Jansen Nye
CS/ML undergrad working on mechanistic interpretability.
I study how neural networks represent and compute — reverse-engineering the circuits inside transformers rather than treating them as black boxes. This site collects my research write-ups, notes, and side projects.
Research
All research →- Placeholder: Toy Models of Superposition, Replicated Jun 2026 · ongoing
A stand-in research entry showing the writeup format — replace with a real project.
- Placeholder: What Do SAE Features Actually Encode? Mar 2026 · write-up
A second stand-in entry so the research page shows more than one writeup — replace with a real project.
Writing
All writing →Currently
- Research
- Replicating toy models of superposition and probing SAE feature geometry.
- Coursework
- Deep learning theory, convex optimization, and a linear algebra deep-dive.
- Reading
- The Transformer Circuits thread, and whatever Neel Nanda posted this week.
Contact
The fastest way to reach me is email: jansennye989@gmail.com. I'm happy to talk about interpretability, research collaborations, or grad school.