PatentsIBM patent application trains LLM guardrails on production dataThe filing clusters detected risk cases and retrains the filter. If quality drops, model distillation is meant to counter the decline.September 13, 2026·2 min read▲—▼
AI & ChipsNASA and IBM Open Their Lunar AI — but Pretraining Code Is MissingThe model, weights and datasets are public. Performance varies by task, while the pretraining code is not included.September 11, 2026·2 min read▲—▼