Learn AI · Browse tracks
Advanced Concepts
Deeper LLM theory and systems — scaling laws, LoRA/QLoRA, alignment, MoE, context engineering, reasoning, multi-agent orchestration, guardrails.
How this relates Browse here for Hello Interview–style articles. The guided path below in the sidebar remains the project milestone sequence.
Articles in this track
- Scaling laws and computeLoss scales predictably with parameters, data, and compute — and why your tiny GPT is not ChatGPT.Open
- Fine-tuning with LoRA and QLoRAParameter-efficient fine-tuning — adapters, rank, QLoRA memory math, and when PEFT beats prompts or full FT.Open
- Alignment basics — RLHF and DPOFrom next-token pretrain to helpful/harmless assistants — preference data, reward models, RLHF, and DPO at a systems level.Open
- Mixture of Experts (MoE)Sparse models activate a few experts per token — more parameters without full dense compute.Open
- Context engineeringBudgeting tokens across system, memory, retrieval, and tools — packing context like an engineer, not a prompt poet.Open
- Reasoning and test-time computeChain-of-thought, self-consistency, and deliberate test-time compute — spend more inference to raise reliability.Open
- Multi-agent orchestrationRouter, supervisor–worker, and handoff patterns — when multiple agents beat one mega-prompt.Open
- Guardrails and safety systemsInput/output filters, policy classifiers, tool allowlists, and layered defense — product safety beyond hoping the model behaves.Open