Y Combinator Challenge #13 - Online Learning | Hacker News Reader

Y Combinator Challenge #13 - Online Learning

astartupaday.wordpress.com

9 pointscrocus19 comments

I'll add something here, since I'm writing an application focused on language learning: Pick an area to focus on, and provide lots of automation for the user.

Even though my application is very beta, it automates a lot of the drudgery in building studylists, which differentiates my software from all the other spaced repetition systems out there. This has gotten a pretty positive response so far, and has the added advantage of being very difficult to duplicate, giving you a big competitive edge.

Not sure how this could be applied to subjects other than language, and doing it for language is hard enough, but that's what makes this an interesting problem. :)

Also, I've dug around quite a bit, and talked to a few other researchers, and have yet to find any proper studies on the effectiveness of SuperMemo algorithm itself. There are lots of testimonials, but nothing involving a control group, and so until I see some hard data, I'm going to remain a little skeptical.

If anybody knows of such a study (with associated data), please let me know.

Of course, since I'm building a learning system, this puts me in a good position to undertake such an experiment. grin Which I plan to do next semester, using a few hundred students' worth of Japanese classes.

Upvote for using hundreds of student hours. Free labor is good for startups!

I also resent the fact that your time is viewed as worthless if you don't have a college degree.

It's not even about free labor. The students in the classes will get free access to my learning system, which has all their class materials built-in, and so no matter what the result of the experiment, the people who use the system will at least reap the same level of benefit that they would with traditional flashcards, and with far less work.

So, I'm doing something to help them, and in return, getting some data to improve the system. Truly the epitome of a win-win scenario.

And, actually, I do have a college degree. Two of them, to be precise, and I rather wish they had been printed on toilet paper, so they'd be worth something.

I've met many college-educated idiots, and many self-educated geniuses. I've also met a similar number of fantastically intelligent college graduates, and self-educated morons. The true test of intelligence, in my experience, is action. Those with intelligence act, create, build, and contribute to making humanity better, whether they want to or not.

So, you've been upvoted for being agreeable. :)

I'm definitely looking forward to your launch. Sounds great!

Wozniak is a published researcher and Supermemo is basically the fruit of his PhD thesis.

I'm aware that Wozniak is a published researcher, as I've read a good chunk of his work; what he's done on memory plasticity is very solid, and provides a really good argument for spaced repetition.

But nowhere in the doctoral thesis is there a link to any sort of scientific study on the effectiveness of the various iterations of the SuperMemo algorithm. If I'm wrong, please tell me where I can find the information I'm looking for, because it would make my life much easier.

Jesus, Mohammed, Yaveh, and Gordon Ramsay could all come down from on high and hand me a learning method that they all swear by, but until I saw some solid data proving its effectiveness, I would remain skeptical.

If you are looking for the most efficient algorithm, then I can say with good confidence that nobody has it figured out yet. The biggest problem with SM is that it is designed for declarative knowledge. I consider it revolutionary in this area, but nevertheless, focusing entirely on declarative knowledge isn't the best strategy either.

I'm not so sure about plasticity -- what kind of plasticity is this? How would you test it? Synaptogenesis is out of the question (because this is not a cellular-scale operation). Standardized tests? That itself is full of confounds.

I have come across some studies regarding memory decay. Perhaps some algorithm is involved. If you dig around I'm sure you can find something, but conclusive data about the effectiveness of various aglos... you'd think that if it exists, the whole world would know by now.

Having failed a couple papers in educational theory classes for proposing vaguely similar ideas, I'll give you the standard criticism: How does this maximize the child's intrinsic motivation to learn? Which isn't the same question as 'will this be fun,' which you make the case for. But rather, will this make kids the kind of people who want to check out non-fiction books from the library on their own ten years from now.

This standard criticism makes a very strong case, doesn't it?

I was recently talking to someone experienced with making educational software. He said that Indiana Jones is full of bad archaeology, but causes surges in archaeology major enrollment numbers. Inspiration probably precedes motivation. Before inspiration: understanding, control, and application?

In this sense an educational application with a competitive/game component is not enough to build inspiration that lasts beyond the duration of the game, which would have very narrowly defined goals.

>But rather, will this make kids the kind of people who want to check out non-fiction books from the library on their own ten years from now.

That's a really tall order.

While we don't know much about building intrinsic motivation, we do know a lot about the factors that undermine intrinsic motivation. So in practice if you can avoid all these, and you make the process both enjoyable and useful, then at least you're on the right path.

Were your papers failed because they had radical ideas, or debunked ideas (that said, can any educational theory ever be thoroughly debunked, all hard facts accounted for)?

Well, I turned one of them into this K5 article so you can judge for yourself:

http://www.kuro5hin.org/story/2004/12/8/2135/17844

Actually, that one I wrote in October 2004, and one of the biggest problems with it was that the professor had never heard of Wikipedia and didn't think that the site could ever possibly succeed. (The first drafts were more Wikipedia-centric, but a lot of that got cut out of the final version.)

Great article, thanks for providing a point of view from someone who's not talking out of their a$$ (aka me). The wiki idea is very, very interesting - it's got that same "I'm not doing this for money but to help the world be a better place" vibe that (imho) is what makes Wikipedia successful. There are valid criticisms of this approach in the comments, namely that it would require a very strict quality control process before educators could trust it. Otherwise, millions of children would grow up thinking that Stephen Colbert really did single-handedly triple the elephant population.

I'm not sure I would still agree with everything I wrote though, so take it with a grain of salt. (Although I haven't actually reread it recently.)

Given its brevity (in terms of length and citations), I suppose this writing is making a personal case for PI, from a mostly theoretical standpoint?

The mention of Skinner threw me off in the beginning, but my feeling is the PI you mention isn't a behaviorist one, but a metaphorical one? If so, I absolutely agree with this idea, and think we have enough technology to make it a reality.

Yeah there is a lot of stuff in there I wouldn't say today. That's why I linked to it, as an example of my failing to understand intrinsic motivation.

nitpick: spacing algorithm does not "dramatically increase memory." Your memory is not so much increased. Your retention rate is what increases.

Another nitpick: 10 minutes a session is too short for anything of any intensity. For single sessions, 40 minutes would be better. But again there isn't a hard rule. Oh well, ideas surely are cheap.

Thanks for the correction, I've updated the post accordingly.

The idea behind the 10 minute sessions was to address issues with short attention spans. Maybe that could be another dynamic variable - the lessons change in duration once they reach a point of diminishing returns.

Reply on news.ycombinator.com