I would like to explore how interpretability is claimed in frontier models and how we can verify these claims and operate them through policy. What I mean is I want to investigate what evidence based standards should we require from these labs that claim interpretability, transparency safety etc.

The recent hype influenced me as AI is being treated as some super intelligence going rogue whereas the real story is the incompetence of the lab not creating a strong enough sandbox. Enforcing accountability on these labs prevents abuse of cyber security threats stemming from AI agent swarms. AI is here to stay but in the wrong hands it is dangerous. It is an input output tool at its core and a whole new architecture would need to be invented to circumvent this to produce even a notion of AGI. Understanding these tools and keeping everyone safe while also keeping the industry competition is important for us to coexist with AI. Safety agencies need data driven decision making to govern artificial intelligence. The essence of data science.