Netflix Never Used Its $1M Algorithm Due to Engineering Costs | Hacker News Reader

Netflix Never Used Its $1M Algorithm Due to Engineering Costs

saintlad.com

53 pointsjakekovoor29 comments

Also not mentioned in this article: as netflix moved to being primarily a streaming business, the catalog decreased significantly. That means the algorithms to find desirable movies in the long tail of the catalog aren't so useful, as there isn't such a long tail of the streaming catalog.

This is because of copyright law: if you buy a physical disk, it is legal to rent that disk out. It was legally feasible for netflix to have an absolutely huge catalog of physical disks.

With streaming, there is no corresponding legal theory. The only way to legally stream content is to negotiate a license with the copyright holder. You can't just buy a disk. There is no mandatory licensing. And many of the studios want to compete with netflix, not license to them. So the catalog is much smaller.

Further, they've got more reasons and motivation now to manipulate you into watching some shows instead of others. Most obviously first-party vs. licensed content. Selling placement in search/browse/recommended feeds or using it as free house advertising. And burying certain content in search/browse interfaces could be useful in certain contract negotiations.

So there's more than one reason recommendations are no longer just about what you might like best.

I've noticed this as well. It actually makes me somewhat angry.

IMO Netflix breached my trust when their recommendation engine stopped trying to accurately predict what I'd like, and instead also considered their agenda.

I really miss the Netflix of about 7-8 years ago, when the company was cool and honest, and we could show our queues to our friends.

I really miss the original Netflix too.

If someone else were to start a DVD-by-mail business like the original Netflix, I'd sign up.

> If someone else were to start a DVD-by-mail business like the original Netflix, I'd sign up.

Netflix still has a DVD-by-mail service: https://dvd.netflix.com/

But not like the original Netflix. Now it's just 2 out at a time (including in the mail).

Actually, they have a variety of tiers. Basically more $/month = more concurrently outstanding DVD's.

Where are the tiers with more than 2?

"Sign Up" only shows three tiers, with a max of 2 discs at a time.

When I go to my account settings, and click on "Change Plan", I have three options for "unlimited" plans: 1, 2, or 3 discs out at a time.

IIRC, there used to be up to 5 discs out at a time available, so I was surprised to see only 3 such plans now.

Perhaps new members have fewer choices? I'm not sure.

I've given a lot of consideration to doing just this. But the logistics of such an enterprise seems pretty daunting without a lot of investment up front. I suppose I could start small with a slower service (mail and processing times) and build up. Netflix does still have their DVD business- https://www.dvd.netflix.com/ (looks like they finally spun it off into a separate company, this time with little fanfare- and not the stupid name.)

Another option I have thought about is combining the old Netlfix model with the Redbox model- have kiosks all over the place, let customers pay for a subscription, then they could select what they want to watch and have it sent to a kiosk near them... I don't know why Redbox hasn't offered an unlimited subscription service.

Really? Cor, you couldn't pay me to use a DVD in 2016. The un-skippable FBI warnings, the ads, the ads, the ads, the awful, confusing menus, the load times. Netflix may not be perfect, and may be getting worse, but at least it's not a fucking DVD.

If you use VLC, you can skip a lot of the "unskippable" garbage.

VLC certainly helps with that. It does have one downside relative to streaming: VLC can't seem to remember where I stopped watching a DVD.

Every time I bring my Linux box back from sleeping, the DVD has to get remounted (or something like that), and VLC has amnesia about where I had stopped it.

So I end up having to write down where I stopped, and re-navigate the just-put-a-DVD-into-the-drive menus, every time I come back to the movie.

I would have expected this to be a solved problem, but perhaps my Google Fu is just weak.

So go to your grocery store. Milk is in the corner so you have to all past the candy and chips to get to it. Did they breach your trust? Customers want milk - why isn't it in the first aisle?

Retail stores have done this for years. Why wouldn't an online store do the same thing?

That's a highly debatable factoid. Milk is in the back of the store because that's where deliveries come in, so it's easier to ensure it stays refrigerated at all times that way.

http://www.npr.org/sections/money/2014/07/23/334076398/episo...

It's slightly different with Netflix and its recommendation engine.

To extend your analogy the grocery store would need to be telling you "Hey shopper, based on your personalized history of buying milk we think you'd like to buy cereal." But instead of giving you all of the cereal in the same aisle the store brand cereal is up front and costs the same as the brand names, and all the cereal the store doesn't get a cut on is in the farthest corner.

Retail store methods, utilizing tricks and subtle manipulations, may be legal but it doesn't make them unambiguously ethical and above criticism. They are permitted to function that way for practical reasons of interest to all of us, but it's not an argument against it being a breach of trust.

(And it would likely be a net economic benefit to society if people bought less junk food, and could get in and out of a grocery store quicker.)

Actually, our milk is delivered weekly, in glass bottles, by the dairy truck. It's pretty epic.

And they used to have a good API...

> And many of the studios want to compete with netflix, not license to them.

And even the ones that don't want to compete with Netflix field offers from Netflix's competitors (and, naturally, Netflix itself) for exclusive streaming license for top tier content.

I'd argue this challenge was one of Netflix's better investments.

Sure - most of the winning algos never actually got used - but here we are, still talking about this almost 10 years after it's inception.

I've been mocking this contest for years, since Netflix apps have horrendous discoverability and lack basic sort/filter features. I often wondered if Netflix deliberately omits features to make their catalogue seem bigger and better.

Spoiler: yes.

The free PR and the recruitment value almost entirely offsets the prize cost.

Maybe I'm not understanding the premise of this article. I always thought a bit over $1m seemed cheap to get access to self selecting, really bright AI people to hammer on an issue for a couple of years. Easy to confirm the results. Lots of publicity. Sets a bound for production to compare against, and stop worrying about it.

The only "algo" I am expecting on the homepage can be condensed in "SELECT * FROM movie ORDER BY added_date DESC" and a "feature" to "Hide forever this item". Of course, I would unsubscribe after seing the staticity of the results. ;)

Did the check cash or not? They didn't mention that part...

The cheque is usually a novelty item. They usually do large transfers digitally or via wire, probably divided to each of the winners. It would have been bigger news if they didn't.

Reply on news.ycombinator.com