Projects - AI Alignment

Decision Transformer Interpretability

In this project I show that A Mathematical Framework for Transformer Circuits, a method used to understand how Large Language Models work, can be extended to Decision Transformers, a reinforcement learning method designed to produced AI which can simulate players of an arbitrary quality.

You can find an initial write up here. I’m currently working on this project and hope to publish more soon, including analysis of circuits in decision transforms which involve memory/language and variable goals.

ARENA

ARENA (Alignment Research ENgineering Accelerator) was 9 week research engineering accelerator I participated in, during which we completed a series of increasingly sophisticated projects, culminating with my capstone on Decision Transformer Interpretability. These projects included:

• Implementing GPT2 and training it to generate Shakespeare text.
• FineTuning BERT
• Implementing our own optimizers from scratch
• Implementing DQN and PPO
• Reverse Engineering small transformers performing algorithmic tasks

Projects - Computational Biology

I studied computational biology at university and have worked on a number of projects in this area. These include:

• Mass Dynamics. I worked as a data scientist for 2 years producing 1 first author paper as well as several R packages. I also developed a novel algorithm for solving the protein-inference problem (see poster presentation below).
• Buckle Protein Engineering Lab. I was a research assistant at the Buckle Protein engineering lab. I co-first-authored 1 paper on structural dynamics of immune proteins.