LLMs seem pretty smart, but they suck at seemingly basic tasks. Why?

The train-test split does not work for classification tasks at the frontier of LLM capability.

Prompts look like documentation, but they're deceptively complex. How can we document the choices that went into them?

An exploration of historical time through three lenses and one chess player.

I spent high school staring out the window and waiting for lunch time. I felt like there was nothing I could change about school. This report is the culmination of my nationwide research on what students can change.

I am traveling across the country researching how trust is built, working to empower student voices.

Pakistan should be the flagship success of China’s economic aid and investment programs abroad. Instead, it casts doubts on the viability of a Chinese economic world order.

"Nothing escapes his keen analytical gaze... He has immersed himself in an incredible volume of seemingly-dry historical sources and not only wrestled them into a clear and well-structured piece of writing, but also condensed them into a pithy but substantial interpretive formula... articulated with flair."  
– Natasha Wheatley, Princeton History Professor

I created a haiku-generating algorithm and argued that the future of great writing isn't human, computer, or computer-trying-to-emulate-human, it's human-computer collaboration.

Further Writing