Hey all. The lecture meetup last week was a success, and I’d like to thank Johannes again for coming all the way from Berlin to expound the way of superrationality. Should there be some entity that wishes to maximise knowledge in the multiverse, I’m sure they’ll be very pleased In other news, MIRI emailed us an unpublished paper on corrigibility. I’m glad we can include it in our next meeting! …which will be in September. August is a bit of a slow month for RAISE, as people are on holidays and such. There will be no meeting. It’s the quiet before the storm; we will start having a meeting every 2 weeks after the holidays. But there is still work to be done. Dony is compiling the two streams of video we made last week, and someone needs to (decide how to) write assignments. I’m trying to fully delegate the latter — to test the waters of delegation, if you will. Let’s see if someone (not you Remmelt!) feels compelled to pick it up. If you do, please start with sharing your ideas at #assignments on the Slack. In the meantime, let me have a word on the setup of the lectures, and why it deserves a word. The setup is important; the setup is hard When it comes to maximising learning, there are a lot of subtleties. Humans are good at communicating informally, but it’s not trivial to formalise a method of communication that engages our intuitions. We’re dealing with the intricacies of system 1 here. Consider these failure modes:
- The instructor has slides that are projected on the wall, but they look at their own screen, so they don’t really establish reference for the students.
- The instructor presents a nice proof of a theorem with the formulas already pre-written on the screen, but they usually engage with the proof through writing it out. They’ve never seen the proof in this context, so they fail to really activate their subagent that clearly understands the proof. The explanation becomes confused.
- The instructor stands, signaling authority, making the students and the instructor a little bit uncomfortable, which diminishes learning.
- The instructor sits among the students and writes out a proof on his tablet, of which the screen is projected onto the wall. They share a gaze with their students but don’t establish eye contact, which is a tool for soliciting attention.
So, to start with, we want sitting, writing, a shared gaze and eye contact.
There is one important incompatibility here. You can’t have a shared gaze or eye contact if one sits down writing out a proof on a tablet sitting under their nose. More generally, I would relate this to the divide between extrovert and introvert mode of the lecturer. Both are mental tools that have their benefits and drawbacks. Extrovert mode is more engaging for the students. Introvert mode interfaces better with complex material.
The solution that we have in mind is to split up the setup in two forms corresponding to these two modes: the “broad” and “deep” setup. “broad” setup The lecturer stands up next to a screen showing digital content (like slides). We have a camera pointed at the lecturer, which also has the receiver of a microphone attached to it. The microphone itself is worn by the lecturer. We also keep a screencapture of the slides, and we merge these two streams of video as we later deem appropriate.
This setup is most suitable for more “philosophical” topics, or things about macrostrategy, and it is less suitable for maths.
“deep” setup Example: https://www.youtube.com/watch?v=ihN… The lecturer sits down with a tablet or laptop and demonstrates content while talking about what they are doing. There is a camera pointed at them, but we only use that stream for sound. There is also a screencap of the tablet. The video of the screencap is later combined with the audio of the camera to produce the end result. The tablet is also connected to a bigger screen, but that’s only to make things convenient for the people physically attending the meeting. This setup is most suitable for maths. Bonus: this setup could be implemented by people who aren’t based in the Netherlands. Pretty sure this will work! With a bit of luck, we will have our first unit perfected by the end of October.
Thanks for being involved everyone! Cheers, Toon