If you are new to artificial intelligence, the hardest part is not the math — it is the vocabulary. The terms arrive faster than the explanations, and most introductions blur AGI, machine learning, and generative models into one undifferentiated cloud of “AI.” That cloud is where beginners get stuck: not because the concepts are hard, but because nobody drew the boundaries between them first. This guide draws those boundaries. It is a map, not a curriculum — a way to place the foundational concepts relative to one another so the next thing you read makes sense. We point to deeper explainers along the way, because no single page should pretend to teach everything from definitions through image generation through interpretability. The goal here is orientation. What Are You Actually Trying to Distinguish? The single most useful thing a beginner can do is stop treating “AI” as one thing. It is a family of distinct ideas that get the same label. Sorting them once saves months of confusion. Start with the relationship that trips up almost everyone: artificial intelligence versus machine learning. AI is the broad goal — systems that perform tasks we associate with human intelligence. Machine learning is one approach to that goal, where a system improves at a task by being shown examples rather than being explicitly programmed. Machine learning sits inside AI, not beside it. We unpack this nesting in detail in our explainer on artificial intelligence vs machine learning, which is the right next stop if the boundary still feels fuzzy. Then there is the tier of terms people use to talk about how capable a system is. Here a clear hierarchy helps. What Is the Difference Between AGI, ASI, and Generative AI? These three terms get mixed together constantly, and they answer completely different questions. Term What it describes Status today Generative AI A type of system that produces new content — text, images, audio — by learning patterns from training data Widely deployed; this is what powers tools you use today AGI (artificial general intelligence) A capability level: a system that can learn and reason across arbitrary tasks at human breadth, not just one narrow domain Hypothetical; not achieved ASI (artificial superintelligence) A capability level beyond AGI — exceeding human ability across essentially all domains Hypothetical; further still The key insight: generative AI is a kind of technology, while AGI and ASI are thresholds of generality a technology might someday cross. A generative model that writes fluent essays is still narrow — it does one family of tasks. Breadth, not fluency, is what AGI names. Our deeper treatment of what artificial general intelligence would mean and how far away it is walks through why that distinction matters and why credible researchers disagree sharply on timelines. How Close Are We to AGI Today? Honestly: nobody knows, and anyone who gives you a confident date is selling something. What we can say without overreaching is that today’s systems, however impressive, are still narrow. A model that generates photorealistic images cannot reason about a legal contract; a model that drafts code cannot drive a car. They are specialists wearing a general-purpose interface. The reason this matters for a beginner is that the gap between “looks general” and “is general” is exactly where hype lives. Fluent output creates an illusion of understanding. The systems are pattern-completion engines of remarkable sophistication, but generality — transferring competence to genuinely novel domains without retraining — remains unsolved. Treat AGI as a research horizon, not a product roadmap. That single reframe will inoculate you against most of the noise. As for “real-world examples of artificial general intelligence” — there are none yet. The phrase describes a capability that no deployed system has demonstrated. What you will find labelled “AGI” in marketing copy is almost always a broad-but-still-narrow generative system. Knowing that the examples don’t exist is itself one of the more valuable things to learn early. A Beginner’s Starting Sequence The most common question we hear is some version of “where do I even begin?” Beginners usually drown because they start with tooling — picking a framework, installing PyTorch — before they have the concepts to know what the framework is for. Reverse that order. Here is a sequence that respects how the ideas build on each other: Get the vocabulary straight. AI vs ML, supervised vs unsupervised learning, what “training” and “inference” mean. This is reading, not coding. Understand one model family end to end. Image generation is a forgiving entry point because the output is visual and the feedback loop is immediate. Our comprehensive guide to generating images with AI is a concrete place to see prompts, models, and outputs connect. Run something small yourself. A free notebook environment and a pre-trained model beats six months of theory you never apply. Hands on the keyboard changes how the concepts land. Learn to ask why a model did what it did. Interpretability — understanding the so-called black box — is what separates someone who uses AI from someone who can be trusted to deploy it. Our piece on making sense of AI’s black box is the right entry to that habit of mind. Pick a domain and go deep. Generic AI knowledge is shallow knowledge. Real competence comes from applying it to a specific problem — vision, language, or a scientific workflow. How Can a Beginner Start Learning AI Online for Free? You can get remarkably far without spending anything. Free notebook environments give you a GPU for short sessions; open-weight models and open-source libraries like PyTorch, TensorFlow, and Hugging Face’s ecosystem are free to download and run. The paid tier becomes relevant only when you need