AI Fundamentals
Core concepts behind modern artificial intelligence.
- Generative AI
Foundation models, autoregressive and diffusion generation, and the engineering patterns required to ship reliable applications on probabilistic outputs.
- Large Language Models
Learn how LLMs like GPT, Claude, and Llama process and generate human language at scale.
- Embeddings
Discover how AI converts text, images, and data into numerical vectors that capture meaning.
- Tokens
Learn why LLMs use tokens, how BPE and SentencePiece work, and how token budgets govern context, latency, and cost.
- Transformers
Transformer architecture that powers modern LLMs — self-attention, encoder/decoder stacks, KV cache implications, and why transformers replaced RNNs.
- Attention Mechanism
Understand scaled dot-product attention, Q/K/V, multi-head attention, KV caches, and the production trade-offs behind long-context LLMs.
- LoRA
Low-Rank Adaptation — train small adapters on frozen LLM weights for efficient domain specialization and multi-adapter serving.
- QLoRA
Quantized LoRA — fine-tune large models on limited GPUs by combining 4-bit (NF4) quantization with low-rank adapters.
- PEFT
Parameter-Efficient Fine-Tuning hub — LoRA, QLoRA, adapters, and when PEFT beats full fine-tuning, RAG, or prompting alone.
- RLHF
Reinforcement Learning from Human Feedback - the three-stage pipeline (SFT, reward model, PPO) that aligns LLM behavior to human preferences, plus when to use DPO instead.
- DPO
Direct Preference Optimization - a simpler, more stable alternative to RLHF that aligns LLM behavior on preference pairs without a separate reward model or PPO loop.
- GPT Models
OpenAI GPT family — GPT-6 Astra as the staged peak (gpt-6-astra), plus GPT-5.6 Sol/Terra/Luna for volume and cost routing.
- Claude Models
Anthropic Claude family — Sonnet 5 for most production, Opus 5 for high-stakes escalation, Fable 5.1 for peak GA agents, Haiku 4.5 for volume, plus MCP and workload-based routing vs GPT/Gemini.
- Gemini Models
Google Gemini — Gemini 3.8 Flash (GA), Pro-class escalation, Gemini 3.5 Transcribe preview, native multimodal, long context, and Vertex AI integration.
- Llama Models
Meta Llama open-weight models — Llama 4 Scout/Maverick for new deployments, Llama 3.x baselines, plus Muse Spark 1.3 (closed API / Muse Code) and Muse Glimmer (on-device open 30B).
- Mistral Models
Mistral AI models — commercial Large/Codestral tiers, efficient open weights, European residency options, and workload routing vs frontier closed APIs.
- DeepSeek Models
DeepSeek models — V4-Pro-0813 GA and Flash-0731 API tiers, V3 MoE open checkpoints, R1 reasoning, and peak/off-peak pricing from 2026-08-16.