We’re freeing the 774 million parameter GPT-2 language mannequin after the discharge of our small 124M mannequin in February, staged free up of our medium 355M mannequin in Might, and next analysis with companions and the AI neighborhood into the mannequin’s possible for misuse and societal get advantages. We’re additionally freeing an open-source felony settlement to make it more uncomplicated for organizations to start up model-sharing partnerships with every different, and are publishing a technical document about our revel in in coordinating with the broader AI analysis neighborhood on e-newsletter norms.
