Tumblr will sell data to OpenAI & Midjourney, to train AI: Report

Just after Reddit signed a massive deal to sell user data to train AI, it seems Tumblr might be about to do the same thing. A report from 404media alleges that an inside source has all but confirmed the deal. 404media has also reviewed leaked internal communications between Tumblr’s parent company Automattic, and the aforementioned AI companies. The report claims that users can choose to opt out of sharing their data.

An internal source has shared details about the impending deal

A source with inside knowledge shared the details with 404media, including internal communications. According to 404media, the reports make it clear that the deal is “imminent”. Automattic, also the owner of WordPress, has publicly stated that it will also share data from that site. The public statement does not mention which companies Automattic would be sharing data with.

In the documentation there was also talks of a mistake that happened recently. Apparently, Tumblr accidentally scraped a lot of user data that was supposed to be private or on deleted posts. This data scrape collected user data from 2014 to 2023, and will be given to OpenAI and Midjourney. Cyle Gage, product manager at Tumblr, writes, “the way the data was queried for the initial data dump to Midjourney/OpenAI means we compiled a list of all Tumblr’s public post content between 2014 and 2023”. It is only data from now on that users can choose to not share with the AI companies. Whether the accidentally scraped private data was shared as well remains unknown.

What Tumblr selling data could mean for AI

Tumblr is a social networking site centered around microblogging. Though this can take a lot of forms, it is almost all visual. OpenAI has an image generation AI model named DALL-E. Midjourney’s entire business model revolves around its image generation AI. To these companies, massive dumps of data detailing casual interaction is priceless. Tumblr and WordPress provide them with exactly that.

With data from Tumblr, both companies will see a massive boost in image generation capabilities. Furthermore, the specific data Tumblr provides is quite unique, especially in this volume. Memes, captioned images, and unique art styles not found anywhere else. Tumblr has been blocking AI crawlers for a long time now, as mentioned in Automattic’s statement. With this deal, the company can not only capitalize on their data, but perhaps also raise Tumblr’s worth. Yahoo bought the company for $1 billion in 2013, but Automattic acquired it for only $3 million in 2019.