Three Papers Accepted at ICLR 2026Donato CrisostomiLast updated on Feb 7, 2026Three papers accepted at ICLR 2026! 🇧🇷 See you in Rio!MASS: MoErging through Adaptive Subspace SelectionLanguage Models are Injective and Hence InvertibleImplicit Inversion Turns CLIP into a DecoderDonato CrisostomiIncoming AI Safety Fellow @ AnthropicMy research interests revolve around artificial intelligence, in particular model merging and representational alignment.