no stance on this piece and model welfare, but I do genuinely love that there’s a lab willing to commit resources to these experiments https://www.anthropic.com/research/end-subset-conversations