Yeah, I said myself I didn't think it would work (outside of the context of somewhere like China). I'm trying to redirect the discussion from simply obsessing about it to something more material. The Luddites accomplished jack shit, and we're headed down the same path right now without specific demands or plans.
I also explicitly said this particular application is a waste of energy and I'm not defending it at all. What I'm saying is, there ARE legit use cases, which makes the issue more slippery.
There's never going to be actual progress with this technology because of capitalism
You realize China is researching this technology too? They're putting billions into it. And also it has progressed wildly in the last couple of years in the US (which is not good honestly for them to be in the lead). This shit is happening if we like it or acknowledge it or not.
But you're here defending it. China may be researching this tech, but as far as I know, they aren't deploying it willy-nilly for any techbro to just swoop in and start making spam bots.
A major element in this "AI" fad is that it's not actually artificial intelligence. It's a label that gets applied for marketing purposes. If we want to get into material analysis, we need to look at what's actually happening and not what we think might happen in the future. And what's actually happening is already existing tech is getting a rebrand to justify using up more resources.
I think we agree on several fundamental points, such as the inefficiency of current AI applications and the problematic capitalist motivations driving much of its development.
That being said, completely dismissing the technology because of its current misuse may prevent us from exploring and identifying genuinely valuable applications. For instance, in research and industrial settings, AI has shown promise in optimizing processes, reducing waste, and performing complex data analysis. These are areas where the potential benefits might rightly justify the resource expenditure. China's approach, which includes heavy state involvement and regulation, might offer some insights.
Let's focus on creating specific and actionable plans that address the root issues rather than being caught in the binary of defense vs. criticism.
And again, here you are defending things lol. Nobody is dismissing the technological advancements of the future. What we're talking about is what's happening right now. And right now the """"AI"""" tech doesn't justify its resource consumption.
You want there to be "specific and actionable plans" but so far you haven't said any of those. What algorithms are you hoping to implement that reduce carbon footprints? Which models have real world applications that humans or existing tech don't already do?
All you've done is "Okay it sucks now but if we keep doing this same thing it's gonna get better just trust me China will solve it and then American companies will totally be left behind (please ignore solar panels and electric cars and how the US just puts a tariff up or actively punishes green energy)."
I'm going to ignore the sort of hostile tone you're taking with me for the sake of an earnest discussion.
But yeah, to get more specific... for example, we could try to advocate for a law that companies like X and Facebook that have consumer-facing image generation to at least fund it with renewables as opposed to just burning natural gas generators. A blanket law that new data centers can only use nuclear or renewables period, regardless of application. I know you'll just say that's "lib-shit" but what do you want me to do say we should start kidnapping politicians or something? It's hard to target this stuff since you can do it at home on a gaming GPU, so I think going after larger providers like social media companies makes sense.
Also energy usage HAS gone down significantly for a lot of these inference requests since some of the original findings. We're finding out that a mixture of experts (small models trained on specific tasks that can be routed to by a top level model that specializes in routing) can rival the output of previously very expensive foundational models for a fraction of the cost. There's new hardware coming out that makes the matrix math more performant and I think it's not unreasonable to say we should put public funds behind such research for the sake of the planet.
And I'm not just saying "it sucks now". That's my whole point... it's more nuanced than that. You're seeing a lot of the most wasteful applications of it on social media because you're not seeing how individuals and businesses aren't necessarily publishing their workflows that use this tech.
To give an example, video generation, something we didn't have even two years ago worth a shit, is getting better every week. If you're a filmmaker and you're in the field filming something, but realize you need a few seconds of establishing shot, or maybe you notice a continuity fuckup or something, you'd normally have to go back on site and reshoot more footage. That includes getting a crew into one or more vehicles, driving somewhere, and using probably lots more fossil fuels. With transformer based video tech, you can use generative fill or just generate a transition or clothing change for those few seconds and save loads of time and money. It's not all black and white is all I'm saying.
What I'm advocating for is a more nuanced conversation that goes beyond the binary of "it's all bad" vs. "it's all good." Expressing frustration and concerns is absolutely valid.
