Policies

Lectures

Lectures are on Mondays, 6–8 PM in Wheeler 120. Each meeting includes a lecture portion followed by in-class paper reading and small-group discussion. Each week’s material includes mandatory readings (30–60 min per week) to complete before class, plus optional supplementary readings.

Attendance

Attendance is mandatory. Students may have at most 2 unexcused absences from weekly meetings. If you cannot make it to a lecture, please email the instructors in advance. Staff will be accommodating regarding reasons for absences.

Weekly Reflections

Reflections are based on the assigned readings and must be submitted before each class. Students will answer specific questions about the material and provide their own insights — points of interest, connections to other ideas, and areas of agreement or disagreement. Completing these reflections demonstrates engagement with the material and prepares students for in-class discussions.

Reflections and readings together should take approximately 90 minutes per week. Two reflections may be missed without a grade penalty.

Paper Presentations

Students will be required to summarize one of the papers they read in-class in a brief presentation, enhancing their understanding of the material.

Signup sheet and slide template will be linked here — EDIT docs/policies.md.

Final Project

Students will complete a coding project focused on replicating results from technical AI/ML safety papers. Students can earn extra credit by creatively extending the ideas presented in the papers. The project is designed to take around 8 hours to complete, may be completed in groups of 1–3 (groups must be approved beforehand), and should be accompanied by a conference/workshop-style research paper or a video presentation. We will provide the necessary resources (API keys, GPUs, etc.) for the project.

Grading

Students will receive credit for:

Attendance at weekly meetings — 40%
Completion of core readings and reflections — 30%
Final project — 30%

To pass, students must:

Have at most 2 unexcused absences from weekly meetings
Complete the Final Project with a passing grade (at least 70%)

Prerequisites

This course assumes fluency with linear algebra, multivariable calculus, and basic programming at a level equivalent to MATH 54 and CS 61A, though these prerequisites will not be formally enforced. Since the class moves quickly through the material and involves significant ML paper reading, familiarity with machine learning is strongly recommended. Prior familiarity with ideas and material in the field of AI Safety is not required. Applicants will be reviewed holistically.

Academic Integrity & LLM Use

Reports and reflections must be your own original work. You may reference outside sources, but cite them. Cite any non-trivial code (more than a few lines) from outside resources.

Do not use LLMs to write your code unless you have explicit permission from the instructors. Explicit permission will be granted on a select case-by-case basis, for final projects that go above and beyond the course material (e.g. original research). You may use LLMs to discuss the material and receive conceptual guidance.

All reflections must be done individually. Final projects may be completed in groups of 1–3 (groups must be approved beforehand).

Accommodations

If you require course accommodations due to a physical, emotional, or learning disability, contact UC Berkeley’s Disabled Students’ Program (DSP). Notify the instructors via email of the accommodations you would like to use. You must have a Letter of Accommodation on file with UC Berkeley to have accommodations made in the course.

UC Berkeley has activated the ALLY tool for this course. You will be able to download reading materials in your preferred format (PDF, HTML, EPUB, MP3). For more information, see the alternative formats link in bCourses or watch the “Ally in bCourses” video.

Inclusion

The instructors are committed to creating an inclusive learning environment where everyone, regardless of background, feels comfortable sharing their ideas and learning from each other. If you have a concern about the class, or want to give feedback, feel free to email the instructors or fill out our anonymous feedback form.