1
Random “Einstein quote”: “If you can’t explain something well, you don’t really understand it.” Modus tollens to modus ponens: “If you really understand something, you can explain it well.” Had we ever asked Einstein how to create a great course, this could have been his sentiment.
But we need not go beyond the comfort of our tribe to find this meme. Our very own Peter Singer expressed it just as well when he mentioned his suspicion that “whatever cannot be said clearly is probably not being thought clearly either.
If we want high-quality explanations, we need high-quality understanding. Not just quickly scrolling through a few papers the night before a meeting, but perusing them and summarizing them with attentive review.
2
Buddhism distinguishes between the paths of Mahayana (= great vehicle) and Hinayana (= small vehicle). The former differs from the latter in that a practitioner vows to first save all other sentient beings before they attain enlightenment themselves. A westerner might have some objections here relating to efficiency, but it seems that this great vehicle is actually more successful. One sets out to save everyone and accidentally saves themselves. “Whoops. Eh, saved one.”
Altruism is weirdly motivating. There are many people in the LessWrong tribe that are very ready to help out, very ready to jump into AI safety to become researchers. Everyone wants to be the hero, right? “I might actually save the world!”
“But God forbid if I fail. I’m probably not good enough to actually become a researcher, so why even try?”
3
So let’s study AI safety together, for the sake of everyone. Let’s literally save all sentient beings, and maybe accidentally “save” ourselves in the process. Let’s do it really well. Peruse the papers, summarize them, read each other’s summaries, review, and discuss.
Veerle is gracefully taking on the responsibility of running the study group. Let me explain the (provisional) setup in detail here:
  • Before the meeting, all resources (= papers, talks, LW posts, etc) are collected.
  • At the start of the meeting, all resources are divided among the attendants, and each attendant is asked to summarize their resource. We continue doing this for some length of time (I’m thinking one or two hours). If you finish your summary, you may summarize another resource. Multiple people may summarize the same resource. If a participant doesn’t have the prerequisite knowledge for something, they may switch to a different resource, or they may spend the time to study the prerequisite instead.
  • Break. Stretch, walk, meditate, whatever.
  • During the second phase of the meeting, we read each other’s summaries. As many as we can given some time constraint. This sounds like people may miss out on some facets of the subject, but realise that there is often a lot of overlap between resources. With some luck, everyone will have a full understanding of the subject at the end of this phase.
  • Break.
  • Then we create mind maps. Attendants form pairs, and compare and possibly adjust mind maps.
  • After the meeting, one volunteer takes the set of mind maps and uses them as an input to deciding the final structure of the course.
This meeting will be mainly online, but there will also be a physical meeting space in Utrecht.
A completely different setup geared towards generating assignments is in the works too.
4
This meeting will tie into the loop that will generate the course in its totality:
  • Choose subject. Generate a list of resources.
  • Have this “mind map meeting”.
  • Use mind maps to finalize the structure of the lecture meeting.
  • Have lecture meeting.
  • Use lecture videos as an input to the assignment meeting, create assignments.
  • Review, repeat.
I’m looking forward to it!

Leave a Reply

Your email address will not be published. Required fields are marked *