Generative AI - tech9tel/ai GitHub Wiki
๐จ What is Generative AI (GAI)?
Generative AI (GAI) is a branch of artificial intelligence that can generate new contentโsuch as text, images, music, code, and even videoโby learning from existing data. Unlike traditional AI that focuses on classification or decision-making, GAI specializes in creativity and synthesis.
๐ง Simply put: GAI is like an imaginative AI that learns patterns and creates brand-new content based on what it has seen before.
๐ Where Generative AI Fits in the AI Workflow
graph TD
A[Traditional AI] -->|Learns to Predict| B[Machine Learning]
B -->|Learns from Large Data| C[Deep Learning]
C -->|Learns to Create| D[Generative AI]
| AI Type |
Purpose |
Output Type |
| ๐ง Traditional AI |
Rule-based decisions |
Fixed responses |
| ๐ Machine Learning |
Predict and classify |
Numeric/text labels |
| ๐ง Deep Learning |
Understand patterns in complex data |
Features or predictions |
| ๐จ Generative AI |
Generate new, original content |
Text, images, audio, video |
๐ง Top Generative AI Architectures
| Architecture |
Description |
| GPT (Generative Pretrained Transformer) |
Language models that generate text |
| GAN (Generative Adversarial Network) |
Competing networks that generate realistic data |
| VAE (Variational Autoencoder) |
Probabilistic encoding and decoding |
| Diffusion Models |
Generate images by reversing noise |
| Transformer Variants |
BERT, T5, and decoder-focused models for generation |
๐ Common Algorithms in Generative AI
| Algorithm |
Purpose |
| Autoregressive Modeling |
Generate next token in a sequence |
| Noise-to-Data Generation |
Used in GANs, Diffusion models |
| Latent Variable Modeling (e.g. VAE) |
Learn hidden features |
| RLHF (Reinforcement Learning with Human Feedback) |
Align generation to human expectations |
๐ฆ Top Generative AI Models
| Model |
Domain |
Created By |
Purpose/Description |
| GPT-4 |
Text |
OpenAI |
Powers ChatGPT |
| DALLยทE 3 |
Images |
OpenAI |
Creates images from text |
| Sora |
Video |
OpenAI |
Generates videos from text prompts |
| Claude |
Text |
Anthropic |
Language generation with safety focus |
| Gemini |
Multimodal |
Google DeepMind |
Understands images, text, code |
| Stable Diffusion |
Images |
Stability AI |
Open-source text-to-image generation |
| MusicLM |
Music |
Google |
Music from textual descriptions |
๐ Real-World Use Cases
| Area |
Example Tools |
| ๐ฌ Conversational AI |
ChatGPT, Claude |
| ๐จ Creative Design |
DALLยทE, Midjourney |
| ๐งโ๐ป Code Generation |
GitHub Copilot |
| ๐ Writing & Content |
Jasper, Notion AI |
| ๐ง Audio Generation |
MusicLM, Jukebox |
| ๐ฌ Video Creation |
Sora, RunwayML |
๐
Evolution Timeline
| Year |
Milestone |
| 2014 |
GANs introduced by Ian Goodfellow ๐ |
| 2017 |
Transformer architecture released |
| 2018 |
BERT redefines text understanding |
| 2020 |
GPT-3 takes NLP to next level |
| 2022 |
DALLยทE 2 & ChatGPT go viral ๐ |
| 2023 |
Stable Diffusion, Claude, Midjourney |
| 2024 |
OpenAI launches Sora for video |
๐ฎ Future of Generative AI
- ๐ค Multimodal models combining text, vision, sound
- ๐ฝ๏ธ Realistic video and synthetic reality creation
- ๐งช Generative models for scientific discoveries
- ๐ง AGI and generalized learning systems
- โ๏ธ Focus on explainability, ethics, and safety
๐ How GAI Compares with Other AI Types
| Type |
Learns? |
Generates Content? |
Examples |
| Rule-Based AI |
โ No |
โ No |
Expert Systems |
| Machine Learning |
โ
Yes |
โ No |
Fraud Detection, Ads |
| Deep Learning |
โ
Yes |
โ ๏ธ Sometimes |
Face Recognition |
| Generative AI |
โ
Yes |
โ
Yes |
ChatGPT, DALLยทE |
๐ Related AI Fields
| Field |
Description |
Closely Related To |
| Natural Language Processing (NLP) |
Understanding/generating text |
GAI (Text generation) |
| Computer Vision |
Understanding images/video |
GAI (Image/video gen) |
| Speech Synthesis |
Generating voice/audio |
GAI (Audio/music gen) |
| Reinforcement Learning |
Decision-making with feedback |
GAI (Alignment via RLHF) |
| Multimodal AI |
Handling multiple input types |
GAI (Unified generation) |
๐งฌ Closely Related Models & Architectures
| Category |
Architectures |
Models |
| Language |
Transformers, GPT |
GPT-3, GPT-4, Claude, Gemini |
| Vision |
CNN, GAN, Diffusion |
DALLยทE, Midjourney, Stable Diff. |
| Audio |
CNN, Transformer |
MusicLM, Jukebox |
| Multimodal |
Unified Transformer |
Gemini, GPT-4V, Sora |
๐ Related Wiki Pages
๐ก Generative AI is not just about creating things โ itโs about reimagining how we interact with intelligence, art, and creativity.