ML safety research in interpretability and model sharing, game-playing language models, and critiques of AGI risk research