Amita Shukla

Hello internet friend! I'm Amita, a technical product manager and AI safety researcher with experience building tools for decision-makers.

I currently study economics at MIT, where I focus on how AI use can support human decision-making, and how we can design multi-agent institutions. Humans and AIs are always reward-hacking! How can we design institutions that are resilient to this? You may be interested to see a quick demo of my thinking on reward-hacking and alignment-faking. On the technical side, I've fine-tuned LLMs using RLVR and RLAIF to study the risks of AI-assisted mental health support, contributed to the LinuxBench benchmarking project through an Apart Research AI control sprint, and designed social impact evaluation frameworks for HuggingFace's EvalEval. I was also a reviewer for NeurIPS 2025 and ICLR 2026.

Previously, I was a senior product manager at Analytic Partners, where I ran experiments reaching hundreds of millions of users. I led a corporate decision-making tool that helped Fortune 500 companies choose their pricing, product, marketing, hiring, and supply strategies. If you've shopped at a grocery store, bought fast food, or stayed at a hotel in the last three years, you've probably seen my work!

I'm the Strategy Lead for NYC Women in Machine Learning, a non-profit group that runs workshops, bootcamps, and events for over 6,000 women in the greater NYC area. I am a semi-frequent contributor of fiction and non-fiction writing to Protocolized Magazine, the magazine of the Ethereum Foundation. I'm also an alum of Blue Dot Impact - if you're a current student, feel free to reach out! Find me elsewhere on the internet below.