Y'all I just got rageblocked and accused of AI psychosis because I was like "maybe you should understand the technology"
-
Y'all I just got rageblocked and accused of AI psychosis because I was like "maybe you should understand the technology"
If this is how the anti-LLM camp are gonna be...yikes.
Are the people at the helms of the frontier labs probably bad? Yes.
Is continuing to overinvest in this technology to the detriment of local communities and natural resources bad? Yes.
But also...
The open weight models are trained. That compute is used. That water is boiled. That technology will continue to exist. 16 GB OSS models are about the size of 2 hours of 4k streaming. 2 billion inference tokens is also about 16gb. For reference, that is between $10k and $50k of frontier model pricing, depending on output vs input (output is more expensive).
But like, if your beef is with the amount of datacenter power and water consumption from consuming an already trained inference API high definition streaming costs about that. A 4k Netflix habit has the same energy footprint as LLMing for an equivalent time period.
By all means let's pillory what the technology is doing to jobs, communities, and the environment. But uh, if you want total bifurcation of the community, this is how you get that. Which, I guess is probably fine for some folks! But I can tell you that capital-M Money has chosen a side and unless you already have a completely self-sustaining subsistence community (this includes tools, antibiotics, etc), a non-zero percentage of team No LLMs For Anything Ever And Fuck You are going to have to do limited cooperation for resources.
-
Y'all I just got rageblocked and accused of AI psychosis because I was like "maybe you should understand the technology"
If this is how the anti-LLM camp are gonna be...yikes.
Are the people at the helms of the frontier labs probably bad? Yes.
Is continuing to overinvest in this technology to the detriment of local communities and natural resources bad? Yes.
But also...
The open weight models are trained. That compute is used. That water is boiled. That technology will continue to exist. 16 GB OSS models are about the size of 2 hours of 4k streaming. 2 billion inference tokens is also about 16gb. For reference, that is between $10k and $50k of frontier model pricing, depending on output vs input (output is more expensive).
But like, if your beef is with the amount of datacenter power and water consumption from consuming an already trained inference API high definition streaming costs about that. A 4k Netflix habit has the same energy footprint as LLMing for an equivalent time period.
By all means let's pillory what the technology is doing to jobs, communities, and the environment. But uh, if you want total bifurcation of the community, this is how you get that. Which, I guess is probably fine for some folks! But I can tell you that capital-M Money has chosen a side and unless you already have a completely self-sustaining subsistence community (this includes tools, antibiotics, etc), a non-zero percentage of team No LLMs For Anything Ever And Fuck You are going to have to do limited cooperation for resources.
@lapt0r I agree with your sentiment. I've started muting anti-AI bigots, because I too have been attacked for saying something positive about AI.
I was curious if "A 4k Netflix habit has the same energy footprint as LLMing for an equivalent time period." is a supposition, or have you seen a study on this? Would love to share it if you have a source.
-
R relay@relay.infosec.exchange shared this topic
