AWS Certified AI PractitionerAI/ML and Generative AI FundamentalsMedium
An AI/ML team has developed a large language model (LLM) and deployed it for customer service. After initial deployment, they notice that the model sometimes produces responses that are factually incorrect or 'hallucinates' information not present in its training data. Which concept describes this unwanted behavior in generative AI models?
- ABias
- BUnderfitting
- CHallucination
- DOverfitting
Show answer & explanationAnswer & explanation
Correct answer: C. Hallucination
Hallucination in generative AI, particularly LLMs, refers to the phenomenon where the model generates content that is factually incorrect, nonsensical, or not grounded in its training data or the provided context. This is a common and significant challenge in deploying LLMs.
Why the other options are wrong
- A. Bias refers to systematic errors in a model's predictions due to unrepresentative or prejudiced training data, leading to unfair or inaccurate outcomes for certain groups, which is different from generating false facts generally.
- B. Underfitting occurs when a model is too simple to capture the underlying patterns in the training data, resulting in poor performance on both training and unseen data.
- D. Overfitting occurs when a model learns the training data too well, including noise, and performs poorly on unseen data. While related to model generalization, it doesn't specifically describe generating false facts.
Hallucination (Generative AI)
The phenomenon where a generative AI model, especially a Large Language Model (LLM), produces content that is factually incorrect, nonsensical, or ungrounded in its training data or provided context.
- Common issue in LLMs.
- Outputs can sound plausible but be false.
- Challenges model trustworthiness and reliability.
- Mitigation involves RAG, fine-tuning, better prompts.
Memory trick: LLMs can err, sometimes creating air.