How much more accurate a simulation could you run if you had OpenAI's training budget and supercomputer? (≈$100M)

looping__lui 2 days ago | parent | next [-]

As a fan of a sport that greatly relies on accurate weather forecasts with an interest in weather models my impression is that more supercomputer budget isn’t necessarily the issue (eg., more granular models can be less accurate) but the lack of data on the initial state is… Things like snow on mountains, soil conditions, vegetation, soil humidity and simply the problem of just having relatively few data-points makes stuff tricky… That stuff changes with season. I have seen some magic improvements coming from small shop companies tweaking the models here and there a bit - massively impressed. Now, running these simulations even further into the future - I have my doubt much useful prediction is gonna come out.

▲

alwa 2 days ago | parent [-]

To your knowledge is there anybody working on a “littering the globe with huge quantity of cheap self-contained sensors” kind of project? Or is that the kind of work best done from the air or from orbit?

	▲	looping__lui 2 days ago \| parent [-]
		I am no authority to speak about this - but I believe expanding sensors and improving on data gathering is a continuous and on-going effort (e.g., weather station on ships, airplanes, stations deployed on sea etc.). Better weather satellites are also a thing. But then there is just so much more that could be done and all these data-points are still unevenly distributed (e.g., planes and ships follow routes). Weather models are improving quite a bit “behind the scenes” but again - I think we are far away from getting anything reliable for several days into the future; not to speak months or years. Here is a nice interview highlighting a couple of improvements people are driving for a niche sport weather forecast: https://magazine.weglide.org/skysight-interview-matthew-scut...

▲

jjtheblunt 2 days ago | parent | prev [-]

what would you do with a more accurate simulation's data, besides have burned a great amount of money and (earth warming?) power to compute it?