Lecture 10: Creativity and Hallucination
Department of Data and Decision Sciences
Emory University
Source: Nielsen Norman Group
Source: X.com
What is an AI hallucination?
When an AI system generates content that is plausible but factually incorrect, presented with high confidence.
Key characteristics:
The unsettling part:
There’s often no signal that the AI is hallucinating. It sounds just as confident as when it’s correct!
The AI sounds confident, but is completely wrong
It’s not a bug, it’s how LLMs work!
Remember: LLMs predict the next word
The problem:
Source: Medium
LLMs are trained to be helpful…sometimes too helpful!
What is sycophancy?
The tendency to give users the answer they seem to want, rather than the accurate answer.
How this causes hallucinations:
Source: Zvi Mowshovitz (2025)
Not all domains are equally risky.
Research shows hallucination rates vary dramatically by topic:
| Domain | Hallucination Risk | Why? |
|---|---|---|
| Common knowledge | Low | Abundant, consistent training data |
| Coding | Low-Medium | Syntax is verifiable; patterns are clear |
| Recent events | High | Beyond training cutoff |
| Medical/Legal | High | Requires precision; errors are costly |
| Obscure history | Very High | Sparse data → interpolation |
| Citations | Very High | Specific strings rarely memorised |
Source: Chakraborty et al (2025)
| Type | Description | Example |
|---|---|---|
| Factual | Wrong facts | “Einstein was born in France” |
| Fabrication | Invented information | Made-up statistics, fake quotes |
| Citation | Non-existent sources | “According to Smith (2023)…” (doesn’t exist) |
| Logical | Contradicts itself | “It’s 5pm now… earlier today at 7pm…” |
| Temporal | Wrong timelines | Mixing up historical dates |
| Entity | Confusing people/things | Attributing quotes to wrong person |
What do you think is the most dangerous type?
Mata v. Avianca Airlines (2023)
The wake-up call:
“Is Varghese a real case?” Schwartz asked the chatbot. “Yes,” ChatGPT doubled down, “it is a real case.”
Source: CNN
Warning: Asking AI “Is this true?” doesn’t work, it will usually confirm its own hallucinations!
Healthcare hallucinations can be lethal:
Examples documented in research:
A 2024 study found:
Source: The New York Times (2025)
However…
The same mechanism that causes hallucinations also enables creativity!
Consider:
The spectrum:
Different tasks need different points on this spectrum!
Controversial take: Hallucinations can be great, too!
Consider these use cases:
The problem isn’t hallucination per se, it’s hallucination at the wrong time
Match the task to the tool. Use high-creativity settings for creative work, high-accuracy approaches for factual work (obviously!)
Red flags to watch for:
| Warning Sign | Example |
|---|---|
| Very specific numbers | “73.2% of studies show…” |
| Precise citations | “Smith et al. (2022), p. 47” |
| Obscure details | Historical minutiae |
| Too-perfect answers | Exactly what you wanted |
| Confident tone | No hedging or uncertainty |
Verification strategies:
Learn to recognise warning signs
Golden rule: The more specific and confident the AI sounds, the more you should verify!
LLMs only know what they learned during training
What happens when you ask about recent events?
| Option | What AI Does |
|---|---|
| Honest | “I don’t have that information” |
| Hallucinating | Makes something up! |
| Confused | Gives outdated info as current |
This is why chatbots now have:
LLMs are “frozen in time” at their training date
Preview of next class: RAG solves this by giving AI access to real, up-to-date documents!
RAG = Retrieval-Augmented Generation
The core idea:
Why this reduces hallucinations:
Next class, we’ll explore RAG in depth!
Source: Towards AI
Teaser: Tools like NotebookLM, ChatPDF, and Perplexity all use RAG!
