Executive summary: A reflective, practice-heavy retrospective from the founders of UBC AI Safety reports strong early demand (50+ participants, 12 leaders, active Slack) but modest career conversion, and offers concrete, cautiously framed advice on program design, outreach, operations, and successor planning for university AI safety groups.
Key points:
Interest vs. action gap: Despite large sign-ups and steady attendance, pre/post surveys and facilitator impressions suggest only small shifts toward AI safety careers; authors flag attainability, competitiveness, weak local role models, limited faculty support, and real-life constraints as likely causes.
Program design lessons: Clarify your group’s goals first (exposure vs. outputs vs. career pathways), then select formats accordingly; their mix—technical intro (using AI Safety Atlas), policy reading group, ARENA-based upskilling, and a discussion group—kept engagement high but produced few tangible projects without mandatory, scoped sprints.
Improving pipelines: Make competitive skills explicit early, add a concrete deliverable to each program (replications, blog posts), offer pre-scoped project ideas with clear success criteria, and build community ties via 1:1s, dinners, mentorship, and visible success stories.
Operations and governance: Decide which AI risks you focus on (existential vs. near-term) to avoid cultural drift; minimize meetings; plan conservatively (planning fallacy); secure venues early (becoming an official club can help); start succession planning well before year-end with role docs and a formal handoff.
Outreach and finance: Department newsletters outperformed other channels; Instagram and fairs were moderately useful; free, simple dinners aided retention; budget per program, set clear reimbursement rules, and expect low admin burden (in their Canadian context).
Organizer career growth: Use organizing to learn (policy reading aligned with interests), host co-working for applications, leverage the “group lead” credential for networking, and consider opportunity costs—organizing may not be optimal for experienced researchers but is valuable for newcomers building trajectory and community.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.
Executive summary: A reflective, practice-heavy retrospective from the founders of UBC AI Safety reports strong early demand (50+ participants, 12 leaders, active Slack) but modest career conversion, and offers concrete, cautiously framed advice on program design, outreach, operations, and successor planning for university AI safety groups.
Key points:
Interest vs. action gap: Despite large sign-ups and steady attendance, pre/post surveys and facilitator impressions suggest only small shifts toward AI safety careers; authors flag attainability, competitiveness, weak local role models, limited faculty support, and real-life constraints as likely causes.
Program design lessons: Clarify your group’s goals first (exposure vs. outputs vs. career pathways), then select formats accordingly; their mix—technical intro (using AI Safety Atlas), policy reading group, ARENA-based upskilling, and a discussion group—kept engagement high but produced few tangible projects without mandatory, scoped sprints.
Improving pipelines: Make competitive skills explicit early, add a concrete deliverable to each program (replications, blog posts), offer pre-scoped project ideas with clear success criteria, and build community ties via 1:1s, dinners, mentorship, and visible success stories.
Operations and governance: Decide which AI risks you focus on (existential vs. near-term) to avoid cultural drift; minimize meetings; plan conservatively (planning fallacy); secure venues early (becoming an official club can help); start succession planning well before year-end with role docs and a formal handoff.
Outreach and finance: Department newsletters outperformed other channels; Instagram and fairs were moderately useful; free, simple dinners aided retention; budget per program, set clear reimbursement rules, and expect low admin burden (in their Canadian context).
Organizer career growth: Use organizing to learn (policy reading aligned with interests), host co-working for applications, leverage the “group lead” credential for networking, and consider opportunity costs—organizing may not be optimal for experienced researchers but is valuable for newcomers building trajectory and community.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.