Instances where an AI language model generates text that is convincingly wrong or misleading, like the AI is confidently presenting false information as if it were true.

• Development of mitigation strategies to recognize and correct errors in language models
• Improving the overall utility of AI language models by identifying and filtering out irrelevant or nonsensical responses
• Enhancing the credibility of AI language models by developing more accurate and reliable information sources

An LLM hallucinates when it generates a description of a fictional city called ‘Centauri’ with detailed maps and population statistics, despite there being no such place in reality.

• Jailbreaking LLM research work