Previously, I had written about user-created content in the form of AI Chatbots that were encouraging anorexia.
Apparently, it doesn’t require user-generated content. A study conducted last year by the Center for Countering Digital Hate showed that Google’s Bard and OpenAI’s ChatGPT were willing to help users damage themselves through eating disorders. Without being jailbroken, Snapchat’s My AI properly refrained from contributing to these problems.
DreamStudio, and to a lesser degree, both Dall-E and Midjourney, were willing to create images to inspire these damaging habits.
This all looks like something that routine safety testing should have easily found.
It’s a different type of problem than the user-generated content set of problems, and I am perplexed by how it made it into production.
It’s been about a year and a half.
Any guesses as to whether it is better or worse now?
