Safety considerations for online generative modeling — AI Alignment Forum