Wikipedia: Agreements with Mistral, Perplexity & Co. for AI training access
For a long time, AI companies relied on Wikipedia content for AI training, increasing server load. Now, more and more are using an alternative.
(Image: Allmy/Shutterstock.com)
Just in time for Wikipedia's 25th birthday, the Wikimedia Foundation announced that several AI companies are now also paying to train their models with content from the popular online encyclopedia. Microsoft, Mistral AI, Perplexity, Pleias, and ProRata have now signed licensing agreements for specially established access via Wikimedia Enterprise, according to the statement. In addition, the German search engine provider Ecosia is on board. Amazon, Google, and Meta had already used the interface, the company further writes. Together, they support Wikipedia and can access content at the required speed.
Less than a year ago, the Wikimedia Foundation complained that scrapers for training AI models were responsible for a drastic increase in bandwidth for multimedia content downloads. It was particularly problematic that they continuously accessed all content, including that which was rarely retrieved. Wikipedia, it was said at the time, was actually designed to be prepared for sudden increases in interest in individual content. The AI scrapers were then blamed for longer loading times for all visitors during peak access times, simply because the baseline load was too high.
Videos by heise
In the fall, the Wikimedia Foundation therefore began to urge AI companies to use the separate interface (API) created for automated, structured queries for their access. The tariffs for this are not public, and even now, the Wikimedia Foundation does not disclose figures on the revenue that the new contracts promise. However, the approach is likely to pay off for them if the baseline load for access decreases and the associated costs become lower. It took some time to understand exactly what the companies needed, Lane Becker from Wikimedia Enterprise told the news agency Reuters. But now all partners see that they have to commit to maintaining Wikipedia.
(mho)
