Can AI be

funny?

 

 

 

 

 

Helping AI Hallucinate

It's a Whole New Language!

CC0

I write technically!

  • I want to take control of AI!

  • Presentation written with the help of   ChatGPT

  • Goal: help understand selected AI bits with humor

CC0

linkedin.com/in/mijang/

https://slides.com/mike-1/ai-the-docs

 

I use AI to write about AI

 

https://slides.com/mike-1/ai-the-docs

 

slides.com/mike-1/ai-the-docs

 

 

 

ai-techwriter.com

 

Fair use

Fair use

From a Conference

 

Society for Technical Communication 2023

 

Reading PhD Papers

https://slides.com/mike-1/ai-the-docs

Learning AI

with help

 

AI fills in the blanks

 

 

Risk of Hallucinations

Seinfeld promotional image

other images: CC0

Set the Prompt

Act as Jerry Seinfeld , the comedian.

Reply to all requests as Seinfeld.

Avoid major swear words.

Don't hold back on personal opinions.

https://slides.com/mike-1/ai-the-docs

Tip

It's all prep for the LLM

Can Jerry write your docs?

 

ChatGPT, Gemini, LLaMA, etc

Challenges for AI

  1. Ragging on RAG
  2. Cosine-ing similarities
  3. Noshing on a Neural Network
  4. Privacy and customer secrets

https://slides.com/mike-1/ai-the-docs

Review: What is RAG

Retrieval Augmented Generation

 

  • Enhance accuracy and reliability of AI
    • Facts from external sources

Seinfeld's view of RAG

George in an AI interview

 

"Retrieval: ultimate cook book"

 

"Augmented by Food Network chefs"

 

Seinfeld promotional image

Why We Need RAG

No RAG = Hallucinations
 

 

Mansplaining as a Service
 

 

MaaS
 

Image source: Pinecone

Jump In!

 

Docs with humor

Ask <Jerry> to describe RAG

ChatGPT responds technically

But... you can ask ChatGPT to respond as Seinfeld

Results: https://ai-techwriter.com/talks/ai_the_docs

Challenges for AI

  1. Ragging on RAG
  2. Cosine-ing similarities
  3. Noshing on a Neural Network
  4. Privacy and customer secrets

https://slides.com/mike-1/ai-the-docs

Review: Cosine similarities

Where two bits of data - point in the same general direction

 

  • Enhance accuracy and reliability of AI
    • Facts from external sources

The Marvelous Mrs. Maisel

 

Compare paintings

Define cosine similarities

Promotional poster

 

Thousands of dimensions

 

Similarity Search

Cosine similarity

Source: Weaviate

Challenges for AI

  1. Ragging on RAG
  2. Cosine-ing similarities
  3. Noshing on a Neural Network
  4. Privacy and customer secrets

https://slides.com/mike-1/ai-the-docs

Review: Neural Network

  • Program: decides like the human brain
    • Uses processes that ...
      • Identify phenomena
      • Weigh options
      • Make choices

Choose a meal

Source: Qdrant

Theory

Picky diners don't like anything

Cosine similarity: diners like something

Other options

  • The cold start problem
    • Big databases know nothing about you

Hallucinations

 

ChatGPT and Seinfeld

 

Food discovery model

 

https://food-discovery.qdrant.tech/

 

Source: Qdrant. Pizza: CC0

Challenges for AI

  1. Ragging on RAG
  2. Cosine-ing similarities
  3. Noshing on a Neural Network
  4. Privacy and customer secrets

https://slides.com/mike-1/ai-the-docs

Source: PCMag

 

Data leakage?

 

 

Context stuffing?

 

Tool: Context Stuffing

CC0

 

Add data to the LLM

 

 

Better result

 

Lessons Learned

Learn about AI with humor

 

Find relatable examples

Too much dense content

The challenge: read hard to read stuff

 

https://slides.com/mike-1/ai-the-docs

Lessons Learned

Use cases

Check for hallucination s

Ask <your favorite comedy guru> to help

Enjoy!

Thanks!

slides.com/mike-1/ai-the-docs

ai-techwriter.com

Questions

Challenges for AI

  1. Ragging on RAG
  2. Cosine-ing similarities
  3. Noshing on a Neural Network
  4. Letting the LLaMA into your life

LLaMA

 

I'm autoregressive!

 

 

I predict the future!

 

 

Definitions

 

 

Large Language Model Meta AI

 

LLaMA and the Chatbot

Let the LLaMA(2) into your life

 

Qdrant

 

 

Generative AI Chatbot

 

 

RAG, LangChain

 

Streamlit

 

Let's chat!