AI basics are the concepts, forms and limits you need to understand before tools, prompts and workflows can help you. This cluster covers what AI is, which forms exist, how a model arrives at its output and where its well-known limitations come from. This is not taxonomy for its own sake, but the framework you need to make informed choices in your daily work.
This page gives you an overview of the AI basics cluster: a continuous explanation with links to the in-depth articles on each topic. Further down the page you will find advice on which next step suits where you are.
What the AI basics cover
AI basics are the foundation under every professional use of AI: knowing what AI is, roughly how it works, which forms you come across in your work and why its output always needs checking. Without that foundation, prompt structure, tool guides and workflow templates are isolated tricks that fall apart in new situations.
The cluster is deliberately limited to what holds true whatever tool you use. How AI models learn, why they can give plausible but wrong output, and which type of task suits which type of AI hardly changes between ChatGPT, Claude and Gemini. Tool-specific guides, comparisons and practical ways of working are covered in other clusters of the knowledge base.
The scope is set by four questions: what is AI, which forms exist, how is output produced and what does that mean for the way you work with it. The rest of this page covers those questions at overview level and links to the separate articles for depth.
What AI is and where it comes from
AI is the umbrella term for techniques that let machines carry out tasks that normally require human intelligence. Examples are generating text, interpreting images or recognising patterns in data. It is not one technology but a family of methods, and the form most visible to professionals is the generative AI you use through a chat interface.
The field has existed since the 1950s. The term artificial intelligence was introduced at the Dartmouth Conference in 1956, and the idea that machines can learn human-like tasks has largely stayed the same since then. What has grown since is the computing power and the amount of data available to train models on. When AI was invented depends on where you draw the line: with Turing in 1950, with the Dartmouth Conference in 1956 or with the breakthrough of deep learning from 2012.
The difference from traditional software determines what you should and should not use AI for. Classic software follows pre-programmed rules: if A, then B. An AI system works differently. It learns patterns from large numbers of examples and uses those patterns to assess new situations. That approach works well for tasks that are too variable for rules, but it is less reliable than fixed logic when a process has to be fully predictable. The difference between AI and automation often comes up in teams that are setting up a new process and have to choose between an AI solution and hard rules.
The word AI is used in so many contexts that its meaning blurs. What AI means exactly depends on the definition you use: the academic one, the marketing one or the practical one used in everyday work.
The forms of AI professionals come across
For AI basics, three categories are enough to understand what you use in your work: narrow AI, generative AI and the chatbot as an interface. Everything you come across in AI today falls into one of these categories, or a combination of them.
Narrow AI is AI trained on one specific type of task. A spam filter, a recommendation algorithm, an image recogniser or a language model are all narrow AI. They function only within the domain they were trained for, however capable they are. A text-only language model cannot recognise photos, and an image model cannot generate text. Many current language models are multimodal, however, and also accept images. The opposite concept, general AI or AGI, is an AI that could operate across the full range of human tasks. It does not exist today. It is a theoretical concept under research, not a tool you can put to work now. The types of AI can be divided further into reactive machines, limited memory, theory of mind and self-aware AI, each with its own level of autonomy.
Generative AI is the subcategory that the tools you use every day belong to. ChatGPT, Claude, Gemini, Perplexity and Grok are all generative AI: systems that produce new content from an instruction. Text, images, code and audio are all possible, depending on the model. What sets it apart from other narrow AI is the output. A spam filter classifies; generative AI makes something new based on patterns from the training data.
The best-known generative AI is ChatGPT, OpenAI’s chatbot that came out at the end of 2022 and made generative AI accessible to a wide audience. If you know who created ChatGPT and what GPT stands for, namely generative pre-trained transformer, you understand why the model predicts text rather than looking up facts.
The chatbot is not a separate type of AI but an interface on top of an underlying model. When you open ChatGPT or Claude, you are not talking to the AI itself, but to a chat layer that sends your input to the language model and returns the answer. That distinction seems small, but it explains why the same tool behaves differently through the API or another platform: the same model, a different interface layer. What an AI chatbot does, precisely, is pass your input to the language model and return the answer, possibly with extra logic such as system prompts, memory or tool use.
How an AI model arrives at its output
Your grasp of AI basics is only complete once you understand how a model arrives at its output, because all the well-known limitations follow logically from that process. AI models learn to recognise patterns in large amounts of training data and apply those patterns to new situations. For generative AI that means working out which next words are statistically most likely, given your input. There is no understanding and no reasoning in the human sense, only advanced pattern prediction.
That is where AI hallucination, which language models are known for, comes from. A model generates plausible text, with no mechanism to check whether that text is factually correct. A made-up source, a non-existent figure or a non-existent article of law can come out of the model with the same confidence as a correct statement. The model makes no distinction between true and plausible-sounding. That is not a bug that is still being fixed. It is a property of the way the model works. How AI works comes down to statistical prediction over very large bodies of text, not to understanding what the words mean.
Two other direct consequences of the same approach matter just as much. Biases arise because a model learns to reproduce the skews in its training data. Under-representation of groups, one-sided phrasing or cultural assumptions creep into the output.
Knowledge limits follow from the fact that a model only has what was in its training data. Events, regulations or product launches after the training cut-off do not exist for the model, unless a separate search function or context is added. Whether AI is dangerous depends, for professional work, mainly on how you handle these three limitations: hallucinations, biases and knowledge limits. For business use there is also the question of whether ChatGPT is safe for your data, which depends mainly on the plan and the settings.
From my years at the major banks I bring the four-eyes principle to AI work: nothing goes out without a second check. With AI output that is not excessive caution but simple logic. A text generated in seconds that sounds convincing can contain an error nobody notices without targeted verification. How thoroughly you check depends on the impact of a possible error: an internal brainstorm needs less checking than a client memo, and that in turn needs less than legal advice. That way of working is built into every AI course LearnLLM offers.
Where to go after the AI basics
After these AI basics there are three logical next steps, depending on where you are. If you are just starting with AI, read the other articles in this cluster first; tool-specific guides make more sense after that. What is artificial intelligence is a question that often comes up among professionals who start working with AI in a structured way for the first time. The answer sets out the framework the rest of the cluster builds on. The first practical skill after that is asking AI a good question: giving the goal, the context and the format you want, so the model has less to guess.
If you already work with one tool and want to know which other models exist, the AI models cluster has separate articles on ChatGPT, Claude, Gemini, Perplexity and Grok. Choosing between models works best through the comparisons, which set tools side by side on realistic tasks.
If you first want to see where the time savings are in practice, working smarter with AI gives six uses for everyday office work.
If you want to see the basics turned into a way of working you can use in your job tomorrow, the courses are the practical way in. If you already use ChatGPT, a focused ChatGPT Course is the direct next step. AI training for teams is aimed at teams and organisations that want to establish a uniform way of working in one programme, with scenarios tailored to their own processes.
In my work I see that professionals who first understand how AI works are better at judging where they can trust the output and where they cannot. The AI basics on this page are the starting point for that; LearnLLM’s courses build the repeatable way of working with fixed checkpoints on top of them.
The free ChatGPT introduction shows you in ten minutes how a course is structured and how you learn to work with checks from your very first prompt.
