Basic AI Concepts
City Web Talk January 29, 2025

1. What is a Model?
An AI model is like a super-smart brain that can read, write, and answer questions.
- Just like a chef follows a recipe, AI follows a model to generate answers.
- Example: GPT-4 is a model that understands text and responds like a human.
💡 Analogy:
Think of AI models like different car engines:
- Some are fast and creative (GPT-4, Claude)
- Some are precise and reliable (BERT, Llama-2)
- Some are small but efficient (Gemini Nano, GPT-3.5)
2. What is a Token Limit?
Every AI model has a maximum number of tokens it can handle in one conversation.
- If a model has 4,000 tokens, that includes your question + AI’s answer.
- Longer conversations use up more tokens and may cause AI to forget earlier parts.
💡 Analogy:
Tokens are like a whiteboard space. Once it’s full, you need to erase old text to add new.
3. What is Context Length?
- AI remembers your previous messages only within its token limit.
- If you have a long conversation, earlier details might get forgotten.
- Some models, like GPT-4 Turbo, have longer memory (128k tokens), meaning they can remember more.
💡 Analogy:
Think of AI’s memory like a chalkboard—it gets wiped once it’s full!
4. What is Fine-Tuning?
Fine-tuning is like customizing the AI’s knowledge for a specific task.
- Example: If you train AI on only medical data, it becomes an expert in medicine.
- Businesses fine-tune AI for customer support, finance, legal work, etc.
💡 Analogy:
Fine-tuning is like teaching a chef only to cook Italian food—they become an expert in it!
5. What is an API in AI?
API (Application Programming Interface) allows AI to connect with apps or websites.
- Example: When you talk to ChatGPT on a website, it uses an API to send & receive responses.
- Companies use APIs to add AI to their apps, chatbots, and automation tools.
💡 Analogy:
An API is like a waiter in a restaurant—it takes your order (your request) to the kitchen (AI) and brings back your food (the response).
6. What is Latency in AI?
Latency means how fast AI responds.
- Low latency (fast) = Quick, instant replies.
- High latency (slow) = Delayed responses.
💡 Analogy:
Imagine ordering pizza—fast delivery is low latency, and slow delivery is high latency.
7. What is Embedding in AI?
Embeddings help AI “remember” things better by turning words into numbers.
- It helps AI find related words or concepts quickly.
- Example: If you ask AI “Tell me about Apple”, it knows whether you mean the fruit 🍏 or the company 🍎 based on context!
💡 Analogy:
Embeddings are like tags on Instagram—they group similar things together!
8. What is a Prompt Engineer?
A prompt engineer is someone who writes better AI prompts to get high-quality responses.
- Good prompts = Better AI answers
- Example:
- Bad Prompt: “Tell me about space.”
- Good Prompt: “Explain space travel for kids in 5 simple points.”
💡 Analogy:
A prompt engineer is like a DJ who selects the perfect song to make people dance (get the best AI response).
9. What is Reinforcement Learning (RLHF)?
AI improves based on human feedback.
- RLHF (Reinforcement Learning from Human Feedback) helps AI learn what’s right and wrong.
- Example: If AI gives a bad answer, humans correct it so it improves next time.
💡 Analogy:
It’s like teaching a child manners—if they say something rude, you correct them, and they learn.
10. What is an Open-Source AI Model?
Open-source models are free for anyone to use and improve.
- Examples: Llama 2, Falcon, Mistral, Stable Diffusion
- Companies & developers can modify them without restrictions.
💡 Analogy:
Open-source AI is like a free recipe—anyone can use, change, and improve it!