AI-Assisted Development

About AI

Workshop — Part 2: What is AI and how does it work?

2 / 4 — About AI
AI-Assisted Development

What is AI?

Artificial Intelligence simulates human intelligence in machines.

Enables computers to:

  • Understand natural language
  • Recognize patterns
  • Make decisions
  • Solve problems

Many sub-fields — but this workshop focuses on one: Generative AI

2 / 4 — About AI
AI-Assisted Development

Types of AI

Machine Learning
Learns from data to make predictions

Natural Language Processing
Understands and processes human language

Computer Vision
Interprets visual information

Reinforcement Learning
Learns by trial and reward

Deep Learning
Neural networks with many layers

Generative AI
Creates new content: text, images, code

2 / 4 — About AI
AI-Assisted Development

This workshop focuses on

Generative AI

The most relevant type for coding.

2 / 4 — About AI
AI-Assisted Development

What is Generative AI?

Models that generate new content based on patterns learned from existing data.

They can produce text, images, code, and more.

Trained on massive datasets, they understand context to generate coherent, relevant responses.

2 / 4 — About AI
AI-Assisted Development

Notable Models

Language & Code

  • GPT (OpenAI) — ChatGPT, GitHub Copilot
  • Claude (Anthropic) — Claude Code, Claude.ai
  • Gemini (Google) — Google AI features
  • Llama (Meta) — Open-source, self-hostable

Images

  • Stable Diffusion — Open-source
  • DALL-E — OpenAI image model

Code-specialized

  • Codestral (Mistral)
  • DeepSeek Coder
2 / 4 — About AI
AI-Assisted Development

How Do Transformers Work?

The architecture behind all modern AI language models.

  • Process sequential data — text, code
  • Capture long-range dependencies — understand context far back in a document
  • Use self-attention to weigh importance of each token relative to all others
"The bank can guarantee deposits will eventually cover
 future tuition costs because it invests in [?]"

Self-attention understands "bank" = financial institution,
not a river bank — purely from context.
2 / 4 — About AI
AI-Assisted Development

Self-Attention in Plain English

For each token (word/character), the model asks:

“Which other tokens in this sequence are most relevant to understanding this one?”

This happens simultaneously for all tokens — that’s what makes transformers fast and powerful.

Input tokens:

The  cat  sat  on  the  mat
 ↕    ↕    ↕   ↕   ↕    ↕
[attention weights between every pair]

Each word “looks at” every other word to build context.

2 / 4 — About AI
AI-Assisted Development

From Training to Code Generation

  • Pre-training Learns from billions of lines of code and text
  • Fine-tuning Specialized on coding tasks and human feedback (RLHF)
  • Context window Your prompt + conversation history fed in at inference
  • Inference Given your prompt, predicts the most likely next tokens
  • Output Streams tokens one at a time until a stopping condition
2 / 4 — About AI
AI-Assisted Development

Why This Matters for Developers

Because the model predicts plausible output — not correct output:

  • It sounds confident even when it’s wrong
  • It hallucinates APIs and options that don’t exist
  • It defaults to common patterns — which may not fit your project
  • It has a training cutoff — it doesn’t know about recent releases

Treat AI output as a draft, not a finished answer.

2 / 4 — About AI
AI-Assisted Development
The model predicts plausible. Not correct.

You provide the judgment.

2 / 4 — About AI
AI-Assisted Development

Summary:

  • Generative AI generates new content from learned patterns
  • Transformers use self-attention to understand context across a sequence
  • The model predicts plausible output — not guaranteed correctness
  • Training cutoffs and hallucinations are fundamental limitations
  • Understanding this helps you use AI tools more effectively
2 / 4 — About AI
AI-Assisted Development

Further Reading

Interactive

Visual Explanation of Transformers

Explore attention weights in your browser

Deep dives

Attention is All You Need — original paper

The Illustrated Transformer — visual walkthrough