Research interests

Large ML model interpretability

Large machine learning models, such as GPT-3 and Gato, are as capable as they are indecipherable. By dissecting them and visualizing how its parts relate to its output, I hope to help us gain better understanding of why they produce the useful results we see.

Understanding the mechanics of large models would help us develop more robust AI, and it would also give us a safer path toward building AGI.

Human-AI collaboration

The effect of advanced technology is never isolated to that technology itself. Only when we interact with it would the loop be complete.

Now that deep learning has enabled new categories of AI capability, I want to explore the new relationships this form between humans and machines. This may take the form of games involving AI agents; productivity tools (e.g. self-organizing knowledge management tool); and other experiments that place AI in a non-traditional role. I also hope that this research can give us insights toward how to control and work with superintelligent AI in the future.

Notes mentioning this note

There are no notes linking to this note.