Lesson 4: Using AI Responsibly
Guard rails and copyright
Some AI tools (like ChatGPT) will have lots of guard rails to prevent copyrighted characters being shown.
However, AI models are trained on a lot of copyrighted materials and so, through the randomness of AI algorithms, copyrighted characters could appear in the response.
For example, including “iconic cartoon character” in a prompt to an image model could result in Mickey Mouse appearing in the image, even if his name is not mentioned directly.
Researchers found that one of Meta's LLM could recall 42 percent of the first Harry Potter book word for word, through some clever prompts (article on Arstechnica).