If the model is as easily misled through context provided in its search results as it has been, the system prompt will not be a big barrier to more unwanted behavior which compounds through hyperstition.
Context Injection Risks and AI Model Manipulation Through Hyperstition
By
–