Elon Musk wanted Grok developers to give the chatbot «their own faces» — leaked internal correspondence

Elon Musk once thought, why not force his employees to give their appearance to Grok AI? However, the scheme did not work with everyone.

The internal Skippy project stated that xAI employees should help Grok chatbot learn to recognize facial expressions and emotions. Musk wanted the chatbot to learn to understand what human emotions look like: how people react, communicate, and express themselves. To do this, you just had to give your face to be transformed into a «human avatar».

If you think the idea sounds dubious — you’re wrong. Dozens of xAI employees who refused to participate in the project thought the same way, as it became known from leaked correspondence. Although the company promised that the videos would remain inside xAI and would be used only for training, some employees still did not sign the consent form. They were worried that their face might appear in an avatar that would say something they had never said.

This reaction may have been exacerbated by the recent scandals surrounding Grok. For example, the bot recently expressed anti-Semitic ideas and praised Hitler. And xAI is also open looking for developers to create anime companions — with flirting and hints of a relationship with the user. This could also raise doubts. So to speak, a developer can write thousands of pages of code, but once he becomes an anime waifu —, everyone will remember him as an anime guy.

However, not all employees refused. Those who agreed had to record 15-30-minute conversations in pairs: one was «the host», the other — played the Grok user. In xAI, they were asked to speak at length, ask additional questions, and maintain eye contact. This was necessary to collect «imperfect» data — with real emotions, different quality, and ambiguous reactions. The company believed that this would help Grok to work better in real life.

To evoke the right facial expressions, employees were even offered topics for conversation. Sometimes they were provocative, for example: «How do you secretly manipulate people to get what you want?» or «Would you ever date someone who has a child or children?» etc.

Although the company has assured that it will not create digital copies of participants, doubts remain. Still, the role of cybersecurity is not exaggerated, so the risks of data theft remain. If One weak password can destroy a companydigital identity theft has many risks.

Source: Business Insider