Awesome idea. I signed up for both roles (Though I can't teach anything til August). I predict the most in demand subject will be machine learning. After that I would guess stats/probability. I wonder how easy it will be to coordinate ~5 learners to agree to one "class". I would guess this project will wind up having 2-3 people in some "classes", assuming the tutors agree. But perhaps I am wrong. Its hard to get people to agree to both a subject and a difficulty level.
Agreed about the difficulty in class coordination. Multiple time zones and overlapping schedules could make consistent classes difficult.
Note: Everything I write below about pedagogy is anecdotal, and I don't have much experience outside of being a student.
This is a really cool idea! I think it's a good way of bringing people in the community closer together!
That being said, I'm worried the classes might not turn out to be very good because good teaching seems generally pretty hard. (Which isn't to say that I don't think LWers can't rise to the challenge, just that this means my priors on "Video tutoring will turn out well" are low.) (Or perhaps I just didn't have many good teachers).
I think personalization could be a very important niche role we could exploit here to make things better. So rather than the Teacher spending time going over the subject, they could chat individually with Learners to assess where they are at, suggest materials, and then follow up after each iteration. (Think something like the flipped classroom model where we're counting on people to be self-motivated learners, and these video sessions would be more for supplementing self-studying.)
Everything I write below about pedagogy is anecdotal, and I don't have much experience outside of being a student.
As a former teacher I complete agree with your approach. One critical component of teaching is determining your students' next inferential step. Which is easier in a "one to one" setting, and almost impossible in a "one to many" setting.
Various online lessons are an exercise in frustration because always someone complains they are too easy (that they cover only the "obvious" and "well-known" facts), and someone else complains they are too difficult (that they use terms they didn't explain, and go through complex ideas too fast).
Traditional school solves this by making everyone progress along the same path, and the same speed. I guess we all know how frustrating this is for smart people who have a potential to do much faster. On the other hand, LW audience will consist of smart people, but there will be wild differences in background knowledge.
So the approach "this is the outline, read these things in this order, and discuss with me when you complete something or you get stuck" is a good solution that allows you to provide material for many people at the same time, and yet address everyone individually.
Some online courses have separate Q&A threads for each lesson, where students can talk to each other, but only about the specific lesson.
I suspect this will end up being something more akin to self-study groups that produce teaching material as a direct result of learning the material themselves. For example, writing up an explanation of how to do a particular book example. This doubles as an assessment of people's skills since other people that know the topic really well can build on those explanations or correct mistakes.
With a series of such explanations, anyone else trying to go through the material will have a clearer pathway for the level of understanding of a given sub-topic they need to develop to progress: the exercises and readings needed to be able to understand something, or do a particular difficulty of project.
This is indeed something that could happen, and I agree it'd be valuable.
Although, while I think that writeups / explanation can be valuable for both the people writing it / the people reading it, the thing I was actually trying to point to was the sort of benefit you might get from a tutor: that is, if someone knows where you're struggling, they can suggest things that are more tailored to help you succeed vs a typical online curriculum.
(And that this sort of personalized "adaptive curriculum" could be a unique benefit to come out of this activity.)
Video chat probably isn't good enough by itself for many topics. For programming, screen-sharing software would be helpful. For mathematics, some kind of online whiteboard would help. Is there anything else we need? Do any of you know of good resources? Free options that don't require registration are preferable.
Good points, I agree. Screen sharing is possible via Hangouts, Skype and talky.io. I'm not sure what the best online whiteboard software is, but screen sharing + using some sort of notepad type thing should work.
Screen sharing is one-way though. If you both need to draw on the same space, it would be pretty awkward. I've heard of Twiddla, Deekit, and GroupBoard, but haven't used them.
That https://talky.io site looks pretty useful. I've used a similar one called https://appear.in which also uses WebRTC. I don't know if one is better. [Edit: looking this over, talky seems better for our purposes.]
I'm surprised that there aren't any active YouTube channels with LessWrong-esque content, or at least none that I am aware of.
I just started a Facebook group to coordinate effective altruist youtubers. I'd definitely say rationality also falls under the umbrella. PM me and I can add you. :)
I propose that Learners individually reach out to Teachers, and set up meetings.
Consider that it may make sense for you to act as a Teacher, even if you don't have a super strong grasp of the topic.
Why that way instead of the reverse? True, the learners probably have the greater motivation. But, the learners have a better idea of what they want to learn than the teachers have of what they can teach, especially if we're accepting teachers without a super strong grasp of their topics. Thus, I think it would make more sense for the learners to post in detail what they want, and the teachers to look over all of that and make the offer on whatever topics they can help with, even if only a little.
