zlacker

[return to "Had a call with Reddit to discuss pricing"]
1. ccooff+g8[view] [source] 2023-05-31 18:08:28
>>robbie+(OP)
> 50 million requests costs $12,000

I've never worked on a web platform like Reddit, nor with any per-request priced APIs. Reddit's charge of $0.00024 per request still looks like it is _significantly_ above what their own costs are.

Wasn't Reddit's pay-for-API-access announcement originally phrased as a desire to claw back some of the value that LLMs have found in Reddit data? I don't understand how per-request API pricing actually accomplishes that. (I was vaguely anticipating Reddit's API pricing to have some sort of expensive "firehose" endpoint for OpenAI/Google/Meta/etc to pull from.)

It looks like they're instead going to squeeze out all third-party apps instead. I don't think this bodes well for Reddit's future.

◧◩
2. kenhwa+je[view] [source] 2023-05-31 18:30:25
>>ccooff+g8
Their pricing likely includes the cost of potential ad revenue that an API call is displacing. There's no easy way to integrate ads into an API, so they just offload the advertising/monetization problem to whoever buys the content.
◧◩◪
3. fastes+Kf1[view] [source] 2023-05-31 23:33:17
>>kenhwa+je
Couldn't they just return ads from the API? Any apps that don't show them would have their API access revoked.
◧◩◪◨
4. danpal+ZW1[view] [source] 2023-06-01 07:58:55
>>fastes+Kf1
The tricky bit with ads is that someone is paying for a user to see it, so if the user never scrolls to it, or perhaps scrolls past too fast, you probably need to know so you don't bill for it. If the user taps on it you definitely need to know that, regardless of whether the ad link is working, but you can't redirect and slow down the ad link.

You need highly accurate data about user behaviour around the ads, and highly optimised display and linking.

I've built a similar system, albeit not for ads, for recommendations, based on viewing items in a feed. With a simple model, across web and app, all developed in-house, it was still hard and required a lot of care to get good ML signals and signals that people really understood.

Doing this across third party clients may be prohibitively difficult. I'd like to see attempts, but so far I've not seen any.

◧◩◪◨⬒
5. steven+UE5[view] [source] 2023-06-02 08:58:36
>>danpal+ZW1
Doesn't this problem go away if the site uses a CPC model?
[go to top]