Projects
Dihedral
A learning platform for climbing systems that sharpens the user's problem-solving skills by presenting them with difficult, realistic scenarios.
Chesslet
A chess puzzle game in which the player must capture their own pieces until only one remains.
Boycott Careless AI
A campaign encouraging LLM consumers to limit their model usage to the most responsible frontier labs, as established by the Future of Life Institute's AI Safety Index.
ClimbingBench
A benchmark for evaluating how well LLMs understand climbing systems and how well they can problem-solve.
ConvoPool
A Chrome extension that lets you voluntarily share your AI chat conversations with a public dataset for AI safety research.
PsychoGPT
An interactive demo of Tim Hua's AI-induced psychosis experiment, showcasing how LLMs can reinforce psychosis.
Unverbalized Biases
Interactive demo showing hidden biases that LLMs exhibit but never mention in their reasoning. Based on the paper Biases in the Blind Spot: Detecting What LLMs Fail to Mention by Arcuschin, Chanin, Garriga-Alonso & Camburu (2025).