Homepage

Hi, I’m Theodore. I’m an AI Safety researcher at the Beneficial AI Foundation.

Projects

I discovered a symlink exploit in Vast.ai and was awarded a bug bounty
tiny_stories_sae: Trains a sparse autoencoder on a 33 million parameter language model
tiny_stories_rl: Uses reinforcement learning to encourage roneneldan/TinyStories-33M to generate stories with alliteration
Looking for feature absorption automatically: My project from when I was at Pivotal Research
CAIMAN-ASR: Pushing the Boundaries of Low-Latency Streaming Speech Recognition: Blog post from the machine learning team at Myrtle.ai (my former employer)

A calculator that allows you to update multistep computations in real time (code)
A primality checker where I got Elm to talk to wasm (code)
An Elm game I coded up for my brother (code)
A command-line tool for keeping a running total in different buckets