Can someone explain what value new Reddit data has for LLMs?

Presumably the entire 2005-2023 (up to March) dataset which is (was) freely available on Pushshift / torrents is enough?

Also, not that Sam Altman is unbiased, but he did say that social media (Reddit, Twitter) were not super important for developing LLMs (But of course, that's exactly what he would say if they were useful!)

Depends what you're training it on, I suppose. If you're just wanting to train something to generically talk then sure, a slightly outdated data set is fine. If you're wanting it to be able to talk about recent things then it'll, naturally, not be fine. It's not so much a problem right now, only being a couple of months out of date, but once that turns into years it'd become more apparent.

But sure, there are plenty of other data sources as well. On top of the problem of not wanting to train your data on content you know is littered with bots.

I'm now envisioning a sci-fi future where the AIs are an embodiment of xkcd #606. Far better than humans at mundane tasks but with significantly outdated cultural knowledge.

There are humans today that refer to Africa as the 'third world' even though it's a continent with countries more developed than some US states or European countries.

Significant outdated cultural knowledge seems the norm for humans too.

Are they using "third world" in its original sense of non-alignment?

https://en.wikipedia.org/wiki/Third_World

As someone in a US state with blackouts to our power grid, I'm wondering what more developed things you might be referring to and I might be missing out on.

I didn't even know it was available on torrents.

> Still, this could have been avoided by providing free-use keys to trusted developers.

This approach mostly doesn't work because, for third-party client apps that don't go through the developer's own server, scrapers could snag those keys from the client. Scraping has always been a thing and I don't see a foolproof solution that allows good actors to use the product without bad actors getting to scrape content (not that scrapers won't settle for html scraping).

Wouldn't you leave the API key, in the server and proxy thru? That's what I was taught to do, and you give your own tokens to clients to track which client is making that request and block that client if needed. It seems very dangerous to pass on the reddit API key to the client.

with Oauth2, the API credentials don't give you access to other users' accounts or data or anything of that sort - so the only problem with the client id/secret being in the client is malicious people pulling it and abusing the service while using those API credentials. In general this hasn't been a problem, but once it costs (for example) Apollo $x per million API calls, it could be a money drain, and it could cause headache for the service that now has to try to do abuse prevention via IP reputation / other tracking techniques.

And proxying api calls isn't an end-all solution. This introduces a large burden for the developer to not just run a scalable proxy for the API calls, but also must introduce and manage some abuse prevention system, maybe with their user account system with reputation ratings based on number of times they've hit a rate limit or how many different subreddits they try to access every day.

It also is generally not good for users to have to deal with a proxy like this, since it means the app developer has the ability to log all of their users' usage, and could be compelled by law enforcement / courts to hand over user PII or activity logs.

How would those api keys be exposed if it’s an app?

Inspecting the app's content or the network traffic.

Decompiling apps even on iOS is not too hard, typically API keys would either be in resource files or exposed via strings in the binary itself.

There's pretty much no scenario where Reddit wants to encourage 3rd party app developers to run their own intermediate "proxy" service for Reddit's API. For one it adds another barrier between Reddit and their own users, and two it would mean that there's just one more small step for the app developer to take to just implement the backend services themselves and stop using Reddit's APIs completely. I see a lot of suggestions already that app developers should do this. This of course is completely impractical for anything more than a trivial number of users unless the developers want to start paying millions to run their own infrastructure instead of just paying for access to Reddit's backend systems.

While I have issues with mods of various forums being little hitlers, I can see why they are rebelling, evidently only the 3rd party apps provide a decent way to mod on mobile. They're already working many hours for free, why would they put up with using crappy tools to do a job they do for fun and for free?