A Glossary of Popular AI Terms: What Hallucinations, Tokens and Prompts Mean

TechCrunch has published an updated glossary of artificial intelligence terms. With the explosive growth of AI, dozens of new words and slang terms have appeared that can confuse even experienced users.

Core AI terms

When people talk about machine learning and neural networks, words like "training," "parameters" and "vectors" come up often. But a number of specific terms appear particularly frequently in the context of large language models.

• Hallucinations: when an LLM confidently generates false information and presents it as fact. A typical example: ChatGPT invents articles that do not exist or attributes to people actions they never took.
• Prompt engineering: the craft of phrasing instructions for AI. A good prompt can improve the quality of an answer many times over, while a bad one can reduce the result to nothing.
• Fine-tuning: adapting a pretrained model to specific tasks and data. Companies take GPT-4 and train it on their own corporate documents.
• Embeddings: numerical codes AI uses to encode the meaning of words and documents. They underpin search, clustering and semantic analysis.
• Token: the smallest unit of text (a word or part of a word) that a model processes. ChatGPT calculates cost by tokens, so in the API context a "token" equals money.

Architecture and methods

Over the past few years, a standard set of techniques for working with models has taken shape.

Transformer is the architecture underlying all modern LLMs such as GPT and BERT. The transformer let neural networks process data sequences in parallel, which revolutionized training speed.

RAG (retrieval-augmented generation) is a technique in which the model first searches external sources for relevant documents, then generates an answer based on them. This lets an LLM work with up-to-date information without retraining and reduces hallucinations.

Another key term is RLHF (reinforcement learning from human feedback). It is through RLHF that OpenAI made ChatGPT polite, helpful and safe. Human ratings help the model learn not only from statistical patterns in data but also from people's preferences, which changes the model's behavior itself.

Why it matters

Understanding the basic terms helps you read AI news critically. If you know about "hallucinations," you will understand why models sometimes lie. Knowing about "tokens," you can optimize your API spending. Realizing that "prompt engineering" exists should convince you not to rush: phrasing a question well deserves attention.

The TechCrunch glossary does not claim to be complete, but it provides a lingua franca, a shared language for discussing AI with a technical specialist as well as with a journalist or a manager.

What it means

Standardizing terminology is the first step toward a competent discussion of a technology. As AI develops, the language of professionals and the public needs to converge, otherwise it will feel as though they are discussing different things. When everyone speaks the same language, it is easier to find the truth and avoid panic.

If you work with AI tools or simply follow the news, the glossary saves hours of searching for definitions and examples. Instead of wandering through many sources, you get one reference you can open at any moment.