Alternative Colored Links to the AI Training Curriculum

[Chip (OP)]

♟️ 0. The First AI — Samuel’s Checkers Program
https://forum.drugs-and-users.org/index.php?topic=7338

What this section teaches:
The birth of machine learning. Samuel’s system was the first program that improved itself through experience rather than rules.
It introduced:
• Evaluation functions
• Self-play
• Learning from outcomes
• Early reinforcement-learning-like behaviour
This is the conceptual ancestor of modern RL, not LLM pretraining.


------------------------------------------------------------

🧠 1. Pretraining Modern AI / LLMs
https://forum.drugs-and-users.org/index.php?topic=7365

What this section teaches:
How modern AI evolved from early ML ideas. The lineage:
Samuel → TD learning → deep learning → transformers → RLHF.
Clarifies that LLM pretraining is NOT reinforcement learning. Pretraining extracts patterns; RLHF aligns behaviour.


------------------------------------------------------------

🏗️ 2. High‑Level AI Architecture & Data Flow
https://forum.drugs-and-users.org/index.php?topic=7347

What this section teaches:
A systems-engineering overview of the entire AI stack:
• Tokenisation
• Embeddings
• Transformer layers
• Output decoding
This gives the “big picture” before diving into internals.


------------------------------------------------------------

🔧 3. How Neural Networks Work
https://forum.drugs-and-users.org/index.php?topic=7343

What this section teaches:
The foundation of all deep learning:
• Neurons and layers
• Activations
• Backpropagation
• Gradient descent
Everything modern builds on this.


------------------------------------------------------------

🔤 4. The Tokenisation Pipeline
https://forum.drugs-and-users.org/index.php?topic=7350

What this section teaches:
Why text must become numbers:
• Token IDs
• Byte Pair Encoding (BPE)
• Embedding lookup
• Why models operate on vectors, not characters
This bridges raw text → model-usable representation.


------------------------------------------------------------

⚡ 5. Transformers
https://forum.drugs-and-users.org/index.php?topic=7344

What this section teaches:
The architecture that changed everything:
• Attention blocks
• Parallel processing
• Residual connections
• LayerNorm
• Feed-forward networks
Transformers replaced RNNs/LSTMs because they scale.


------------------------------------------------------------

💬 6. LLMs Explained
https://forum.drugs-and-users.org/index.php?topic=7342

What this section teaches:
How transformers become chatbots:
• Pretraining on massive corpora
• Next-token prediction
• Why LLMs “sound intelligent”
• How chat interfaces wrap the model


------------------------------------------------------------

📚 7. RAG — Retrieval Augmented Generation
https://forum.drugs-and-users.org/index.php?topic=7348

What this section teaches:
How LLMs integrate external knowledge:
• Vector databases
• Embedding search
• Document retrieval
• Reducing hallucinations
RAG is the modern solution to static model knowledge.


------------------------------------------------------------

🧭 8. Embeddings & Vector Spaces
https://forum.drugs-and-users.org/index.php?topic=7351

What this section teaches:
The heart of modern AI:
• High-dimensional meaning spaces
• Semantic clustering
• Cosine similarity
• Latent space structure
• Why “cat” and “dog” cluster
• Why hallucinations happen
This is the bridge: Token IDs → Meaning Space.


------------------------------------------------------------

🎯 9. Attention Mechanisms & Self‑Attention
https://forum.drugs-and-users.org/index.php?topic=7352

What this section teaches:
The real magic inside transformers:
• Query / Key / Value vectors
• Attention weights
• Token relationships
• Context windows
• Why attention replaced recurrence
Without this, transformers look like magic.


------------------------------------------------------------

🧪 10. Training vs Inference
https://forum.drugs-and-users.org/index.php?topic=7353

What this section teaches:
One of the most misunderstood topics:
• Pretraining
• Gradient descent
• Static weights
• Fine-tuning
• RLHF
• Why models do NOT learn during conversation
This clears up enormous confusion.


------------------------------------------------------------

🧩 11. Context Windows & Memory
https://forum.drugs-and-users.org/index.php?topic=7354

What this section teaches:
Why chatbots forget:
• Token limits
• Sliding windows
• Context truncation
• Persistent memory systems
• RAG vs memory
Explains real-world chatbot behaviour.


------------------------------------------------------------

⚠️ 12. Hallucinations & Failure Modes
https://forum.drugs-and-users.org/index.php?topic=7355

What this section teaches:
Why AIs make things up:
• Probabilistic generation
• Confidence ≠ correctness
• Distribution gaps
• Mode collapse
• Confabulation
• Prompt injection
• Context poisoning
Most people misunderstand hallucinations entirely.


------------------------------------------------------------

🎥 13. Multi‑Modal AI
https://forum.drugs-and-users.org/index.php?topic=7356

What this section teaches:
Modern AI is no longer text-only:
• Vision transformers
• Image tokenisation
• Audio embeddings
• Cross-modal latent spaces
• Diffusion models
Connects LLMs to image/audio/video systems.


------------------------------------------------------------

🤖 14. Agents & Tool Use
https://forum.drugs-and-users.org/index.php?topic=7357

What this section teaches:
The frontier of modern AI:
• Tool calling
• External APIs
• Planning loops
• Chain-of-thought orchestration
• Autonomous agents
• Memory stores
• Execution environments
This is where AI is heading now.


------------------------------------------------------------

📈 15. Scaling Laws
https://forum.drugs-and-users.org/index.php?topic=7358

What this section teaches:
Why bigger models suddenly worked:
• Parameter scaling
• Data scaling
• Compute scaling
• Emergent behaviour
• Why GPT‑3 changed everything
Explains the “sudden” leap in AI capability.


------------------------------------------------------------

🔀16. Mixture of Experts (MoE)
https://forum.drugs-and-users.org/index.php?topic=7367

What this section teaches:
How modern frontier models scale without scaling compute proportionally:
• What an "expert" actually is (a sub-network)
• The router / gating mechanism
• Sparse activation — only some experts fire per token
• Why MoE allows enormous parameter counts on modest compute
• MoE vs dense models
• Mixtral, GPT-4 (rumoured), Gemini — real-world examples
• The trade-offs: memory vs compute
Explains why parameter counts stopped being a simple proxy for capability or cost.


------------------------------------------------------------

💾 17. Quantisation & Model Compression
https://forum.drugs-and-users.org/index.php?topic=7360

What this section teaches:
How huge models run on ordinary hardware:
• FP32 → FP16 → INT8 → INT4
• GGUF / GGML formats
• Pruning
• Distillation
• Running 7B models on laptops
The practical consequence of scaling laws.


------------------------------------------------------------

🖼️ 18. Diffusion Models
https://forum.drugs-and-users.org/index.php?topic=7359

What this section teaches:
How image generation works:
• Noise schedules
• Denoising
• Latent diffusion
• Classifier guidance
• Why Stable Diffusion works
A completely different architecture family from transformers.


------------------------------------------------------------

🔮 19. The Future of AI
https://forum.drugs-and-users.org/index.php?topic=7361

What this section teaches:
A forward-looking synthesis:
• Frontier model trends
• Agentic systems
• Multi-modal unification
• Long-context architectures
• Self-improving systems
A realistic view of what’s coming next.