Resources: Reinforcement Learning

Exercises: Reinforcement Learning

Resources: Capstone Project

Exercises: Capstone Project

Resources: Transformers and Mechanistic Interpretability

Exercises: Transformers and Mechanistic Interpretability

Resources: Fundamentals

Exercises: Fundamentals

Resources: Prerequisites

Exercises: Prerequisites

The material on this page covers the next 5 days of the curriculum. It will cover transformers (what they are, how they are trained, how they are used to generate output) as well as mechanistic interpretability (what it is, what are some of the most important results in the field so far, why it might be important for alignment).

Each exercise will have a difficulty and importance rating out of 5, as well as an estimated maximum time you should spend on these exercises and sometimes a short annotation. You should interpret the ratings & time estimates relatively (e.g. if you find yourself spending about 50% longer on the exercises than the time estimates, adjust accordingly). Please do skip exercises or look at solutions if you don't feel like they're important enough to be worth doing, and you'd rather get to the good stuff! Additionally, remember that the recommended time for each exercise is just the time it takes to complete the exercise, it doesn't include the time for the recommended readings. 

The first three days are structured group exercises, the final two days will allow you to explore these subjects more independently with some unstructured exercises, where you can choose which subject to pursue. 

All of the material needed to complete these exercises, including readings, instructions, and hints, are available on this section's streamlit page. Please follow along with the streamlit page to complete the exercises, then submit your work here. 

By the end of the unit, you should be able to:
This week will focus on teaching you to: 
\- Build your own transformer from scratch, and use it to sample autoregressive output
\- Use the TransformerLens library developed by Neel Nanda to locate induction heads in a 2-layer model
\- Find a circuit for indirect object identification in GPT-2 small (optional)
\- Interpet model trained on toy tasks, e.g. classification of bracket strings, or modular arithmetic (optional)
\- Replicate Anthropic's results on superposition (optional)


Submit a link to your exercises, through a GitHub repo, or your Collab notebooks, or through any other form you'd like. 



Transformers: Building, Training, and Sampling

Submit a link to your exercises, through a GitHub repo, or your Collab notebooks, or through any other form you'd like. 


ARENA

Resources: Transformers and Mechanistic Interpretability

Resources