Reading Groups

Two weekly reading groups, one form for both.

From the second week of November we run two weekly reading groups at Saarland University, one on AI Safety and one on AI Security. Each session is about an hour of reading together and an hour of discussion. You can join one group or both, with a single interest form.

AI Security: Mondays, 17:00–19:00
AI Safety: Tuesdays, 17:00–19:00
E1.7, Room 0.08
From the second week of November

Fill in the interest form

One form, two reading groups

The same interest form is for both reading groups. You can join one group or both: in the form you choose AI Safety, AI Security, or both, and tell us which topics you’d like to read about.

You don’t have to register to take part. Feel free to just drop by. The form takes about five minutes and helps us plan topics and formats around what people are interested in.

Each session is two hours: about an hour of reading together, then an hour of discussion. Sometimes there are snacks.

AI Safety Reading Group

Led by: Mahalakshmi Raveenthiran
When: Tuesdays, 17:00–19:00
Where: E1.7, Room 0.08
Starts: second week of November
Topics may include: risks from advanced AI and threat models, alignment (RLHF, scalable oversight), interpretability, evals and red teaming, AI control, deception and scheming, governance and policy, and societal impacts

AI Security Reading Group

Led by: Esha Afzal
When: Mondays, 17:00–19:00
Where: E1.7, Room 0.08
Starts: second week of November
Topics may include: jailbreaks and prompt injection, adversarial robustness, data poisoning and backdoors, model stealing, privacy attacks, security of LLM agents, AI for cyber offence and defence, and watermarking and deepfake detection

Choosing what we read

Members propose papers, blog posts, reports or videos, and vote on what to read. A proposal includes:

Title and link
Type: paper, blog post, article, report, video, …
Why read it: one or two lines
Reading time: a rough estimate
Prerequisites: what someone needs to know to follow it, and which parts to read beforehand if it can’t be read in about an hour

The reading with the most votes wins, and the organisers break ties. Proposals that aren’t picked go to a backlog and can be voted on again later.

FAQs

Questions you might have.

Is it one form for both reading groups?

Yes. Fill in the interest form once and pick AI Safety, AI Security, or both.

Can I join both reading groups?

Yes. Come to one or both, whatever fits your interests. Pick “Both” in the interest form so we know to plan for you in each group.

Who runs the reading groups?

The AI Safety Reading Group on Tuesdays is led by Mahalakshmi Raveenthiran. The AI Security Reading Group on Mondays is led by Esha Afzal.

Do I have to sign up to come?

No, you can just drop by. The form helps us plan topics and formats, so we’d still appreciate it if you fill it in.

Do I need to prepare before each session?

Usually not. We read together in the first hour. If a reading builds on earlier work or is too long for an hour, the organisers say which parts to read beforehand.

Do I need a technical background?

No. Some readings are technical papers and others are blog posts or reports. The form asks how comfortable you are with technical papers so we can pitch the readings well.

Two weekly reading groups, one form for both.

One form, two reading groups

Questions you might have.

Questions about the Reading Groups?