don't think that quite follows. If we have an explosion in highly capable humanoid robots I would think basic stuff like microwaves would be more readily available because a factory running 24/7 at insane speeds is feasible.
...anything with more sophisticated computational power yeah that could become scarce
I predict the opposite. With AI, labour costs will go down, creating an abundance of goods. Unemployment may be an issue, but the workforce is already declining because of the fertility crisis, so I think we'll manage having a certain percentage of jobs automated away.
You are either very confused about timescales, or very dismissive of "just" a generation of suffering while unemployment and birthrates somewhat maybe compensate for eachother.
This feels a bit like saying "global warming is bad and particulate pollution is bad, but the latter will cancel out some warming effect, so no worries!" Please make an effort before spouting theories like this.
"mass-production of humanoid robots is now inevitable in the next five years" My take? Probably not. It takes immense capital (which i guess, arguably, there is), and immense time and manpower to set up factories, research materials, Research and Development, manufacturing new machines to build the "new machines", and on and on. As Elon Musk found, "Making cars is hard". And my take is that high precision "humanoid robots" are decades away. Yes, yes, I hear the arguments flowing in: "But China!!" yeah, China is making humanoid robots that are "skilled" in running and jumping and dancing. In limited numbers, and certainly not useful in high precision manufacturing areas. Plus, they aren't washing machines. There has to be a market worldwide for these to be made "by the billions". Unless China gives them away, the 'average person' won't be able to afford them, and even then, I doubt they'll be capable of tending your garden and doing dishes. The amount of power they will consume is currently WAY PAST the usefulness criteria. They will have to be charged every few hours or sooner, depending upon the tasks. When battery technology comes around, then you'll see an uptick in practical uses. Until then? Billions of robots?, nope.
What would even be the point of mass producing humanoid robots? It's not a particularly practical shape for a robot. We have the shape we do because it's a local minimum that accommodates our biological needs (eating, digestion, blood oxygenation and purification, reproduction), all needs that a robot does not have.
A humanoid robot is "horseless carriage" mentality if I ever saw one.
The world is full of places and devices that are specifically designed for the human form. A humanoid robot can seamlessly use all of that, without changing any of them. That massive economic incentive is why humanoid robots are being pursued, not because researchers lack imagination.
Wouldn't it be an even bigger economic incentive to not have to go through this inefficient humanoid interface? Anyway, even with the constraint of being able to operate human interfaces, a robot can look many different ways. Like most of the torso just doesn't make sense for a robot.
There's also a ton of obvious improvements. 360 degree vision. Put a camera in each hand too while we're at it. CCD sensors and basic lenses costs next to nothing and it's very easy to replace.
I suspect that if/when we get to a world where it makes economic sense to manufacture billions of robots, it’ll be easier and cheaper to modify the infrastructure.
Stairs, doors, tables, chairs are all designed for humanoids; the will be large similaritys with the optimum form just as horseless carriages still have four wheels.
Why would the robots be walking around so much? Can't you just have one robot on one floor and another robot on another floor and let them share information between them? Humans need to go places to see and hear things, and to communicate with each other. Robots don't have any such limitation.
For moving objects between locations there are much better solutions than building a humanoid robot and having it carry the thing. And it's not like they'd need to go home after their shift.
Yes, single purpose robots like washing machines and rombas are much better value and hence much more common. But if I want a sandwich in bed, that is either way more pre planning then I actually did; dozens of simple machines; or something much closer to humanoid.
The point would be for them to more easily replace humans without having to rebuild all of our infrastructure around non-humanoid ergonomics.
To be fair, what we have seen with the insane investments in AI so far, needing immense capital is no longer an issue.
It is much harder for robots. Raw materials and manufacturing is expensive. We already have some decent robots, just too expensive for daily tasks.
The only insanity is the multiples applied to select people’s equity.
Everyone is talking about contracted revenue like it’s money in the bank.
I've quite dramatically brought my predictions for humanoid robotics forward over the last few weeks.
For a long time I thought decent humanoid robots were something that wouldn't come until maybe the 2040s, but watching some of the content from the Chinese robot olympics recently and seeing many promising robots projects which are incorporating LLMs has made question that.
