The technology industry has always loved its acronyms, but the artificial intelligence boom has produced a lexicon so dense that even seasoned executives find themselves smiling through meetings while quietly Googling "what is RAG" under the table. This is not a knowledge gap that will correct itself — the terminology is proliferating faster than comprehension, and the consequences extend well beyond social embarrassment.
Understanding the difference between a large language model and a foundation model, or knowing when someone is misusing "AGI" to describe a chatbot that can book dinner reservations, has become a genuine competitive advantage. Investors, journalists, hiring managers, and policymakers are all making consequential decisions based on claims they cannot fully evaluate.
The terms that actually matter
Start with the fundamentals. A large language model (LLM) is a neural network trained on vast quantities of text to predict the next word in a sequence — a deceptively simple mechanism that produces surprisingly sophisticated outputs. A foundation model is the broader category: any large model trained on diverse data that can be adapted to many downstream tasks. All LLMs are foundation models; not all foundation models are LLMs.
Fine-tuning means taking a pre-trained model and training it further on a specific dataset to improve performance on particular tasks. Retrieval-augmented generation (RAG) is a technique that grounds a model's responses in external documents, reducing hallucinations by giving the model actual sources to cite rather than relying solely on its training data.
Inference is what happens when you actually use a model — the computational cost of generating each response. Parameters are the adjustable weights inside a neural network; when someone says a model has 70 billion parameters, they are describing its scale, though more parameters do not automatically mean better performance.
The terms people misuse
Artificial general intelligence (AGI) refers to hypothetical AI systems that can perform any intellectual task a human can. Current systems, however impressive, are not AGI — they are narrow tools optimized for specific domains. When a company claims to be "on the path to AGI," treat this as marketing rather than technical disclosure.
Hallucination describes when a model generates confident-sounding but false information. It is not a bug that will be patched; it is an inherent property of how these systems work. Alignment refers to efforts to ensure AI systems behave according to human intentions — a field of genuine research that has also become a rhetorical shield for companies facing criticism.
Open source in AI often means something different than in traditional software. Many "open" models release weights but not training data, or permit use but not commercial deployment. Read the license before assuming anything.
Our take
The AI industry benefits from public confusion. Vague terminology allows modest improvements to be announced as breakthroughs and genuine limitations to be waved away as temporary. Learning the vocabulary is not about becoming a technical expert — it is about developing the critical faculty to distinguish substance from spectacle. In a landscape where billions of dollars flow toward claims that few can evaluate, fluency is a form of self-defense.
