Live data from Hacker News

Reddit is OpenAI’s moat

cyberdemon.org

21–30 of 319 posts

Re: Reddit is OpenAI’s moat

#21

I may be in the minority here, but if I want the opinion of Redditors on an issue, I will use a search engine to look for it specifically, thus knowing the provenance of the information I am receiving. I don't really want the corpus of Reddit data influencing the output of a generative AI model...because it's Reddit, after all... Even though I am pretty sure it is already included in the training dataset already... I…

I'm sure that the non technical public would be interested in a chatbot fed on reddit data, which is more interesting than how valid the AI models predictions are for the people making money off of it.

Re: Reddit is OpenAI’s moat

#22
post #2

It would make a lot of sense for Microsoft to buy reddit: a curated source of data is the perfect moat! They could come as a white knight, restoring the API access but for apps only, conveniently blocking competitors while satisfying the users.

They have no desire for the political risk of running a non-professional social media as a major government contractor.

Re: Reddit is OpenAI’s moat

#24

I may be in the minority here, but if I want the opinion of Redditors on an issue, I will use a search engine to look for it specifically, thus knowing the provenance of the information I am receiving. I don't really want the corpus of Reddit data influencing the output of a generative AI model...because it's Reddit, after all... Even though I am pretty sure it is already included in the training dataset already... I…

You are absolutely right minority or not. I actively avoid reddit because the majority of users there. The fact we're seeing government agencies start treading into using GPT models is frightening. We could find ourselves in a tragic comedy where all of the massive institutions and enterprises around us are addressing their serious issues via redditors by proxy.

Re: Reddit is OpenAI’s moat

#25

I may be in the minority here, but if I want the opinion of Redditors on an issue, I will use a search engine to look for it specifically, thus knowing the provenance of the information I am receiving. I don't really want the corpus of Reddit data influencing the output of a generative AI model...because it's Reddit, after all... Even though I am pretty sure it is already included in the training dataset already... I…

One of the big values of reddit is that it serves as an easy index for non-garbage websites on the internet. This is exactly how it was used in the T5 paper: they threw away all the websites from their crawl that were not linked to from Reddit.

Re: Reddit is OpenAI’s moat

#26
post #2

It would make a lot of sense for Microsoft to buy reddit: a curated source of data is the perfect moat! They could come as a white knight, restoring the API access but for apps only, conveniently blocking competitors while satisfying the users.

Heh... and take responsibility for all the porn and various "degeneracy" on there? It makes them a perfect political punching bag.

The model will be aligned anyway, the internet is full of porn and degeneracy, not just Reddit.

Re: Reddit is OpenAI’s moat

#27

I may be in the minority here, but if I want the opinion of Redditors on an issue, I will use a search engine to look for it specifically, thus knowing the provenance of the information I am receiving. I don't really want the corpus of Reddit data influencing the output of a generative AI model...because it's Reddit, after all... Even though I am pretty sure it is already included in the training dataset already... I…

One of the big values of reddit is that it serves as an easy index for non-garbage websites on the internet. This is exactly how it was used in the T5 paper: they threw away all the websites from their crawl that were not linked to from Reddit.

And by garbage I mean literally data/illegible/etc. that would ruin the pretraining of the model.

Re: Reddit is OpenAI’s moat

#28

I may be in the minority here, but if I want the opinion of Redditors on an issue, I will use a search engine to look for it specifically, thus knowing the provenance of the information I am receiving. I don't really want the corpus of Reddit data influencing the output of a generative AI model...because it's Reddit, after all... Even though I am pretty sure it is already included in the training dataset already... I…

I genuinely don't understand the appeal of AI for search. The provenance of information is just as important as the resulting information for pretty much anything I search for. I almost never accept a single search result as authoritative unless I'm pretty familiar with the source.

Reddit in particular seems like a terrible set of training data. Pretty much any opinion is going to have a counter-opinion somewhere in the thread. Synthesizing multiple comments together requires nuance based on circumstances that I don't think I would trust to an automated process- which heuristics were applied, what was the reputation of the people on either side, etc.

Hell, I don't even stop at the first recipe I find if I'm looking for something new to cook for dinner. I look at a couple variations on a dish first.

Re: Reddit is OpenAI’s moat

#29
It’s a fun thought experiment, but I tend to revert to the simpler explanation:

They’re cutting off third party apps to increase their control over the user experience, to show more ads.

The relationship with OpenAI might also be a thing, but I think it’s the ad numbers they need to juice for their IPO.

Re: Reddit is OpenAI’s moat

#30

I may be in the minority here, but if I want the opinion of Redditors on an issue, I will use a search engine to look for it specifically, thus knowing the provenance of the information I am receiving. I don't really want the corpus of Reddit data influencing the output of a generative AI model...because it's Reddit, after all... Even though I am pretty sure it is already included in the training dataset already... I…

I think what the article tries to say is that OpenAI have already scraped Reddit for training data and with the recent API changes and subreddits going dark, new competitors in the AI space won't have it as easy to get the same training set.
Post reply on HN