EXPLORING THE BOUNDARIES OF ARTIFICIAL INTELLIGENCE HALLUCINATION

AI systems are being increasingly used to perform a wide variety of tasks. Large organization typically build foundation models, which are models trained on very large sets of data. These models are then fine-tuned for specific tasks in specific domains. A major issue with using such systems are hallucinations which refers to issues like incorrect detection, incorrect fact generation, etc. Such issues have been widely studied for identifying potential mitigation. In this study, we come up with a two category classification for the causes of hallucination. We then discuss about the ways to handle such hallucination. We briefly look at ways to identify hallucination and then discuss potential future research areas to reduce the hallucination.