CS 7xxx, University of Utah

This seminar is for anyone eager to learn mechanistic interpretability, collaboratively with others.

Participants should have strong understanding of the transformer architecture and training stages of LLMs (pretraining, instruction finetuning, alignment, finetuning). Participants should be ready to dig into mathematical details.

Instructors:

• Professor ****Ana Marasović ([email protected]) leads the seminar
• 4 student volunteers give lectures every other week [thank you!!!]

When: Wednesday 11:50am-12:50pm (remote not available)

Where: MEB 3515

Seminar format:

• 1 topic over 2 weeks
• 1st week: A deep dive in the core technical parts.
• 2nd week: A paper discussion of some paper on the same topic.

Requesting a permission code as an undergraduate: To receive a permission code, you must have already taken CS 5340 — Natural Language Processing. Please email the instructor asking for a permission code.

You’d like to participate without enrolling officially? Please come to the seminar.

Grading for those who officially enroll is based on participation.

Schedule (subject to change)