sustained compute or hosted APIs at volume — which, for a beginner, is later than you think. Start with the free path; it is not a watered-down version, it is the real thing at small scale. How Much Does AI Cost? This is the question with the least honest answers online, because the cost depends entirely on what you mean by “AI.” Learning it costs roughly nothing beyond your time, given free tools and tutorials. Using a hosted model (calling an API) costs per request — often fractions of a cent for small calls, scaling with how much text or how many images you process. Building and running your own system is where costs become real: compute for training or inference, engineering time, and ongoing operation. The trap beginners fall into is assuming a single price exists. It does not. A useful framing: cost tracks control. The more you hand off to a hosted provider, the more you pay per unit but the less infrastructure you own; the more you run yourself, the more upfront work but the lower the marginal cost at scale. For anyone learning, the honest answer is that the entry cost is near zero and rises only when your ambitions do. What Are the 5 Rules of AI? There is no official rulebook, and any list claiming to be the five rules is a simplification. But a few principles genuinely serve beginners well, and they are worth internalizing early: A model is only as good as the data it learned from. Garbage in, garbage out is not a cliché here; it is the dominant failure mode. Fluency is not understanding. A system that sounds confident can be confidently wrong. Narrow beats general, today. Specialized systems work; general intelligence does not yet exist. You should be able to question the output. If you can’t probe why a model decided something, you can’t trust it in anything that matters. Match the tool to the problem. Not every task needs a large model — or any model at all. These are habits of judgment, not laws. They will keep you from the most common beginner mistakes better than any framework tutorial. A Note on Reading Specialised Explainers Some guides you encounter will be domain-specific in ways that look opaque from outside — lab-method explainers, for instance, that decode named protocols rather than general AI concepts. Our walkthrough of making lab methods work, with Q2, R2, and Q14 explained is an example of how a how-to can be narrow and technical while still being a genuine entry point for someone in that field. The lesson generalizes: the best guide is the one scoped to the problem you actually have, not the broadest one you can find. FAQ How to start AI as a beginner? Start with concepts, not code: get the vocabulary straight first — what AI, machine learning, training, and inference mean — then run one small thing yourself in a free notebook environment. Pick a forgiving entry point like image generation where the feedback is immediate, and only then go deep on a specific domain. Tooling makes sense once you understand what it is for. How much does AI cost? There is no single price — cost depends on what you mean. Learning AI costs roughly nothing beyond your time given free tools; using a hosted model costs per request, often fractions of a cent for small calls; building and running your own system carries real compute and engineering costs. As a rule, cost tracks control: more hand-off means higher per-unit cost but less infrastructure to own. What are the 5 rules of AI? There is no official rulebook, but a few principles serve beginners well: a model is only as good as its training data; fluency is not understanding; narrow systems work while general intelligence does not yet exist; you should be able to question a model’s output; and you should match the tool to the problem rather than reaching for the biggest model by default. How close are we to AGI today? Nobody knows, and confident timelines should be treated with suspicion. Today’s systems are still narrow — impressive within one domain but unable to transfer competence to genuinely novel tasks without retraining. Generality, not fluency, is what AGI names, and it remains unsolved. Treat AGI as a research horizon, not a product roadmap. What is the difference between AGI, ASI, and generative AI? Generative AI is a type of system — one that produces new content like text or images from learned patterns — and it powers the tools in use today. AGI and ASI are capability thresholds: AGI describes human-breadth generality across arbitrary tasks, and ASI describes capability beyond human level. Generative AI exists now; AGI and ASI remain hypothetical. How can a beginner start learning AI online for free? You can get far without spending anything. Free notebook environments provide short GPU sessions, and open-weight models plus open-source libraries like PyTorch, TensorFlow, and the Hugging Face ecosystem are free to download and run. The paid tier matters only when you need sustained compute or hosted APIs at volume — later than most beginners expect. The free path is the real thing at small scale, not a watered-down version. What are some real-world examples of artificial general intelligence? There are none yet. AGI describes a capability no deployed system has demonstrated — general reasoning across arbitrary domains at human breadth. What gets labelled “AGI” in marketing is almost always a broad-but-still-narrow generative system. Recognizing that the examples don’t yet exist is itself one of the more valuable things to learn early. The terms will keep multiplying as the field moves, but the map underneath stays stable: distinguish the kind of system from its level of generality, treat fluency and understanding as separate things, and scope your learning to a problem you actually care about. Get those three right and the rest of the vocabulary stops being intimidating.
