First-year PhD in CS at Purdue; co-founder of Effective Altruism Purdue and AI Safety Purdue; aspiring AI alignment researcher
Nathan Reed
Karma: 88
Some ideas I had:
Advice I found useful for my PhD decision
Transformers Struggle to Learn Search and my model of LLM capabilities
Why EA and AI Safety groups should recommend watching Pantheon
A meta-analysis of university group organizer advice posts
Lessons Learned from Running EA Purdue
Thanks! We took an approach of making the fellowship application form as minimal as we could, so most of our applicants were able to put fairly high-quality responses to the remaining questions. We also have experienced a somewhat higher attrition rate for the actual fellowship (4->2 last semester, 50->15 this semester), which we suspect is due to the shorter forms, but we haven’t seen any other new problems.
I’m the co-founder and one of the main organizers of EA Purdue. Last fall, we got four signups for our intro seminar; this fall, we got around fifty. Here’s what’s changed over the last year:
We got officially registered with our university. Last year, we were an unregistered student organization, and as a result lacked access to opportunities like the club fair and were not listed on the official Purdue extracurriculars website. After going through the registration process, we were able to take advantage of these opportunities.
We tabled at club fairs. Last year, we did not attend club fairs, since we weren’t yet eligible for them. This year, we were eligible and attended, and we added around 100 people to our mailing list and GroupMe. This is probably the most directly impactful change we made.
We had a seminar sign-up QR code at the club fairs. This item actually changed between the club fairs, since we were a bit slow to get the seminar sign-up form created. A majority of our sign-ups came from the one club fair where we had the QR code, despite the other club fair being ~10-50x larger.
We held our callout meeting earlier. Last year, I delayed the first intro talk meeting until the middle of the third week of school, long after most clubs finished their callouts. This led to around 10 people showing up, which was still more than I expected, but not as much as I had hoped. This year, we held the callout early the second week of school, and ended up getting around 30-35 attendees. We also gave those attendees time to fill out the seminar sign-up form at the callout, and this accounted for most of the rest of our sign-ups.
We brought food to the callout. People are more likely to attend meetings at universities if there is food, especially if they’re busy and can skip a long dining court line by listening to your intro talk. I highly recommend bringing food to your regular meetings too—attendance at our general meetings doubled last year after I started bringing chips.
We sent out more departmental outreach emails earlier. Again, this was delayed until the third or fourth week of school last year, so most people were no longer looking at the mailing lists. This year, we sent out departmental outreach the week before school started, and ended up in a lot of inboxes while people were still paying attention. We also increased the number of departments we reached out to, and gained more mailing list sign-ups accordingly.
We had more experience. Last year, I was a first-time organizer, and was learning as I went with the help of CEA’s University Group Accelerator Program and Organizer Support Program. (I highly recommend signing up for these if you get the chance—I would not have taken many of the my effective steps last year and this year if I didn’t have their advice!) This year, I have a better idea of what I’m doing, and can more accurately prioritize between outreach opportunities and events.
Most importantly, we had a second organizer. Last year, I was largely running the group by myself, and most tasks were blocked on my own time and energy. Having a second organizer gave each of us the chance to focus on tasks we found easier, and allowed us to get things out the door faster. Our discussions also improved our strategy from what I would’ve been able to devise alone. (Thanks, @Matthew M!)
Early group organizing can often feel demoralizing and hopeless. You may put a lot of effort into setting things up, only to get a single-digit number of applicants to your first seminar. But the work is worth it—I wouldn’t have had the success of this year if I hadn’t spent the last year running seminars and filling out registration paperwork. Good luck with your own organizing, and may you find similar success!
Nathan Reed’s Quick takes
I’m fairly convinced by the scale arguments for animal welfare, but have a slight hesitation due to worldview diversification considerations, optics, and a possible lack of room for more funding. If I had to irreversibly allocate the $100m in the next ten minutes, however, I would choose animal welfare.
Pantheon is a show that ended up on Netflix recently and covers a singularity caused by uploads. I think it gives really good intuitions on what a fast AI takeoff might look like.