forethought's new post is pretty GSAI pilled (not in those words)

[Quinn Dougherty]

https://www.forethought.org/research/ai-tools-for-existential-security

Just needs a little nudging and it's totally on board with GSAI/FV:

Automating research into AI safety, such as theoretical alignment, mechanistic interpretability, or AI control	If these areas are automated early enough relative to the automation of research into AI capabilities, safety techniques might keep up with increasingly complex systems. This could make the difference in whether we lose control of the world to misaligned power-seeking AI systems.3
AI tools for greatly improving information security	Strong information security could limit the proliferation of powerful AI models, which could facilitate coordinating not to race forwards as fast as possible. It could also reduce the risk of rogue models self-exfiltrating.