Also fyi, ours run it on their Raspi 400 via dosbox, they love it too. The DOS games library is full of educational-ish gems (they really like Lemmings and The Settlers, too).
In nvfp4, it's about 300 gigs once you offload n-grams, 491 without offloading, you can run it pretty well on 4x DGX Sparks, which last I checked was about $20k. So, it's definitely runnable.
Or you can just use any of the neoclouds' shared hosting. The thing for them to be freaked out is that these models are getting good enough very quickly, and all the shared hosting providers can run them for a tiny fraction of what the frontier model companies charge.
To follow up on this, US PhD programs are not generally packed with US citizens, they're frequently the cream skimmed from the rest of the world. For many subspecialties, there's just not going to be enough US citizens to do all the cutting edge work that could/should be done.
We should clip a green card to every PhD we grant.
I mean, we also break a lot of PhD programs because due to glut of foreign workers, the work environment is hell. I've known several citizens who dropped out because it was 996 type environment and since they didn't need to visa to remain in the US, they could just move on.
Clipping a green card to every PhD is probably political non starter. As American, I'd be very leery of such a scheme.
What does it mean to "break a lot of PhD programs"? Research is competitive, and hard work. Yeah, there are easier options, but in the long run, the countries with the people willing to work their butts off academically are going to outperform ours if we can't get their best to come here and work their butts off here.
>Clipping a green card to every PhD is probably political non starter. As American, I'd be very leery of such a scheme.
There is actually bipartisan support for trying to make it easier for PhDs to stay - both parties have repeatedly introduced legislation to help with that - but it seems to keep getting derailed by arguments about other aspects of immigration reform.
"break a lot of PhD programs" meaning PhD look less attractive day by day for citizens while relying on foreign money to stay afloat because funding is no longer available.
Maybe you disagree and I never considered getting a PhD but from what I understand, the workload has gone from hard to insane. If you are saying PhD only work with 18 hour days, 7 days a week, alright I guess.
>There is actually bipartisan support for trying to make it easier for PhDs to stay - both parties have repeatedly introduced legislation to help with that - but it seems to keep getting derailed by arguments about other aspects of immigration reform.
BTW, I've seen the legislation, it gets derailed a ton because many voters, myself included, believe it would just backdoor into Universities happily taking money for people to get PhD in low demand subjects to get the green card.
I think those proposals are usually scoped to science/tech PhDs, I guess to avoid that scenario, but I only have a headline-level knowledge of the bills. I don't know if it's that insane at most labs? But my first/second hand knowledge is a decade out of date at this point.
This seems to imply that training will ever be done? But yeah, I think the idea is that the appetite for thinking-on-tap will be enormous.
Even for smaller models, I think they’ve found that training an enormous, inefficient model and then distilling it internally to something much more efficient to serve is the way to go.
There's way more anxiety here than is necessary. If you don't want to optimize, then it's pretty simple right now, just use Opus 5.5 or GPT 6.1 Sol. If you want to quickly test out the alternatives and see if the decreased ability in long horizon tasks is offset by the decrease in costs, toss a few bucks into your favorite neocloud (Fireworks was fine last time I tried them) and give GLM 5.3 and Deepseek v4.1 a spin, and see if they're good enough. OhMyPi makes it trivial to swap models without swapping your harness.
Everyone has a great researcher piped to their desk now. You can have it follow the Twitter zeitgeist for you, you don't have to do it yourself.
It's running a normal central home heat pump/air conditioner for one hour, or eight hours of playing on a gaming PC. Yeah, the propaganda around datacenter resource usage has massively outrun the truth, which makes me think there are some very interested parties pushing behind the scenes.
I don't discount the possibility, but I think part of it is that no one was interested in knowing how many resources normal computing costs. Any one of us who's worked on a video platform or cloud storage providers already knew how the sausage was made, but as long as it provided unlimited videos and photo storage, the public didn't care.
Though I do suspect that golf course owners in Arizona and alfalfa farmers in California must be ecstatic.
reply