Question 1

Do I need a technical background to join?

Accepted Answer

It depends on the program. The AI Policy Fellowship and Policy Reading Group require no prior background in AI or policy. The Technical AI Safety DeCal, however, assumes fluency with linear algebra, multivariable calculus, and basic programming, and strongly recommends prior ML experience at the level of CS 189.

Question 2

What programs does BASIS run?

Accepted Answer

BASIS runs three main programs: a Technical AI Safety DeCal (a course covering ML safety research), an AI Policy Fellowship (exploring governance frameworks through weekly projects and discussions), and an AI Policy Reading Group (open to Berkeley students and the broader community). There is also an ongoing speaker series featuring guests from organizations like OpenAI, DeepMind, and Anthropic.

Question 3

What does the Technical AI Safety DeCal actually cover?

Accepted Answer

The 12-week course covers topics including AGI timelines, RLHF and preference training, adversarial robustness, LLM agent safety, reward hacking, inner misalignment and deception, and interpretability. Students complete 2-3 coding projects replicating results from AI safety papers and give a final project presentation. Guest lectures are held in weeks 11-12.

Question 4

Can members publish real AI safety research?

Accepted Answer

Yes. The research page lists several published papers co-authored by BASIS members, including contributions to 'Humanity's Last Exam,' 'Safetywashing,' and 'Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs,' among others.

Question 5

How is the club structured and how do I stay in the loop?

Accepted Answer

Most club communication happens through Slack. Students can also join a Discord server. Signing up through the interest form adds you to the newsletter and keeps you updated on recruitment, events, and program deadlines.

AI Safety at Berkeley