Consulting on AI policy (volunteer)

Worked on SB 1047, a landmark AI regulation bill in California which advanced the conversation around AI safety. The Center for AI Safety Action Fund was one of the bill's co-sponsors. The bill passed both houses of the California legislature before being vetoed by the Governor.Primarily worked on amendments to the legislation in collaboration with Senator Wiener’s legislative team, committee staff, legislative counsel, external legal experts, and outside stakeholders.Comments on the amendments from both supporters and opponents of the legislation (comments do not imply any endorsement):“In our assessment the new SB 1047 is substantially improved, to the point where we believe its benefits likely outweigh its costs.” – Dario Amodei, CEO of Anthropic“All these changes are great. This has shaken out into a very reasonable bill.” – Sam Hammond, Foundation for American Innovation“I now think SB 1047 is not a bad bill.” – Charles Foster, Finetune“Some of these revisions have made the bill less bad.” – Andrew Ng“Many changes, including some that were just announced, I see as strict improvements.” – Zvi Mowshowitz“We appreciate the Senator’s collaboration, along with many of the positive changes made throughout the legislative process.” - Mozilla"Overall, SB 1047 is considerably better than it was before—they weakened or clarified many ofthe key regulations." - Representative Zoe Lofgren

Worked on the Foundational Research Grants team. I helped with grantmaking, writing technical explainers for policymakers, and projects involving agentic AI systems. My fellowship placement was funded by the Horizon Institute for Public Service.

- Worked on a research project involving Trojan attacks in large language models. Co write the ML safety newsletter.- Answered technical questions for philosophers in the CAIS Philosophy Fellowship. Assisted with policy engagement. - Co authored An Overview of Catastrophic AI Risks paper and contributed to the forthcoming AI Ethics, Safety, and Society textbook.

• Collaborated with the CEO, Dan Hendrycks, and supported growth of the organization from 1 employee to 10• Co-wrote the organization’s 150 page research agenda and contributed to machine learning research projects in machine ethics and legal natural language processing (see publications below).• Authored blog posts, grant proposals, and other written content for the organization.• Founded and directly managed the online Machine Learning Safety Scholars summer program, involving ~65 college students from 20 countries. Hired and managed 8 teaching assistants.• Provided feedback on language for the $20 million NSF Safe Learning-Enabled Systems program.

• Ran more than 2000 experiments using the lab’s computer cluster in support of one of Professor Stuart Russell’s PhD students. • Worked on a research project that aimed to help reinforcement learning agents avoid unsafe states by following human instructions.

• Interned at a startup working on mobile authentication based on gait (walking patterns).• Cleaned and integrated an external academic dataset into the startup’s internal testing data.• Rewrote a tool used to graphically compare the performance of trained machine learning models.

Worked on improving a convolutional neural network to infer strength of the green wavelength in satellite imagery from other sensor readings. This tool made it possible to produce color satellite images.

Analyzed web event data to recommend optimizations to Kiva's lend page. Developed a system to automatically classify common types of support tickets.

Contracted to mentor current students on content for the Data Analyst Nanodegree and Self-Driving Car Nanodegree.Answered over 2000 student questions.

Wrote a program to distribute some of Kiva’s data publically to interested researchers; gained extensive knowledge of Kiva’s database and data warehouse.Used gradient boosting and a long short-term memory network to make predictions about future loans on Kiva.

Developed software for Looma, an affordable and low power consuming audio-visual technology device access to educational content to Nepalese schools.Worked on a team tasked with developing a lightweight, offline slideshow tool that would allow teachers to make customizable presentations using preloaded images.