Yeah, I said myself I didn't think it would work (outside of the context of somewhere like China). I'm trying to redirect the discussion from simply obsessing about it to something more material. The Luddites accomplished jack shit, and we're headed down the same path right now without specific demands or plans.
I also explicitly said this particular application is a waste of energy and I'm not defending it at all. What I'm saying is, there ARE legit use cases, which makes the issue more slippery.
You realize China is researching this technology too? They're putting billions into it. And also it has progressed wildly in the last couple of years in the US (which is not good honestly for them to be in the lead). This shit is happening if we like it or acknowledge it or not.
But you're here defending it. China may be researching this tech, but as far as I know, they aren't deploying it willy-nilly for any techbro to just swoop in and start making spam bots.
A major element in this "AI" fad is that it's not actually artificial intelligence. It's a label that gets applied for marketing purposes. If we want to get into material analysis, we need to look at what's actually happening and not what we think might happen in the future. And what's actually happening is already existing tech is getting a rebrand to justify using up more resources.
I think we agree on several fundamental points, such as the inefficiency of current AI applications and the problematic capitalist motivations driving much of its development.
That being said, completely dismissing the technology because of its current misuse may prevent us from exploring and identifying genuinely valuable applications. For instance, in research and industrial settings, AI has shown promise in optimizing processes, reducing waste, and performing complex data analysis. These are areas where the potential benefits might rightly justify the resource expenditure. China's approach, which includes heavy state involvement and regulation, might offer some insights.
Let's focus on creating specific and actionable plans that address the root issues rather than being caught in the binary of defense vs. criticism.
And again, here you are defending things lol. Nobody is dismissing the technological advancements of the future. What we're talking about is what's happening right now. And right now the """"AI"""" tech doesn't justify its resource consumption.
You want there to be "specific and actionable plans" but so far you haven't said any of those. What algorithms are you hoping to implement that reduce carbon footprints? Which models have real world applications that humans or existing tech don't already do?
All you've done is "Okay it sucks now but if we keep doing this same thing it's gonna get better just trust me China will solve it and then American companies will totally be left behind (please ignore solar panels and electric cars and how the US just puts a tariff up or actively punishes green energy)."
I'm going to ignore the sort of hostile tone you're taking with me for the sake of an earnest discussion.
But yeah, to get more specific... for example, we could try to advocate for a law that companies like X and Facebook that have consumer-facing image generation to at least fund it with renewables as opposed to just burning natural gas generators. A blanket law that new data centers can only use nuclear or renewables period, regardless of application. I know you'll just say that's "lib-shit" but what do you want me to do say we should start kidnapping politicians or something? It's hard to target this stuff since you can do it at home on a gaming GPU, so I think going after larger providers like social media companies makes sense.
Also energy usage HAS gone down significantly for a lot of these inference requests since some of the original findings. We're finding out that a mixture of experts (small models trained on specific tasks that can be routed to by a top level model that specializes in routing) can rival the output of previously very expensive foundational models for a fraction of the cost. There's new hardware coming out that makes the matrix math more performant and I think it's not unreasonable to say we should put public funds behind such research for the sake of the planet.
And I'm not just saying "it sucks now". That's my whole point... it's more nuanced than that. You're seeing a lot of the most wasteful applications of it on social media because you're not seeing how individuals and businesses aren't necessarily publishing their workflows that use this tech.
To give an example, video generation, something we didn't have even two years ago worth a shit, is getting better every week. If you're a filmmaker and you're in the field filming something, but realize you need a few seconds of establishing shot, or maybe you notice a continuity fuckup or something, you'd normally have to go back on site and reshoot more footage. That includes getting a crew into one or more vehicles, driving somewhere, and using probably lots more fossil fuels. With transformer based video tech, you can use generative fill or just generate a transition or clothing change for those few seconds and save loads of time and money. It's not all black and white is all I'm saying.
deleted by creator
What I'm advocating for is a more nuanced conversation that goes beyond the binary of "it's all bad" vs. "it's all good." Expressing frustration and concerns is absolutely valid.
deleted by creator
disengage