Slides: Week 1 What AI Is · Mehmet Sait Gümüş

AI and LLMs from zero

What AI Is

Week 1 of 12, AI and LLMs from zero
By the end you can explain what AI is and how it is built.

Outcomes

What You Will Be Able To Do

Explain AI in plain words
Describe the main eras of AI
Compare rules with learning
Name today's kinds of AI systems
Test a chat assistant on a real task

Vocabulary

Three Words To Keep Apart

Program: a list of instructions
Algorithm: a clear procedure for a problem
Model: stored numbers used to predict
AI systems often combine a model and a program

Definition

AI Is A Family, Not One Thing

Tasks that normally need human judgment
Spam filters, maps, dictation, photo search
Chat assistants are one part of the field
This course narrows to language models

History

1950 To 1980: Symbols And Rules

1950: Turing asks whether machines can think
1956: Dartmouth workshop names the field
Programs reason with hand-written logic
Expert systems encode human knowledge

History

Cycles Of Hope And Pause

Rules were costly and broke on new cases
Funding rose and fell in cycles
Quiet progress in statistics and learning
Machine learning replaces hand-written rules

History

Milestones That Changed Direction

1997: Deep Blue beats Kasparov at chess
2012: AlexNet wins an image recognition contest
2017: the transformer paper from Google
Weeks 3 to 5 explain these ideas

Year	Event	Why it matters
1950	Turing publishes Computing Machinery and Intelligence	Poses the question of machine thinking
1956	Dartmouth workshop	The term artificial intelligence is adopted
1966	ELIZA	First famous chatbot
1997	Deep Blue beats Kasparov	Machine wins at chess
2012	AlexNet	Deep learning takes over vision
2017	Transformer paper	Basis of modern language models

Core idea

Two Ways To Build A Program

Rules: a person writes the logic
Learning: the computer finds logic from examples
Both are useful tools
Engineers choose by task and risk

	Rules	Learning
Who writes the logic	A person	The data
Strength	Predictable, easy to explain	Handles messy, fuzzy tasks
Weakness	Breaks on cases nobody listed	Needs data, can inherit its flaws
Example	If subject contains 'prize', mark spam	Learn spam from labeled emails

Rules

Example: Spam Filter By Rules

Easy to read and easy to fix
Predictable output
Spammers change wording quickly
Too many patterns to list by hand

if "free money" in subject:
    label = "spam"
else:
    label = "ok"

Learning

Example: Spam Filter By Learning

Collect emails already labeled by people
Adjust the model until guesses match labels
Apply the result to new emails
Finds patterns a person might miss

Choosing

Trade-Offs

Rules: human effort, fully predictable
Learning: needs data, sometimes wrong
Combine: model proposes, rules check
Pick by cost of a mistake

Scope

Narrow AI And AGI

Today's systems are narrow
Language models handle many text tasks
AGI means nearly any intellectual task
No such system exists today

Map

The Landscape Today

Language models: read and write text
Image generators: text to picture
Speech systems: audio to text and back
Vision and recommendation systems
Multimodal models: text plus images

Access

Who Builds It, How You Reach It

Few companies train the largest models
Open-weight models can be downloaded
Reach models by chat, API, or local run
Week 10 covers building with an API

Habit

Treat Output As A Draft

Fluent text can still be wrong
Check the facts that matter
Test claims on small examples
Keep a notebook of prompts and results

Summary

Recap

AI is a family of methods
History moved from rules to learning
Rules are predictable, learning is flexible
Current systems are narrow but broad in text

Next week

Next Week: Data And Machine Learning Basics

Features, labels, and training
Overfitting and how to spot it
Measuring a model with evaluation sets

Wrap up

Questions And Exercise

Exercise: rules versus a chat assistant
Label 10 comments both ways and compare
Read the Turing paper introduction
Bring one question to week 2