AI.info
Generative AI
Generation as a system rather than a magic box: distributions, tokens, grounding, and the controls around the model call.
- Generative AI as a System, Not a Magic Box
- From Probability Distributions to Generated Artifacts
- Tokens, Tokenizers, and the Boundaries a Model Sees
- Embeddings and Representation Spaces
- Encoder, Decoder, and Encoder–Decoder Model Families
- Inside a Transformer Block: An Operational Anatomy
- Autoregressive Language Modeling and the Next-Token Objective
- Pretraining Data, Objectives, and the Behavior They Create
- Scaling Laws, Data Budgets, and Compute Allocation
- Benchmarks, Contamination, and Capability Claims
- Post-Training as Behavior Shaping
- Instruction Tuning and Supervised Fine-Tuning
- Preference Data, Rubrics, and Reward Models
- RLHF, DPO, and Preference Optimization Tradeoffs
- Decoding, Search, and Sampling
- Prompting as Interface Engineering
- Few-Shot Demonstrations and Prompt Datasets
- Task Decomposition and Workflow Prompting
- Reasoning, Test-Time Compute, and Verifiers
- Context Windows, Position Effects, and Effective Recall
- Context Engineering: Selection, Ordering, and Compression
- Structured Outputs and Constrained Generation
- Function Calling and Tool Contracts
- Conversation State, Memory, and Privacy Boundaries
- Retrieval-Augmented Generation as an Evidence Pipeline
- RAG Corpus Design, Chunking, and Metadata
- Sparse, Dense, and Hybrid Retrieval
- Reranking and Context Assembly
- Grounding, Citations, and Claim-Level Support
- RAG Evaluation and Debugging
- Prompt, RAG, Tool, or Fine-Tune?
- Fine-Tuning Data Curation and Release Design
- Parameter-Efficient Adaptation: LoRA, QLoRA, and Adapters
- Synthetic Data and Model-Generated Supervision
- Multimodal Foundation Models
- Generative Media, Editing, and Provenance
- Evaluating Generative AI Systems
- Hallucination, Uncertainty, Verification, and Abstention
- Security, Privacy, and Abuse-Resistant Design
- Model Selection, Routing, and Economics
- Serving LLMs: Prefill, Decode, KV Cache, and Batching
- Monitoring, Feedback, and Change Management
- Red-Teaming, Incident Readiness, and Release Governance
- Generative AI Capstone: Design and Defend a Grounded Product