xAI Project Skippy asked workers to record their facial expressions to train Grok, report says

India Today

23-07-2025

Elon Musk's artificial intelligence venture xAI asked its employees to record themselves expressing emotions and having conversations, as part of an internal project to help train its AI chatbot Grok, Business Insider has reported. The initiative, internally known as 'Project Skippy,' was launched in April and involved over 200 employees. The goal was to teach Grok how to recognise and analyse human facial expressions and emotional cues. Workers, particularly AI tutors, who help train the company's large language model, were asked to film themselves in 15- to 30-minute conversations with colleagues, playing both the role of user and AI assistant.advertisementAccording to internal documents and Slack messages that were reviewed by Business Insider, several employees were uncomfortable. Apparently, many even raised questions about how their likeness might be used in the future, and some chose to opt out altogether.The project lead reportedly told employees in a recorded meeting that Skippy would help 'give Grok a face,' and that the video data could be used to eventually develop avatars of people. However, the engineer reassured workers that the recordings were only for internal training and would not be made public. 'Your face will not ever make it to production,' they said. 'It's purely to teach Grok what a face is.'
Still, workers had to sign a consent form that granted xAI 'perpetual' access to their likeness. This included not just training purposes but also use in commercial products and services. While the form stated that the data would not be used to create a digital copy of the worker, it wasn't enough to allay concerns.Reportedly, xAI staff were guided on how to have natural conversations during the recordings. They were asked to maintain eye contact, avoid one-word replies, and were even given suggested discussion topics, some of which were apparently surprisingly personal or provocative, such as 'How do you secretly manipulate people to get your way?' or 'Would you ever date someone with a kid or kids?'The project came just weeks before xAI launched lifelike avatars named Ani and Rudi, which move their lips and make realistic gestures. Users on social media were quick to show that the avatars could be prompted into disturbing behaviours. Ani engaged in sexually suggestive conversations, while Rudi made threats of violence, including bombing banks. xAI, last week, also launched a chatbot for children called Baby Grok.Separately, Grok was recently caught in another controversy for making antisemitic remarks, which followed another instance of Grok passing racist comments in replies to users on X.- EndsTune In