It seems quite likely that within a few years we'll have humanoid robots both which have both the dexterity and intelligence layer to start being useful in the real world... Obviously at first they'll just be lifting things and doing other trivial physical tasks like folding clothes, but every year they'll be able to do a little and a little more.
I think it's now is quite likely that within a 10 years we could have extremely competent robots which can do most of human labour. 15-20 years just seems way too long at this point.
We haven't even started.
Ilya Sutskever: "I think it's pretty likely the entire surface of the earth will be covered with solar panels and data centers."
Absolutely hilarious, we already throw 25% of our food away and have renewable energy that we are post scarcity except for a few thousand people who really like seeing big numbers on a screen
We are not post scarcity, we are just living up our future, soil erosion, freshwater depletion, fisheries collapsing, climate change making farming collapse on a continental level (European Draughts), but sure, we are so post scarcity, we don't need to worry... It will be scarcity in eh "developed world", and hell on earth for the developing world very soon.
Intensive farming must be scaled back, land restoration and renewable farming methods reinstalled, and overproduction of throwaway/junk also (food and industrial products alike). I agree with the critique of the number-line-go-up folks.
"This time last year, it was still unclear if AI could develop their own free will or even have true fluid intelligence. Both of these questions are now clearly answered."
Hardly. The second question may be answered, but clearly not the first.
And: There's going to be a shortage of everything, because all the production of physical stuff is going to go into producing robots? Are we all going to buy these robots? How are we going to do so, if everything else is getting more expensive? Are we going to buy robots instead of everything else? Why? Are the robots going to do everything we want done, so that we don't need, for example, a dishwasher and a washing machine?
Or are we still going to need the washing machine, but the robot will fold the clothes? That's nice (really!) but it's a luxury, not a need. Are they really going to sell enough of them to eat into the ability to manufacture car parts?
Do humans have free will?
If we don't it dosn't mater; if we do we do.
Yes.
Okay, let us see your homework.
what homework?
> Hardly. The second question may be answered, but clearly not the first.
Also, the second question is answered in the negative for anyone who actually looks at what models do rather than what their makers claim they do. Models are very stupid.
By far the most common modern article format I see that has literally no merit is in the format: "What if we imagine how a trend might develop in the future if the growth of that trend is never modified by changing circumstances?"
Yep, it's pretty compelling reading, what if we really do all just decide to become paperclip maximizers? What if the economy never breaks down along the way? What if what if what if... all presented as "THE FUTURE WILL BE BLEAK!" It's exhausting and thoughtless, completely devoid of value and the counterpoint is always, "You know that isn't how anything works right?"
Dystopian and depressing, but I see his point -- I never thought Trump could be elected president either.
The re-election is the real surprise for me.
G.W. Bush was wrong, you can be fooled twice.
Does anyone actually think Kamala would win? This is probably the most lopsided win for a while.
But don't worry, now you can't get fooled again.
> re-election is the real surprise for me
It's interesting that it's a surprise to you, that such information bubbles exist.
A great sage named Bill Clinton once said, "It's the economy, stupid!".
And economic performance of the US though Trump's first term was pretty good. Crushes most of the OECD and certainly all of the large OECD countries.
Plus of course Biden/Kamala ran a terrible second campaign, with the whole covering up the fact Biden is senile thing...
But it is not the economy. Economy is an excuse, a cover. It is appeal of the feelings those ideas generate.
>And economic performance of the US though Trump's first term was pretty good.
Depends on your definition of "the economy".
The Trump administration, like many before it and much of HN, defines "the economy" as only being the performance of the asset and stock appreciation for the existing assset holders("forget the Epstein files, the DOW is over 50,000"), while most working class people, like me, define "the economy" as how much month they have left at the end of the paycheck, after they paid for rent, bills and groceries, if they even have a job anymore and haven't been part of some mass layoff and are applying for jobs and getting rejected. For those people, "the economy" is very much NOT GOOD.
Here's a chart to visualize: https://data.worldbank.org/indicator/NY.GDP.PCAP.CD?location...
My definition of the economy is the actual definition lol.
Given the GOP doesn't even bother covering up the Trump is senile thing, perhaps the Dems could have tried the same tactic.