Chrysos, G. Beyond Neuron-Level Sparsity: Achieving Faithful and Interpretable LLMs With Mixture of Decoders. AAAI 2026, 40, 39811-39812.