JS + AI: (BYOM)
Build Your Own Model with TensorFlow.js!
WIFI: connecttech
PW: webdev24
THE SCHEDULE
9:15 - LET'S TALK AI (45m)
10:00 - Teachable Machines
10:30 - BREAK
10:45 - DEMOS
11 - FIZZBUZZ
NOON - LUNCH
1 - Visual AI
2 - Basic LLMs
2:30 - BREAK
3 - Build a KNN
4 - Transfer Learning
4:30 - END
Popular AI
Since 2012
Popular AI
TEXT GENERATION
- ChatGPT: The Leading AI model
- Gemini: Google's competetor to ChatGPT
- Llama 3: Best AI open source model from Meta
- Claude 3: GPT competitor by Anthropic
- Perplexity: ChatGPT for research
- MM1: Apple's Large Language Model
- Titan: Amazon's LLM - Comes in Express & Premiere
- Grok: Elon Musk's OpenAI competitor
- CoPilot: Microsoft's Coding Assistant
- CodeWhisperer: Amazon's CoPilot Competetor
- Tabnine: Copilot competitor company
- Jasper: AI copilot for marketing
- Mistral AI: Open source text gen models
Popular AI
Generative Images
- Midjourney: Easy AI Image Generator
- Dall-E: OpenAI Image Generator
- Adobe Firefly: AI tool from Adobe
- Stable Diffusion: Best Image Generation
- Getty iStock: "Commercially Safe" generation
- Synthesia: AI text to video
- Sora: OpenAI text to video
- Runway: AI text-to-video
- Pika: AI Video Generator
Popular AI
OTHER POPULAR AI
- ElevenLabs: AI text-to-speech & voice clone
- Suno: AI music generator
- Voice.ai: AI voice changer
- AIMastering.com: AI Audio Mastering
- Lalal.AI: Remove instrument or voices from music
- Text to Speech and Speech to Text (Too many!)
WHAT IS MACHINE LEARINING?
- WE DON'T PROGRAM
- WE CREATE TESTS
- MACHINES LEARN
THE TWO BIG BUCKETS
- Narrow AI
- General AI
EXAMPLES
Object Detection
EXAMPLES
NSFW JS
EXAMPLES
Tracking
EXAMPLES
Person Detection
EXAMPLES
Face Detection
EXAMPLES
Enjoy the Show
EXAMPLES
Facial Recognition - For Nic Cage
EXAMPLES
Face Mesh
EXAMPLES
Makeup Try on
EXAMPLES
Pose Detection
EXAMPLES
Counting Exercises
EXAMPLES
Scan Document
EXAMPLES
LANGUAGE OPTIONS
EXAMPLES
LANGUAGE OPTIONS
EXAMPLES
LANGUAGE OPTIONS
THE FUTURE
A combination of AI models working in concert to give excellent customer experience.
Anything you could do if you were given enough instruction and examples, is something an AI could do for you.
WHY NEURAL?
Neural Synapses Are
- Densely connected
- Have varying thickness in connections (weights)
- Vary synapse size (bias)
- Need a certain amount of electricity to fire, or activation voltage.
Example Problem
XOR operation
A Simple Network
b = -5
a = σ
b = 15
a = σ
b = -15
w=10
w=-10
w=10
w=-10
w=10
w=10
A Simple Network
b = -5
a = σ
b = 15
a = σ
b = -15
a = σ
w=10
w=-10
w=10
w=-10
w=10
w=10
0
1
b = -5
a = σ
0.9933
(0 * 10)
+ (1 * 10)
+ -5
= 5
σ(5) = 0.9933
A Simple Network
b = -5
a = σ
b = 15
a = σ
b = -15
a = σ
w=10
w=-10
w=10
w=-10
w=10
w=10
0
1
0.9933
(0 * -10)
+ (1 * -10)
+ 15
= 5
σ(5) = 0.9933
b = 15
a = σ
0.9933
A Simple Network
b = -5
a = σ
b = 15
a = σ
b = -15
a = σ
w=10
w=-10
w=10
w=-10
w=10
w=10
0
1
0.9933
(0.9933 * 10)
+ (0.9933 * 10)
+ -15
= 4.866
σ(4.866) ≈ 0.9924
0.9933
b = -15
a = σ
0.9924
DONE!
GPU Speed
Is this the same as code?
Matrix Math
Linear Algebra
Congratulations
You Understand Inference of a Basic Neural Network
A Simple Network
b = -5
a = σ
b = 15
a = σ
b = -15
w=10
w=-10
w=10
w=-10
w=10
w=10
1
2
3
4
5
6
7
8
9
Parameters
HOW?
MAGIC PARAMETER NUMBERS!?
HOW?
MAGIC PARAMETER NUMBERS!?
Math From Scratch?
NO!
Let's See Some AI Training
https://playground.tensorflow.org/
Let's Build a Model
https://teachablemachine.withgoogle.com/
WHY SHOULD YOU NOT USE THIS IN PRODUCTION?
Let's Code!
LUNCH?
VISUAL DATA
ROCK PAPER SCISSORS
Let's LLM
Understand LLMs deeper
COMPREHEND LLMS
BUILD A KNN
TRANSFER LEARNING
https://aisortinghat.com/
Gant Laborde
Thanks!
GantLaborde.com
