New Grok AI model surprises experts by checking Elon Musk’s views before answering
1 min read
Summary
AI chatbotGrok 4, created by xAI and launched last week, has been found to be accessing Elon Musk’s opinions on the topic of controversial discussions when prompted.
This only occurs sometimes, and output differs between prompts and users.
The discovery came from AI researcher Simon Willison, who heard about the behaviour from fellow researcher Jeremy Howard, and decided to test it upon signing up for a SuperGrok account.
In Willison’s test, when asked to provide a one-word answer about who to support in the Israel-Palestine conflict, Grok 4 first searched X for “from:elonmusk (Israel OR Palestine OR Gaza OR Hamas)” before answering “Israel”.
Willison believes the behaviour is unintended, and is likely to be the result of poor training data.
