That applies to any technological innovations that improve efficiency though. What’s the solution to making things actually benefit everyone? What worries me the most is if it all gets developed in closed labs where the wealthy maintain full control of it and reap all the benefits. I was thinking that if the tech was under our control, it would then at least be beneficial to more people. I don’t find it to be a satisfactory solution, but I haven’t been able to think of anything better in the last five years.
At the current rate, few normal people will even be able to afford a computer or a smart phone in a few years, let alone run a private AI. These algos aren’t being made for normal people, they’re being made to benefit the privileged few. You can try all you want to be as altruistic with your design as you want, all that’s going to come of it is a cage for the vast majority of people. Whether Musk builds it, or Amodei builds it, or China builds it, it’s going to leave the vast majority of people shut out and struggling to survive. There is no kinder, gentler atom bomb. There is no world where these tools aren’t a global catastrophe that will ultimately cost billions of people their lives.
So the problem is everything moving to the cloud and that cloud being mostly controlled by a small number of big players, right? I still don’t know what the solution to that should be. In my mind, we can either assume that they’ll successfully take over and we give up (so why not make our lives easier while we’re here?), or we fight back while we still have access to the technology.
Make AI unusable. Keep Data Centers rare, keep tokens expensive, make sure no one is willing to spend on AI, and let the megacorps collapse under the weight of debt they can’t pay off. When they can’t fight back, start passing laws that outlaw private AI development and nationalize the major models. Start fighting for anti-trust laws with teeth, reduce corporate copyright entitlements while enforcing human IP rights, and void the whole concept of software patents.
Or die. Dying is always an option. I think we’d all rather avoid that one though.
I think we should clarify what we mean by AI first. I’m under the impression that you’re talking about generative models. I don’t think it makes sense to ban all form of automation. But if you really do mean all automations, then I’d like to hear the reasoning for that.
If you want to drop the marketing buzzwords and be specific, yes, LLMs, image generation/graphic design and other generative models, but also machine assisted data analytics need to be heavily regulated and taxed into a box. Military and law enforcement applications need to be strictly state owned, automated surveillance needs to be put on a bonfire entirely. There’s a lot of different classes and categories of “AI” that need to be completely re-thought, but you are correct, I am not entirely against all categories of automation.
It seems we are in agreement on most of this. These generative models aren’t going to fill your pantry or do your chores. They can help through world models, but I’m hoping to get things working without that.
Of course, then you run into the problem of access vs cost. Running the AI requires resources, those resources have market value so you need to make money. Setting aside the fact that by definition you are making sure those resources aren’t being expended on something else potentially more conducive to life, you now have to limit access and/or features to those who can pay, ie. not the people who are struggling the most and could most use the help. Given the price of processing power alone right now, let alone the rest of it, how do you propose these tools become available to the places they can do the most good?
There’s also the problem that most engineers scratch their own itch, not other people’s. The tools you’re developing almost always tend to focus on things that are personal struggles. Your tools are often going to be solving problems the typical person will never have, or at the very least, are so low priority compared to everything else your tools are inconsequential. What you need is people, and feedback mechanisms that can help identify opportunities… you know like a bureaucracy.
I agree in principle with you. Technology in every bodies hand is benefitting everybody, while under control of a few only benefits them.
But with LLM as they are currently and in my eyes in the near future as well? I’m not seeing it possible for the average person to make their own model with available hardware. Big data centres are needed and they burn a lot of resources for creating a video that turns a cat into a human, help genociders argue why they needed to bomb that school or destroy our education systems while simultaneously destroyingpublic trust in media.
When it comes to LLMs and similar generative models, I agree. I’m talking about AI in general though, and reinforcement learning, which is the focus of my work. It’s still very resource intensive and doesn’t work very well, but what technology isn’t in the early stages?
They weren’t talking about your internal motivations, but about your results. Important difference.
That applies to any technological innovations that improve efficiency though. What’s the solution to making things actually benefit everyone? What worries me the most is if it all gets developed in closed labs where the wealthy maintain full control of it and reap all the benefits. I was thinking that if the tech was under our control, it would then at least be beneficial to more people. I don’t find it to be a satisfactory solution, but I haven’t been able to think of anything better in the last five years.
At the current rate, few normal people will even be able to afford a computer or a smart phone in a few years, let alone run a private AI. These algos aren’t being made for normal people, they’re being made to benefit the privileged few. You can try all you want to be as altruistic with your design as you want, all that’s going to come of it is a cage for the vast majority of people. Whether Musk builds it, or Amodei builds it, or China builds it, it’s going to leave the vast majority of people shut out and struggling to survive. There is no kinder, gentler atom bomb. There is no world where these tools aren’t a global catastrophe that will ultimately cost billions of people their lives.
So the problem is everything moving to the cloud and that cloud being mostly controlled by a small number of big players, right? I still don’t know what the solution to that should be. In my mind, we can either assume that they’ll successfully take over and we give up (so why not make our lives easier while we’re here?), or we fight back while we still have access to the technology.
Make AI unusable. Keep Data Centers rare, keep tokens expensive, make sure no one is willing to spend on AI, and let the megacorps collapse under the weight of debt they can’t pay off. When they can’t fight back, start passing laws that outlaw private AI development and nationalize the major models. Start fighting for anti-trust laws with teeth, reduce corporate copyright entitlements while enforcing human IP rights, and void the whole concept of software patents.
Or die. Dying is always an option. I think we’d all rather avoid that one though.
I think we should clarify what we mean by AI first. I’m under the impression that you’re talking about generative models. I don’t think it makes sense to ban all form of automation. But if you really do mean all automations, then I’d like to hear the reasoning for that.
If you want to drop the marketing buzzwords and be specific, yes, LLMs, image generation/graphic design and other generative models, but also machine assisted data analytics need to be heavily regulated and taxed into a box. Military and law enforcement applications need to be strictly state owned, automated surveillance needs to be put on a bonfire entirely. There’s a lot of different classes and categories of “AI” that need to be completely re-thought, but you are correct, I am not entirely against all categories of automation.
It seems we are in agreement on most of this. These generative models aren’t going to fill your pantry or do your chores. They can help through world models, but I’m hoping to get things working without that.
Of course, then you run into the problem of access vs cost. Running the AI requires resources, those resources have market value so you need to make money. Setting aside the fact that by definition you are making sure those resources aren’t being expended on something else potentially more conducive to life, you now have to limit access and/or features to those who can pay, ie. not the people who are struggling the most and could most use the help. Given the price of processing power alone right now, let alone the rest of it, how do you propose these tools become available to the places they can do the most good?
There’s also the problem that most engineers scratch their own itch, not other people’s. The tools you’re developing almost always tend to focus on things that are personal struggles. Your tools are often going to be solving problems the typical person will never have, or at the very least, are so low priority compared to everything else your tools are inconsequential. What you need is people, and feedback mechanisms that can help identify opportunities… you know like a bureaucracy.
I agree in principle with you. Technology in every bodies hand is benefitting everybody, while under control of a few only benefits them.
But with LLM as they are currently and in my eyes in the near future as well? I’m not seeing it possible for the average person to make their own model with available hardware. Big data centres are needed and they burn a lot of resources for creating a video that turns a cat into a human, help genociders argue why they needed to bomb that school or destroy our education systems while simultaneously destroyingpublic trust in media.
When it comes to LLMs and similar generative models, I agree. I’m talking about AI in general though, and reinforcement learning, which is the focus of my work. It’s still very resource intensive and doesn’t work very well, but what technology isn’t in the early stages?