AI Basics Lesson 0: What Are ChatGPT, Codex, and Agents, Exactly?

For anyone encountering AI for the first time, the biggest barrier is rarely the technology itself — it's the pile of terms that sound familiar yet somehow confusing when seen together: GPT, ChatGPT, Codex, Agent, AI workflows… Which ones are products? Which are models? Which ones are actually doing things on your behalf? A Bilibili course series called Learning ChatGPT from Scratch opens with this Lesson 0 precisely to help people who have zero AI background — or who get scared off by jargon the moment they open a tutorial — place these concepts in their proper positions.

Why AI Beginners Need a "Concept Map"

The instructor makes an observation worth paying attention to: although many people around us and online are genuinely benefiting from AI-driven productivity gains, the number of people who can actually use AI fluently in real life is surprisingly small. The reason isn't a lack of interest — it's that the learning curve is steeper than it looks. Jargon walls, unclear entry points, and not knowing what to learn next all shut beginners out before they even get started.

With that in mind, the course series has two explicit goals: first, to give people who have never used AI a simple, clear learning path; second, to make sure learners come away not just knowing "which buttons exist," but genuinely able to bring ideas in their heads to life. That captures the core value of AI tools — turning ideas that were previously shelved because you lacked the means to execute them into real, tangible outputs.

One point the instructor returns to repeatedly is worth emphasizing: AI is ultimately just a tool. The direction is yours to decide, and important information is yours to verify. It's a powerful assistant — but the person at the helm is always you.

Breaking Down the Core Concepts: GPT, ChatGPT, Codex, and Agents

The key to understanding the AI ecosystem is getting the hierarchy of these terms straight. The instructor uses a particularly vivid analogy to explain it.

How AI, GPT, and ChatGPT Relate to Each Other

Artificial Intelligence (AI) is the broadest category — and the term we hear most often in everyday life.

GPT is a family of models developed by OpenAI. Think of it as a "car engine" — it's the core power source, but not a finished product in itself. Version updates like GPT-5 and GPT-5.5 are essentially engine upgrades.

ChatGPT is the "finished car." It's the product most people interact with day to day, and many people's first impression of it is a simple back-and-forth chat tool. But ChatGPT's capabilities have expanded dramatically — it can now handle images, documents, web search, voice, and more, and its image model has been upgraded significantly. Its scope is actually quite broad.

What Is Codex? An AI Agent Built for Programming

Codex is a separate product aimed primarily at programming and software development — and it's essentially an AI Agent. Think of it as an executor that works on your behalf: it understands your requirements, reads your project files, creates and edits code, runs development tools, reviews its own output, and delivers a working result to you.

What Do "Agent" and "AI Workflow" Mean?

At the beginner level, an Agent can be understood as "a way of getting things done": AI works toward a specific goal by reading resources, calling tools, executing steps, checking results, making corrections, and finally delivering an output to you.

An AI workflow isn't a standalone thing — it's the process of connecting people, AI, data, tools, steps, and checkpoints into a repeatable, continuously improvable pipeline. This is the critical step that transforms scattered AI capabilities into real productivity.

A quick distinction: ChatGPT starts from conversation and covers a broader range of capabilities; Codex leans more toward end-to-end execution in programming and development. But both share one important caveat — neither can guarantee a perfect result on the first try. You need to review the final output yourself, and you can deploy review scripts to verify repeatedly until you get something usable.

A Learning Path for AI Beginners: From First Steps to Building Workflows

The instructor lays out a step-by-step learning path that starts from the very beginning:

• Getting started: Find the right entry point, understand the basic interface, and complete one genuinely useful conversation.
• Communication skills: Learn to articulate your questions clearly and develop practical prompts (the instructor specifically notes that "a one-size-fits-all prompt doesn't exist").
• Expanding capabilities: Make full use of ChatGPT's file, image, search, and voice features — and learn to verify the results AI gives you.
• Building workflows: Turn scattered Q&A sessions into a personal, real-world workflow you can apply to learning, daily life, work, content creation, and more.
• Advanced stage: Move into Codex, build knowledge of files, terminals, and projects, then get acquainted with APIs, Agents, MCP, and more complete AI workflows.

The course is roughly 70% hands-on practice and 30% theory. Theory is explained only to the extent needed to understand why something works a certain way — everything else is demonstrated through real interface walkthroughs.

Three Methodological Principles Every AI Beginner Should Remember

This introductory lesson also lays out a few methodological points that every AI learner should keep in mind.

Separate Three Types of Information

Because ChatGPT and Codex update extremely fast, the instructor commits to clearly distinguishing between three types of content: official documentation (with source and date noted), actual screen recordings (real, recorded operations), and personal experience and judgment (flagged explicitly as subjective recommendations). When course content conflicts with official documentation, defer to whatever the official source says at the time you're watching. This kind of rigor is especially valuable in a field that moves as fast as AI.

AI Is Not Always Right

The instructor repeatedly reminds viewers: using AI isn't just about knowing how to ask questions or give instructions — you must check what it produces. "Don't assume something is true just because it sounds convincing." This is the most common trap for beginners — the fluency of AI output often masks the errors inside it.

The Direction Is Yours to Decide

AI can gradually bring the ideas in your head into reality, but that still requires your own effort and judgment. No matter how powerful the tool, the person steering the ship is always you.

Conclusion

As the Lesson 0 of the entire series, this episode doesn't pile on definitions. Instead, it helps beginners build a clear "concept map" in their minds: GPT is the engine, ChatGPT is the finished car, Codex is the programming executor, Agent is a way of getting things done, and AI workflows are the process that ties everything together.

For anyone who genuinely wants to get started with AI, this "understand it, do it, verify it" approach is far more practical than diving straight into advanced theory. In the next lesson, the instructor will start from the correct official entry point and walk through — step by step — how to register and log in on both Windows and iOS, and how to complete your very first conversation. That's exactly what most people need to take that first step.