> I'm being asked to do things that have never been done before -- and I love this, it is the ultimate puzzle -- but how the fuck do you expect me to give accurate ETAs and to do this 3-5 times a year and launch ground breaking work with some 2080Ti nodes and maybe one A100 node?
If it is work that some industry is highly interested in, and this industry has deep pockets, it is very hard to compete with it in academia. Look for more fundamental questions to research (there exist insanely many, just don't follow the hype) that can be investigated with a lot less ressources.
Fwiw, this is the route most academics take, but it is still not easy to publish in this domain. I've seen workshop papers get rejected from very clearly academic work because they don't use real world or large datasets. I've seen theory papers get rejected because they don't have experiments. Yes, read that last one again. And mind you, these big companies also do a lot of theory and foundational work too. The process is just noisy.
If it is work that some industry is highly interested in, and this industry has deep pockets, it is very hard to compete with it in academia. Look for more fundamental questions to research (there exist insanely many, just don't follow the hype) that can be investigated with a lot less ressources.