There’s word going around that X just enabled a setting that lets it train Grok on public tweets, as well as any interactions they have with the chatbot. That’s not entirely true: a help page instructing users how to opt-out of X using their data to train Grok has been live since at least May. X just never exactly made it crystal clear that it was opting everyone into this, which is a sketchy move. If you don’t want a bad chatbot to use your bad tweets for training, it’s thankfully easy to switch that off.
You just need to uncheck a box from the Grok data sharing tab in the X settings. If that link doesn’t work, you can go to Settings > Privacy and Safety > Grok. For the time being, the setting isn’t accessible through X’s mobile apps (the company says it will be soon), so you’ll have to uncheck the box on the web for now. It’s also worth noting that Grok isn’t trained on any tweets from private X accounts.
All X users have the ability to control whether their public posts can be used to train Grok, the AI search assistant. This option is in addition to your existing controls over whether your interactions, inputs, and results related to Grok can be utilized. This setting is…
— Safety (@Safety) July 26, 2024
One of X’s selling points for Grok when it rolled out the chatbot was that it had the advantage of using real-time information that’s published on the platform — in other words, users’ tweets. That only works if users opt-in or are automatically enrolled into sharing their data with the chatbot. But X isn’t exactly the pinnacle of truth and accuracy. It’s full of pranksters, and lifting their jokes might be one of the reasons why Grok keeps on getting stuff wrong. In any case, it’s not exactly uncommon for AI models to be trained on material without explicit permission from the original creators.