•SiTech AI Team
Open 27B Model Rewrites Its Own Weights Without Training, Paper Says
A new arXiv preprint says an open 27B language model can diagnose its own failures and rewrite its weights through reasoning alone, improving on unseen problems without gradients or training.