> To improve the safety of the model, we filtered the data for harmful content in pre-training, especially around hazardous biosecurity knowledge, by reusing th

> To improve the safety of the model, we filtered the data for harmful content in pre-training, especially around hazardous biosecurity knowledge, by reusing the CBRN pre-training filters from GPT-4o. Our model has a knowledge cutoff of June 2024. This would be a great "AGI" test. See if it can derive biohazards from first principles