AI Safety Researcher based out of India, broadly interested in Scalable Oversight, Natural Abstractions and Scientist AI. My answer to Hamming’s Question is designing truth-seeking agents which discover the causal abstractions in an open-ended world in an external, interpretable and intervenable way.

I’m currently working on causally modeling composition of goals in open-ended environments, delving in shard theory, open-endedness, and causal incentives. I also try to support the development of an AI Safety ecosystem in India. I’ve previously shipped two products, completed some fellowships, recently graduated, and attended FAEB — the Finnish iteration of ARENA. These days, I spend most of my time thinking about the threat models arising from Open-Endedness and red-teaming research agendas to work with Scalable Oversight.

Website: diksha-shrivastava13.github.io

Note: I’m not active on social media — the best way to reach me is by email.