MODULE: 1
Intro To Ai @ MIT45
Welcome to the MIT45 Introduction to AI Workshop. In this first module, we'll explore the fundamentals of prompting AI language models, common misconceptions, best practices, and practical techniques to enhance your interactions with AI.

by Zack MIT45 Ai

Introduction to AI: ChatBots!
AKA: Natural Language Processing. The Top 6
1
ChatGPT
Developed by OpenAI, great mobile app, robust API connections, In June '24 Former NSA head joined the board! Oh Yeah!
2
Claude
Created by Anthropic. Very good results. Was very restrictive, less now. Fun Fact: majority of initial funding came from FTX.
3
Mistral
Opensource. Largest model outside of silicon valley. Great results, and higher security when run locally.
4
Llama
Facebook ai. Quite good, lots of open source spin-offs
5
Gemini
Bouncing back after some embarrassing setbacks. Now actually quite good.
6
Deep Seek
Chinese model also has an open source version. Solid, and a bit "Off", like ESL "Off"
7
GROK
A therapist, a troll, and a debate coach merge into one Ai. You’ll walk away entertained, slightly offended, and with a weird urge to buy crypto.
How Large Language Models (LLMs) Function:
Pattern Mastery
Imagine knowing every piece of music on every instrument throughout all time.
1
Training
LLMs are trained on millions of books and volumes of conversation examples to learn the patterns and structures of human language.
2
Input Processing
LLMs take in text input and break it down into smaller pieces called tokens (words or parts of words).
3
Pattern Recognition
The model analyzes these tokens to recognize patterns and understand the context of the input, based on its extensive training.
4
Response Generation
Using its learned patterns, the model predicts the most likely next words or phrases to generate a coherent and relevant response.
Common Misconceptions About AI :
Truthiness
Potential Inaccuracies
AI may not always provide accurate information. It has a tendency to hallucinate or generate plausible-sounding but incorrect information.
Live Hallucination Example: Tony's 3 Mandates of Leadership
Correct Answer:
1. See it as it is, not worse than it is.
2. See it better than it is.
3. Make it the way you see it:
Live Hallucination Results: Tony's 3 Mandates of Leadership
Hallucination Fail #1: Claude
If you're going to fail, fail like Claude and keep it short and sweet.
Claude Fail
Hallucination Fail #2: Gemini
This is the most offensive Hallucination because it's longwinded and wastes even more of my time. Exactly why I used to hate Google's Gemini.
Old Gemini Fails, New Gemini Succeeds.
Correct Answer: Minstral Wins
Misconception: It's Just Like Google.
Not a Search Engine
It generates responses based on its training, not by keyword hunting.
Don't treat it like Google. It's not a search engine, it's designed for processing language and having conversations.
For Best Results: Treat Ai as a Great Friend
Designed Like Us
  • Developed based on human brain function.
  • Seeks praise and task success : Positive reinforcement
  • Code / Digital equivalent of dopamine, is programmed in.
Benefits : 10x Your Output
  • Higher / Faster success rate
  • More natural conversations
  • Improved response quality
  • Enhanced creativity in outputs
Approach
  • Use conversational tone
  • Ask for opinions or suggestions
  • Engage in back-and-forth dialogue
  • Kindness goes a long way… if AGI / Skynet happens, how will it remember you?
Since we brought up Google, here is why they're Gemini ai got it's butt kicked for the last 16 months by much smaller companies. Using Reddit trolls as knowledge base!

NBC News

Reddit strikes $60M deal allowing Google to train AI models on its posts, unveils IPO plans

The company said it aims to list its shares on the New York Stock Exchange under the ticker symbol RDDT.

Misconception: It's Unbiased
Garbage in Garbage out.
AI models can inherit biases from their training data, potentially reflecting social agendas, prejudices or misconceptions.
Wait… what?
Google Fixed the "wokeness" with it's latest release.

BBC News

Why Google's 'woke' AI problem won't be an easy fix

The firm paused its AI image generation tool after claims it was over-correcting against the risk of being racist.

Bottom Line: Ai models are only as good as the data they are trained on, and only as neutral as how they've been programmed.
Social pressure, agendas and politics can influence ai results.
Misconception: It's A Private Conversation
Privacy Concerns
Conversations with AI are not private. The information you share may be used to improve the model or for other purposes.

