3 ms·
Some may remember that Reddit used to have an item recommender a long time ago, back in its first year or so. It was a Bayesian classifier that, since it neede
by zach 15y ago
Some may remember that Reddit used to have an item recommender a long time ago, back in its first year or so. It was a Bayesian classifier that, since it needed a bunch of input, only worked for the most hardcore members — who had already seen almost all of the recommendations!
This was originally the "hard problem" at the center of Reddit.
Let me explain what I mean by that. There used to be a quaint notion that to be a respectable tech startup, you had to have a "hard problem" (technologically speaking) at your core, which you had an innovative "secret sauce" solution for, preferably one you were patenting. After all, if not, then someone can just copy you and squash you like a bug, right?
Since then, YC's insistent focus on making something people want, Eric Ries' lean startup gospel and many entrepreneurs' own experiences have thankfully gone a long way to convince people (most importantly SV investors) that focusing on a "hard problem" is not only unnecessary, but may end up being a fatal distraction.
This is a pretty good example of how the "hard problem" can turn out to be completely irrelevant. Once it was clear that the recommendation engine wasn't a growth vector, the Reddit team seemed to drop it out of sheer pragmatism. They just needed to keep the site running.
I can't recall many who cared or even noticed that the "recommended" tab was gone. But from that point on, Reddit was more free to become not just a quirky "personalized news" startup, but what it has aspired to since: the front page of the internet. And only now, just now, do a good chunk of the millions of users think a recommender might be nice.
It's the startup version of "you aren't gonna need it" — if it doesn't drive growth, push it aside.
- citricsquid 15y agoI think you went too far. You're correct that reddit didn't/doesn't need any form of recommendation engine for individual items -- that's what voting + subreddits are -- it most definitely did (and still does) need something for recommending subreddits. This is a problem they should have been addressing from day 1, it isn't make or break but it takes the site way beyond its current usefulness to hardcore users.
- defrost 15y agoIt was something like three years before reddit had subreddits. Hardly a day 1 problem.
- DanielRibeiro 15y agoEarlier today, on Clojure West, the founder of GetPrismatic[1] made a very interesting presentation. He is trying to solve this problem. Real time machine learning is a hard problem. Hopefully it will be something people want, because I want it as well. But if it is not, I'd rather he pivoted into something where he can succeed than taking his startup to the ground because of me. A startup that fails helps nobody. [1] http://getprismatic.com/ http://getprismatic.com/
- achompas 15y agoIs this presentation online somewhere? Sounds v. interesting, especially if it's Bradford Cross.
- DanielRibeiro 15y agoNot yet: https://github.com/strangeloop/clojurewest2012-slides https://github.com/strangeloop/clojurewest2012-slides InfoQ taped it, but they may not release it for up to 6 months. And yes, it was Bradford Cross.
- philwelch 15y agoThank god tech startups no longer have to actually develop technology. I suspect the "hard problem" idea might not actually be that antiquated. Most of the newer batch of startups are either not profitable at all or not profitable enough to justify VC investment. There's almost no technology risk, but the tradeoff in market risk is such that even if you succeed, you're not as profitable. This is the trap Reddit is in.
- rayhano 15y agoJust on a slight tangent - have we gone too far and forgotten to solve hard problems; leaving them to the likes of Apple, Samsung and sequestered Universities?