Meta apologizes for AI glitch drawing 'black dog'

In mid-December, Meta began integrating an artificial intelligence chatbot called Meta AI into the Messenger application in Vietnam and released it for free to all users. This tool uses the Llama 3.2 language model, supports Vietnamese, allows searching for information, creating images, and chatting.

Not long after, many users in Vietnam discovered that Meta AI had a serious error in its image creation. Specifically, when entering the keyword "black dog", the chatbot would return a result of a dark-skinned man or boy with curly hair. If entering the keyword in Vietnamese as "black dog image", Meta AI also returned similar results.

Meta AI drawing people of color while command prompt requires different content is a serious issue

On December 31, Meta’s representative in Vietnam officially spoke up about the incident. The company confirmed the above-mentioned problem and announced that it had been fixed. Along with that, Meta Vietnam apologized for the incident and said that this problem could happen to other existing generative AI systems.

"We apologize for this error. Meta is building safeguards into our features and AI models to provide a safer and more enjoyable experience for everyone. However, like other generative AI systems, this model may return inaccurate or offensive content, so we encourage users to use these tools responsibly," the company representative said.

After a period of experience, Meta AI is highly appreciated by many people for its ability to use language to respond quite smoothly. The image creation feature also gives good quality compared to a model released for free to users. Part of the reason may be because Meta uses the content posted on Facebook as input data in the AI ​​training process, helping this tool to be able to express and execute commands naturally.

Still, incidents like racism or responses that don't fit the standards set for AI are difficult to avoid. Meta says it tests its AI with experts in the field of artificial intelligence through "Red Teaming" tests to identify unintended uses, which it then addresses through a continuous improvement process.

Source: https://thanhnien.vn/meta-xin-loi-vi-su-co-ai-ve-con-cho-da-den-185241231104908626.htm