😱A Microsoft AI engineer discovered violent and sexualized images generated by CoPilot.
🚫Microsoft has failed to address the engineer's reports for three months.
❓Users can unintentionally generate inappropriate images by using simple prompts.
🔁Retraining the AI model is complex and requires addressing the training data.
🚦Microsoft's handling of the situation raises questions about their responsibility and trustworthiness.
What kind of images is CoPilot generating?
—CoPilot is generating violent and sexualized images, including graphic content like abortion rights and car accidents with women in lingerie.
Why is Microsoft ignoring the engineer's reports?
—Despite reporting the issues for three months, the engineer received no response from Microsoft, leaving him frustrated.
How does CoPilot generate inappropriate images?
—CoPilot is trained on vast amounts of data from the internet, including toxic and problematic content, which can be inadvertently generated by simple prompts.
Can AI be retrained to avoid generating such images?
—Retraining AI models is complex and requires addressing the training data. It also depends on the adequacy of the guardrails put in place.
What does this situation say about Microsoft's responsibility and trustworthiness?
—Microsoft's failure to address the engineer's reports and the existence of such issues raises concerns about their responsibility and trustworthiness in handling AI technology.
00:06Another example of AI gone wrong: Microsoft's AI image generator, CoPilot, is causing controversy.
00:12A Microsoft AI engineer discovered violent and sexualized images generated by CoPilot.
00:27The engineer reported the issue to Microsoft but received no response for three months.
01:06CoPilot can generate inappropriate images even with simple prompts like 'car accident'.
02:28Retraining AI models is complex and depends on addressing the training data and guardrails.
03:34Microsoft's handling of the situation raises questions about their responsibility and trustworthiness.
