>but they are extremely useful, and there are no sane arguments against that.

I'm not arguing usefulness. We have many useful tools, programs, and institutions out there. I'm sure some parts of this community directly contributed to such tech.

I'm arguing cost. Costs raise too much (and as you mention, they don't give a flying fuck about those who can't afford it) and we throw LLMs into the pile of "useful yet inaccessible" tools that increases the disparity of the passing and failing kids. So we're back to square one.

>by whom

Shareholders. All the frontier models are going public, and we've seen this playbook being run. They compete until competitors peter out, or they collude to make sure prices all increase in coordination. It's happening now with memory, happened years back with the streaming wars. I see no reason why LLMs are different.

>fighting against them being used in education is as absurd as fighting against computers and Internet had been.

I'm simply asking two questions long term

1. Who foots the bill?

2. Who gets the benefit.

From what I see you haven't answered #2. And for #1 you're just hoping competition keeps prices low despite it being well established that LLMs are not sustainable at current costs. Something's going to give, and history tells me it's not going to be the privileged families.

▲

vlyan 5 hours ago | parent [-]

>Shareholders. All the frontier models are going public,

what I meant is that there isn't a singular entity capable of doing such a rag pull. even if/when closedai and misanthropic do it, open-weights models obviously won't follow.

>Who foots the bill?

primary schoolers don't need $50/1M SOTA. dirt cheap "flash" tier models are more than enough, and even models that can run on consumer hardware are enough for great many things if you know their limitations.

>Who gets the benefit.

students who use it.

again, a LLM can essentially act as a fairly proficient and infinitely patient tutor on every subject, and you have unlimited access to it. that's the application I see for them in education. it would've been amazing to me to have that when I was in school.

>despite it being well established that LLMs are not sustainable at current costs

not a single third-party provider on openrouter is subsidizing their tokens.

	▲	johnnyanmac 4 hours ago \| parent [-]
		>dirt cheap "flash" tier models are more than enough And do you trust companies to keep subsidizing the "cheap" models? The reason for the memory crisis isn't due to consumer DRAM but affects it nonetheless. Remember, the kinds of families being talked about here are on districts that apparently cannot afford basic learning supplies and it's a "fantasy" to get them more funding. Who's footing the bill? >and you have unlimited access to it. You keep handwaving the costs but not providing good argument for who's covering it. There's no such thing as a free (or in this case, unlimited) lunch. A disadvantaged family may not be able to bear yet another subscription (and again, that train has already left on subscription based syst3ms), a school district would need to fund this for thousands of students, likely funded by property taxes. >not a single third-party provider on openrouter is subsidizing their tokens. Yes. Their earnings reports reflect this. The companies are running defecits to fund this. They are the ones subsidizing it. We all know that will not last forever in this post ZIRP economy. Proponents for AI love to argue about the capabilities in 3-5 years whenever flaws in AI models arise, but seem sheepish to talk about the costs and monetization models. But the latter is the real important part when talking on the scale of an entire education system. I think we can all agree that the products that win such educational bids are rarely the highest quality ones.