Concepts
24 articles in this category — generated in March 2026, not editorially reviewed. Every article carries this label in its header.
AI Safety and Alignment
RLHF, DPO, Constitutional AI — how models become "safe", jailbreaking, prompt injection, defenses, EU AI Act connection
2026-03-21 · generated, not reviewed
Attention Mechanisms
Self-Attention, Cross-Attention, Multi-Head, Flash Attention, Multi-Query, Grouped-Query, Sliding Window, Sparse Patterns
2026-03-21 · generated, not reviewed
Chain-of-Thought Prompting
CoT explained — why it works, zero-shot CoT, few-shot CoT, self-consistency, tree-of-thought, extended thinking, when CoT is overkill
2026-03-21 · generated, not reviewed
Context Windows
What context windows are, 2026 model comparison, long-context strategies, needle-in-haystack problem, effective vs advertised lengths
2026-03-21 · generated, not reviewed
Diffusion Models
Forward/Reverse Diffusion Process, Noise Schedules, DDPM, DDIM, Latent Diffusion, Classifier-Free Guidance — Theory and Practice
2026-03-21 · generated, not reviewed
Embedding Models
How Embeddings Work, Training (Contrastive Learning), MTEB Benchmark, Comparison — Sentence-Transformers, OpenAI, Nomic, BGE, E5
2026-03-21 · generated, not reviewed
Fine-Tuning LLMs
Full fine-tuning vs LoRA vs QLoRA — when to use each, dataset preparation, hyperparameters, costs, tools like Unsloth and Axolotl
2026-03-21 · generated, not reviewed
Function Calling
How Function Calling Works, LLM Tool Use Protocol, OpenAI Format, Anthropic tool_use, Open-Source (Hermes, Gorilla) — JSON Schema, Parallel Calls
2026-03-21 · generated, not reviewed
Knowledge Distillation
Teacher-Student Paradigm, Soft Labels, Temperature Scaling, Distillation for LLMs — When to Distill, Code Examples
2026-03-21 · generated, not reviewed
LoRA and Adapters
Low-Rank Adaptation math simplified, rank selection, QLoRA vs DoRA, merging adapters, practical 30-minute training examples
2026-03-21 · generated, not reviewed
Mixture of Experts (MoE)
MoE architecture — what experts are, routing mechanisms, sparse activation, models (Mixtral, DeepSeek), memory vs compute tradeoffs
2026-03-21 · generated, not reviewed
Model Merging
SLERP, TIES, DARE, Linear Merge, mergekit Tool, Practical Merge Recipes — When Merging Works
2026-03-21 · generated, not reviewed
Multimodal Models
Vision-Language Models, Audio, Video Understanding — architecture, comparison (GPT-4V, Claude 3.5 Vision, Gemini, LLaVA), practical use-cases
2026-03-21 · generated, not reviewed
Quantization
Compress large models — FP32 to INT4, formats (GGUF, GPTQ, AWQ), quality vs VRAM tradeoffs with real numbers, quantizing with llama.cpp
2026-03-21 · generated, not reviewed
RAG Explained
Retrieval Augmented Generation — retrieve relevant documents, augment prompts, generate answers. Embedding models, vector search, chunking, reranking strategies
2026-03-21 · generated, not reviewed
Reinforcement Learning
RL Basics (States, Actions, Rewards, Policies), PPO, DPO, RLHF Connection, Robotics & Games, LLM Alignment
2026-03-21 · generated, not reviewed
Reranking
Cross-Encoder vs Bi-Encoder, Reranking Pipeline, Cohere Rerank, BGE-Reranker, ColBERT — RAG Integration
2026-03-21 · generated, not reviewed
Speculative Decoding
Draft-then-Verify Approach, Acceptance Criteria, vLLM/llama.cpp Implementation, Medusa, Eagle, Lookahead Variants
2026-03-21 · generated, not reviewed
Structured Generation
Constrained Decoding, JSON Mode, Grammar-Based Generation — Outlines, Guidance, Instructor, LMQL, Regex Constraints
2026-03-21 · generated, not reviewed
Token Efficiency in AI Agents
Deep reference on token optimization. Token bloat problem, prompt caching, context window management, model selection, and concrete cost-saving strategies.
2026-03-21 · generated, not reviewed
Tokenization
BPE, WordPiece, SentencePiece explained — token counting vs words, context windows, tokenizer differences, saving tokens with compression
2026-03-21 · generated, not reviewed
Transformer Architecture
Self-Attention mechanism step-by-step, encoder/decoder patterns, positional encoding, scaling laws, and practical parameters that define modern LLMs
2026-03-21 · generated, not reviewed
Vector Databases
Embeddings and similarity, comparison (ChromaDB, Pinecone, Weaviate, Qdrant, Milvus, pgvector), self-hosted vs cloud, HNSW and IVF indexing strategies
2026-03-21 · generated, not reviewed
What is an Agent Harness?
Conceptual deep dive into agent harnesses. Definition, components, history, design patterns, and how Claude Code functions as a harness.
2026-03-21 · generated, not reviewed
Continue the learning path
The learning path puts these articles in order, and the Hub carries the building blocks we have checked in our own operations.
- Local and self-hosted
- Documented and verifiable
- From our own operations
- Made in Austria