Most of your group’s impact will come from the people who end up most highly engaged and motivated to work in AI safety. Therefore, building a thriving AI safety group can be seen as the problem of expanding the core group.

<aside>

One group, Boston University, did this particularly well. Read their memo here: The Unreasonable Effectiveness of a Strong Core Team

</aside>

One of the primary goals of your club should be helping the core group become kick-ass AI safety people through:

• Upskilling
• Networking
• Increasing AI Safety context
• Doing research
• Building key soft skills
• Etc.

The people who are attracted to your group will be roughly as good as your average group member. If you have a core of kick-ass people, others will notice this and want to emulate you and join your club. This also means that once a core group’s composition has been defined, it’s much harder to change it in the future (see our article on Founder Effects).

In order to attract good people, your people need to have deep knowledge of AI safety.* *People want to talk to people who are roughly “on their level”. By leveling up, you will attract higher-level people. Not knowing AI safety sufficiently well will lead to failure in the most high-stakes scenarios, such as grad students and professors. It’s also unlikely your group will be super impactful if it’s not attracting the top 1-10% of students at your university.

Knowing a lot about AI Safety is not very hard. Taking something like ARENA (and a bit more ML) and reading and discussing ~100 AIS papers will make you much more knowledgeable in AI safety than most of the rest of the world, and people will notice. You can do this with friends, turning a possibly boring one-person activity into a group activity, which further strengthens your core group.