Import AI 454: Automating alignment research; safety study of a Chinese model; HiFloat4

# AI Safety Research Getting a Speed Boost—But Questions Remain About Cost Researchers are developing tools to automate parts of AI safety work, potentially speeding up the process of making AI systems more trustworthy and reliable. A new study examined safety issues in a Chinese AI model, while another breakthrough improved how efficiently AI systems use computer memory. The underlying question everyone's asking: as AI capabilities accelerate, can we actually keep pace with safety research, or will financial pressure push companies to deploy systems faster than we can properly test them?