Reddit has reportedly signed a $60 million deal with an unnamed AI biz to hand over user conversations for model training.
The deal comes as Reddit looks to boost interest in its upcoming IPO. Reddit reportedly told prospective investors about the $60 million contract earlier this year, and indicated that its execs may repeat this type of content-sharing-for-model-training deal in the future.
Bloomberg, citing "people familiar with the matter," noted that both the IPO and AI deal details are subject to change and the site's expected listing might now happen as soon as March. Reddit did not immediately respond to The Register's inquiries.
The site's users, on the other hand, had plenty to say about the rumored $60 million deal. Comments ranged from "Reddit is asking WAY too little" to why would anyone pay tens of millions of dollars for 'shitposts' and "obscure horror artwork."
More cynical posters recalled Reddit's plan to charge for API access, which led to a limited user revolt and some forums going private or shutting down.
"Good to know that Reddit's API clampdown was never because they wanted to protect your data from AI usage," one user noted. "They were just protecting it from unpaid AI usage. Welcome to the dullest cyberpunk hell."
Reddit wouldn't be the first company to train AI models on content without seeking permission from whomever generated the data, and scraping text and images to train models has spawned several lawsuits. But if Reddit has covered this kind of data collection properly in the site's terms and conditions this shouldn't be a problem.
To avoid more of these types of legal battles, OpenAI has secured licensing agreements with the Associated Press, Axel Springer, and is reportedly in talks with CNN, Fox Corp, and Time expressly to use these media org's articles.
However, while Reddit's a superb time suck for burrowing deep into obscure rabbit holes about things like grilled cheese sandwiches and screaming fish, it's not exactly known for fact-checked, accurate information. So this particular deal, if true, does make it seem like an odd choice of content for training data.
Or, as one Reddit user commented: "gonna be one hell of a stupid AI." ®
Source: The register