1:56:20 Let's build GPT: from scratch, in code, spelled out. Andrej Karpathy
6.8M views · 149.3K likes · 3 years ago
What Is Hallucination?
AI Hallucination
An AI hallucination occurs when a language model generates information that sounds confident and plausible but is factually incorrect or entirely fabricated. LLMs don't 'know' facts — they predict likely text sequences, which sometimes produces convincing nonsense.
How Hallucination Works
Common hallucinations: citing non-existent research papers, generating plausible but wrong code, inventing API endpoints that don't exist, or confidently stating incorrect dates or statistics. The model has no concept of truth — it generates probable-sounding text.
Mitigation strategies: RAG (ground responses in real documents), temperature reduction (less creative = fewer hallucinations), chain-of-thought prompting, citation requirements, and human review. No approach eliminates hallucinations entirely.
Key Concepts
- Confabulation — When the model fills gaps in knowledge with plausible but fabricated information — a core hallucination mechanism
- Grounding — Providing real documents or data in the prompt so the model bases responses on facts rather than imagination
- Uncertainty Calibration — Well-calibrated models express uncertainty when they're unsure — poorly calibrated ones state fabrications confidently
Learn Hallucination — Top Videos
1:56:20 
34:05 ChatGPT Tutorial - A Crash Course on Chat GPT for Beginners Adrian Twarog
6M views · 86.2K likes · 3 years ago
3:31:24 Hallucination Educators
Leila Gharani
@leilagharani
AI Coding
Excel. Power Query. Copilot. ChatGPT. Power BI. PowerPoint. You use them every day to automate Excel and your work - s...
3M Subs
625 Videos
117.7K Avg Views
3.06% Engagement
OpenAI
@openai
AI Coding
OpenAI’s mission is to ensure that artificial general intelligence benefits all of humanity.
1.9M Subs
456 Videos
36.2K Avg Views
2.18% Engagement
Academind
@academind
AI Coding
There's always something to learn! We create courses and tutorials on tech-related topics since 2016! We teach develop...
929K Subs
752 Videos
17K Avg Views
2.39% Engagement
Frequently Asked Questions
Can hallucinations be eliminated?
Not completely with current technology. They can be significantly reduced through RAG, better prompting, temperature tuning, and human review — but LLMs can always generate incorrect information.
Why do LLMs hallucinate?
LLMs are trained to predict the next likely token, not to determine truth. They don't have a fact database — they have statistical patterns. When the training data doesn't cover a topic well, the model extrapolates plausibly but incorrectly.
Want a structured learning path?
Plan a Hallucination Lesson →