Mike Jang (NGINX): Helping AI Hallucinate – It's a Whole New Language!
 

This talk was presented at the AI The Docs online conference on April 4, 2024. We are thrilled to share the recording and the summary with you.

Visit the talk summary page to see all of the presentations from the conference.

Mike Jang

Principal Technical Writer at NGINX

Mike's presentation

In this presentation, Mike Jang shares his personal journey with AI and how he leverages it for learning and writing.

Key Takeaways

Understanding AI Hallucinations and Practical Use

• AI Hallucinations: Mike discusses how AI models like ChatGPT can produce incorrect or misleading information, which he finds both challenging and useful for learning.
• Humor as a Learning Tool: He uses humor and comedic perspectives (e.g., Jerry Seinfeld) to simplify complex AI topics, making them more relatable and easier to understand.

Technical Aspects of AI

Retrieval Augmented Generation (RAG)

• Definition: RAG involves combining data retrieval from reliable sources with real-time data augmentation to enhance AI responses.
• Application: Mike explains how RAG can reduce the problem of outdated or hallucinated information in AI models. 
• Cosine Similarity: Cosine similarity measures the similarity between two pieces of data by turning them into vectors and comparing them.

Privacy and Data Leakage

• Issue: Addresses concerns about data leakage through AI tools, using the Samsung case as an example where proprietary code was exposed.
• Solution: Emphasizes the need for careful management of proprietary information and effective context-stuffing techniques to maintain privacy.

Lessons Learned

• Key Insights: Mike shares that learning AI can be enhanced by using humorous and relatable examples, and underscores the importance of context-stuffing to avoid technical jargon.
• Effective AI Documentation: Highlights the need for documentation that is tailored to user needs and includes practical use cases.

Conclusion:
Mike concludes by reaffirming the value of a humorous and contextual approach to understanding AI. He invites the audience to explore further resources on his professional site and emphasizes the importance of balancing technical accuracy with engaging content.