@1a3orn interesting to think about how the design of posttraining forms the model’s traits and personalities. it wouldnt be surprising to me if differences in model characters are also significantly downstream of tasks and envs designs between labs.