The Hitchhiker's Guide to Agentic AI — full-stack practitioner's reference
New arXiv book covers autonomous AI systems end-to-end: from transformer architecture and GPU systems through training, alignment, and reasoning, to production deployment.
Central premise: great agentic systems demand literacy across every layer of the pipeline, not depth in one. Treats LLM substrate as foundation, then builds alignment (RLHF, PPO, DPO, GRPO, reward modeling) and reasoning.