We could certainly do both, but then I worry that each will hope the other initiates. The cure for this is if one individual plays matchmaker to get things started. Due to the bystander effect, I'll name adamzerner as the obvious choice for the role, but you could delegate then abdicate if someone else is willing.
I pinned http://lesswrong.com/r/discussion/lw/p69/idea_for_lesswrong_video_tutoring/ to #productivity on the LessWrongers Slack group.
A chat room could also work better than individual emails. But everybody has to be on the same channel and check it regularly. I don't even have an invite yet (I just asked Elo for one). Is everyone else on?
Maybe relevant (somewhat personal) - I'm in the market for math tutoring. Context: I'm not really a beginner, in fact an ML practitioner with mostly software background. I learned enough math to have useful intuitions but really want to level up in understanding/solidifying probability/linear algebra/optimization and maybe some MIRI-relevant math (though less practically relevant). Ideally I'd want to work with someone who can probe my current knowledge and help me lay out a plan and fill gaps, mostly independently (assign homework) and work through sticky problems over video.
Willing to pay fairly for the service depending on quality, potentially for a long time. PM if capable+interested.
I think this is a great idea, likely to have positive value for participants. So going Hamming questions on this, I think two things are important.
Failure seems like the default outcome. How do we avoid that? Have there been other similar LessWrong projects like this that worked or didn't? Maybe we can learn from them.
Group projects can work without financial incentives. Most contributors to wikis and open-source software, and web forums like this one, aren't paid for that.
Assume we've made it work well, hypothetically. How did we do it?
It reminds me a lot of the "mastermind group" thing, where we had weekly hangouts to talk about our goals etc. The America/Europe group eventually petered out (see here for retrospective by regex), the Eurasia/Australia group appears to be ongoing albeit with only two (?) participants.
There have also been online reading groups for the sequences, iirc. I don't know how those went though.
forums, wikis, open source software
I see a few relevant differences:
We might be able to apply these "differences" to our attempt. A lot of the value we're talking about here is just some basic direction to get started and help when you get stuck. That's a pretty "small barrier to entry", and then "small incremental improvements".
Could we dedicate a Slack channel to video tutoring? My experience with small IRC groups is that there is a small number of experts who check in frequently, or at least daily. Then the beginners will occasionally pop in and ask questions. If they're patient enough to stay on, an expert usually answers within the day, and often it starts a real-time chat when the expert mentions the beginner's handle. We could use the Slack channel to ask questions to get started or when we get stuck. If an appropriate teacher is on, then they can start a video chat/screen share on another site. There would be no obligation for a certain time limit.
I think this is a great idea, mostly as a motivator for people to come with classes.
For example, I have been thinking of doing a workshop on mental math or OKCupid for a while, but the logistics of doing it IRL for more than 2-3 people seem overwhelming. But this gave me the idea that I can plan to do it online, which actually has a lot of benefits, and I can advertise and see who signs up on LW.
For people interested in learning Russian, I can't "teach" teach it, but I can help you practice it.
A question for people asking for machine learning tutors: have you tried just reading through OpenAI blog posts and running the code examples they embed or link? Or going through the TensorFlow tutorials?
By using the word "just", it gives me the impression that you think it's easy to not get lost. In my experience with other fields, it is easy to get lost, and I would assume that the same is true with machine learning.
I meant it as "This seems like a clear starting point." You're correct that I think it's easy to not get lost with those two starting points.
I'm my experience with other fields, it's easy to get frustrated and give up. Getting lost is quite a bit more rare. You'll have to click through a hundred dense links to understand your first paper in machine learning, as with any other field. If you can trudge through that, you'll be fine. If you can't, you'll at least know what to ask.
Also, are you not curious about how much initiative people have regarding the topics they want to learn?
Update 7/9/17: I propose that Learners individually reach out to Teachers, and set up meetings. It seems like the most practical way of getting started, but I am not sure and am definitely open to other ideas. Other notes:
Update 6/28/17: With 14 people currently interested, it does seem that there's enough to get started. However, I'd like to give it a bit more time and see how much overall interest we get.
Idea: we coordinate to teach each other things via video chat.
So, this seems like something that would be mutually beneficial. To get started, we'd need:
Personally, I'm not sure how much I can offer as far as doing the teaching. I worked as a web developer for 1.5 years and have been teaching myself computer science. I could be helpful to those unfamiliar with those fields, but probably not too much help for those already in the field and looking to grow. But I'm interested in learning about lots of things!
Perhaps a good place to start would be to record in some spreadsheet, a) people who want to teach, b) what topics, and c) who is interested in being a Learner. Getting more specific about who wants to learn what may be overkill, as we all seem to have roughly similar interests. Or maybe it isn't.
If you're interested in being a Learner or a Teacher, please add yourself to this spreadsheet.