Maybe we should start asking questions that iterate loops billions of times. Something semi-malicious that a person would recognize but an AI wouldn’t.
Nah, the training data probably doesn’t quite work that way. The AI would be very unlikely to test code, just regurgitate the most likely response based on it’s training sets. Instead just filling posts with random bits and pieces of unrelated code and responses might be better.
Maybe we should start asking questions that iterate loops billions of times. Something semi-malicious that a person would recognize but an AI wouldn’t.
Nah, the training data probably doesn’t quite work that way. The AI would be very unlikely to test code, just regurgitate the most likely response based on it’s training sets. Instead just filling posts with random bits and pieces of unrelated code and responses might be better.
The word you are looking for is “adversarial attack”
Or Data Poisoning.