AI 与智能体

Building a tokenizer from the ground up — UTF-8 byte streams, character-level encoding, and Byte-Pair Encoding explained through working code.

Training a spam classifier, then systematically breaking it with perturbed inputs — a hands-on walkthrough of adversarial examples in NLP.

GPT-4o's multimodal capabilities — native training across text, voice, vision, and real-time latency improvements.

Text-to-image mechanics — diffusion models, text encoding, iterative denoising, and current limitations.

Express.js meets OpenAI API — building a personality-driven chatbot with system prompt engineering.