AI Hallucinations Review

What is an AI hallucination?

• When AI dreams while processing data

• When AI generates false or misleading information

• When AI refuses to answer a question

• When AI only follows pre-programmed rules

Answer explanation

An AI hallucination refers to instances when AI generates false or misleading information, rather than accurately processing data. This can lead to incorrect outputs that do not reflect reality.

Which of the following are reasons AI might hallucinate? (Choose two.)

• Poor-quality training data

• AI getting bored and making things up

• AI always knowing the correct answer

• The way AI generates text can introduce errors

Answer explanation

AI may hallucinate due to poor-quality training data, which can lead to inaccuracies, and because the text generation process itself can introduce errors, resulting in fabricated or incorrect information.

Which of these is an example of an AI hallucination?

• AI translating a sentence correctly

• AI generating a historical event that never happened

• AI refusing to answer a question

• AI checking its response for errors before answering

Answer explanation

An AI hallucination occurs when the AI generates false information, such as a historical event that never happened. This distinguishes it from accurate translations or error-checking, which are correct AI functions.

What are ways to detect AI hallucinations? (Choose three.)

• Fact-check information from reliable sources

• Ask AI to rate its confidence in its response

• Trust AI completely since it's trained on good data

• Know the AI’s limitations and where its knowledge comes from

Answer explanation

To detect AI hallucinations, fact-checking from reliable sources ensures accuracy, asking the AI for confidence levels helps gauge reliability, and understanding its limitations clarifies potential knowledge gaps.

Give an example of an AI hallucination you have encountered or heard about. What was incorrect about it?

Evaluate responses using AI:

OFF

Answer explanation

An example of AI hallucination is when a model generates a false historical event, like claiming a non-existent battle. The inaccuracy lies in fabricating details that never occurred, misleading users about real history.

Which of these statements about AI hallucinations is FALSE?

• AI hallucinations can happen in both text and images

• AI always knows when it’s making a mistake

• AI can sometimes generate misleading or completely false information

• AI hallucinations can happen due to poor-quality training data

Answer explanation

The statement 'AI always knows when it’s making a mistake' is false. AI lacks self-awareness and cannot recognize its errors, leading to hallucinations in both text and images.

Which of these tools can help detect AI hallucinations? (Choose two.)

• Wikipedia or verified fact-checking sites

• Asking AI the same question multiple times

• Trusting everything AI says

• Looking at AI’s answer and assuming it’s correct

Answer explanation

Wikipedia and verified fact-checking sites provide reliable information to verify AI responses. Asking the same question multiple times can reveal inconsistencies, helping to identify potential hallucinations.

Why do you think AI hallucinations can be dangerous in real-world applications (e.g., news, medicine, law)?

Evaluate responses using AI:

OFF

Answer explanation

AI hallucinations can lead to misinformation, misdiagnoses, or wrongful legal decisions, causing harm to individuals and society. In critical fields like news, medicine, and law, accuracy is essential to prevent serious consequences.

What is a factual contradiction in AI hallucinations?

• AI randomly switching topics in an answer

• AI stating false information as fact

• AI misunderstanding the question

• AI generating multiple possible correct answers

Answer explanation

A factual contradiction in AI hallucinations occurs when the AI states false information as fact, misleading users. This is the essence of hallucination, where the AI confidently presents inaccuracies.

One way to reduce AI hallucinations is to improve the ________ AI is trained on. (Fill in the blank.)

Answer explanation

One way to reduce AI hallucinations is to improve the quality of data AI is trained on. High-quality data ensures that the AI learns accurate patterns, reducing the likelihood of generating incorrect or nonsensical outputs.