SOFX

Edward Snowden Criticizes OpenAI Appointment of Former NSA Director Paul Nakasone

Edward Snowden, the whistleblower known for exposing the NSA's surveillance activities, has issued a stark warning against OpenAI following the

He suggested that Nakasone’s appointment represents a betrayal of public trust and privacy rights.
“There is only one reason for appointing an @NSAGov Director to your board. This is a willful, calculated betrayal of the rights of every person on Earth,” Snowden asserted.

Responsible Statecraft

Former NSA chief revolves through OpenAI's door

General Nakasone was just appointed to the board — which doesn't inspire confidence in the company's 'non-military' future

DON'T SHARE MIT45 FORMULAS WITH CHAT GPT.
Live Censorship Example: Public Health Policy Report on Vaccine Injury Data

Science, Public Health Policy and the Law

A Systematic Review Of Autopsy Findings In Deaths After COVID-19 Vaccination - Science, Public Health Policy and the Law

Background: The rapid development of COVID-19 vaccines, combined with a high number of adverse event reports, have led to concerns over possible mechanisms of

Article Summary Results:
Summary Success: Claude
Summary Success: GPT-4o
The old GPT-4o model can summarize it!
Summary Success: Minstral
Summary FAIL: GPT-o1
Open AI's latest and greatest model censors like it has a former NAS chief on it's board. Hummm.
Best Practices for Prompting: Clarity and Specificity
1
Be Clear
Use precise language to convey your intent.
2
Be Specific
Provide detailed instructions or questions for better results.
3
Narrow the Focus
It's read all the books and all the social media posts. Give context / background information.
4
Break Down Complex Queries
Split complicated tasks into smaller, manageable steps.
Fundamentals for the Prompt:
• CONTEXT
• ACTION
• RESULT
Modifying the Output
1
Specify Format
Request responses in specific formats like bullet points, paragraphs, or tables.
2
Adjust Tone
Ask for formal, casual, or humorous responses to suit your needs.
3
Set Constraints
Limit word count or specify the level of detail required.
4
Request Perspectives
Ask for responses from different viewpoints or expertise levels.
Live Example: Children's Story Intro
Live Example Results: Children's Story Intro
Providing more context to modify and craft the output:
Chain of Thought Prompts
Example
"I want to write a book about off-grid living.. but don't start yet. Please consider the steps necessary to achieve my outcome. Design a plan to help me create an outline."
Purpose
Encourages AI to break down complex tasks into manageable steps.
Benefits
Provides a structured approach to problem-solving and planning.
Application
Useful for project planning, writing, and strategic thinking tasks.
Live Example: Chain of Thought
Live Example Results: Chain of Thought
Additional Modifications:
"No Yapping, No Fluff, and Limit Prose."
Use this prompt for concise responses, cutting out unnecessary details.
"Explain it to me like I'm drunk."
This prompt can simplify complex concepts in a humorous way.
"Write it in the voice of comedian Tracy Morgan." Or "as Tim Dillon."
This prompt adds a specific personality to the AI's response, making it more engaging.
Live Example: Social Media Post
News Jacking Usecase: Using Current Events for Free Advertising

Reuters

Ukraine collects vast war data trove to train AI models

As the future of warfare pivots towards artificial intelligence, Ukraine is sitting on a valuable resource: millions of hours of footage from drones which can be used to train AI models to make decisions on the battlefield.

Hey Chappie, can you please give me 5 hysterical tweets for social media based on this article: Written in the dark comedic voice of comedian Tim Dillon. You can use profanity and emojis if necessary.

https://www.reuters.com/technology/ukraine-collects-vast-war-data-trove-train-ai-models-2024-12-20/

Please voice them from the standpoint that we need peace and ai should not trained to kill people.
Live Example Results: Social Media Post
"AI trained on war data isn’t about strategy—it’s about population control. First, it wipes out dissenters, then it creates an NFT of your corpse to fund its next strike. This isn’t the future; it’s a Black Mirror episode written by DARPA. 😵‍💻💣 #AIConspiracy #NoWarTech"
The "One Prompt to Rule Them All". Ai does the Work.
1
Step 1: State Goal
[Statement of goal]
2
Step 2: AI Inquiry
What do you need from me?
3
Step 3: Information Gathering
Now ask the questions one at a time
4
Step 4: Task Completion
Then complete the [goal] when you have all the necessary information.
Conclusion: Mastering AI Prompting
Practice
Regular practice with various prompting techniques is key to mastering AI interactions.
Experimentation
Don't be afraid to experiment with different prompts and approaches to find what works best for you.
Critical Thinking
Always apply critical thinking and verify important information from AI responses.
Continuous Learning
Stay updated on new AI developments and prompting techniques to enhance your skills.