Tilde Research's Aurora optimizer fixes hidden 'neuron death' problem in Muon

Tilde Research released Aurora, a leverage-aware optimizer that addresses a structural flaw in the Muon optimizer which permanently kills MLP neurons during training. The release was paired with a 1.1B-parameter pretraining experiment validating the fix. The publication is one of the more substantive optimizer-research drops of the quarter, with implications for any lab using Muon variants in frontier-model pretraining.