The move is controversial, with many third-party apps having to shut down as a result, but the Reddit CEO has his reasons and doesn’t appear to be backing down
I think the thing that blows my mind about the pricing model is- AI using reddit for training data doesn’t need to up vote, downvote, subscribe, comment, or constantly ingest new data. They could ingest everything on Reddit once and be done, or come back every 6 to 12 months for an update. 3rd Party App developers need to download the same posts from the front page for every user, multiple times a day, constantly. They actually interact with a website. A $1 billion LLM ai might pay $1 million to download all the reddit data they need. Meanwhile Apollo might be worth less than $1 million as a business and is being hit with $20 million per year as operating expenses. The pricing model is set up in such a way that LLM’s are in a much better position to either pay a pretty insignificant fee to get all the data, or just build a scraper since they don’t need to support multiple users or website interaction. Meanwhile the price to app developers is impossibly high.
I’d go with pathological liar given how he alleged that the Apollo developer threatened Reddit - allegations which the Apollo developer proved false by releasing the audio of the phone call - and then Spez doubled down on the allegations while saying the developer shouldn’t have released the audio. Basically, “I’m going to defame you and you providing proof of me lying about you isn’t allowed.”
They are just lying. Blaming LLMs is just a convenient, topical target. If that was really a problem, why did they not react anytime in the last few years when thimgs like ChatGPT were actually gathering their initial data? It’s not like this tech popped up or of nowhere, it’s been around for awhile but just recently became a mainstream story due to the increased access.
I don’t get your argument. A service that uses the api once or twice a year, and uses much less of the offered features, and generally stresses the website much less… Is to pay more, simply because the application is worth more?
I think he’s calling out the fact that the justification for the API pricing changes was partially “we’ve got to stop AI training bots from scraping Reddit.” However, the pricing changes were actually made to hit third party developers hard and not hit the AI modelers that hard.
I think the thing that blows my mind about the pricing model is- AI using reddit for training data doesn’t need to up vote, downvote, subscribe, comment, or constantly ingest new data. They could ingest everything on Reddit once and be done, or come back every 6 to 12 months for an update. 3rd Party App developers need to download the same posts from the front page for every user, multiple times a day, constantly. They actually interact with a website. A $1 billion LLM ai might pay $1 million to download all the reddit data they need. Meanwhile Apollo might be worth less than $1 million as a business and is being hit with $20 million per year as operating expenses. The pricing model is set up in such a way that LLM’s are in a much better position to either pay a pretty insignificant fee to get all the data, or just build a scraper since they don’t need to support multiple users or website interaction. Meanwhile the price to app developers is impossibly high.
Correct. Spez is either a pathological liar, or incomprehensibly stupid. But probably both.
I’d go with pathological liar given how he alleged that the Apollo developer threatened Reddit - allegations which the Apollo developer proved false by releasing the audio of the phone call - and then Spez doubled down on the allegations while saying the developer shouldn’t have released the audio. Basically, “I’m going to defame you and you providing proof of me lying about you isn’t allowed.”
They are just lying. Blaming LLMs is just a convenient, topical target. If that was really a problem, why did they not react anytime in the last few years when thimgs like ChatGPT were actually gathering their initial data? It’s not like this tech popped up or of nowhere, it’s been around for awhile but just recently became a mainstream story due to the increased access.
These ai training data makers actually use scrapers, wich is why the argument is ridiculous anyway.
Exactly.
I don’t get your argument. A service that uses the api once or twice a year, and uses much less of the offered features, and generally stresses the website much less… Is to pay more, simply because the application is worth more?
I think he’s calling out the fact that the justification for the API pricing changes was partially “we’ve got to stop AI training bots from scraping Reddit.” However, the pricing changes were actually made to hit third party developers hard and not hit the AI modelers that hard.