Prompt Engineering Glossary

Crafting prompts designed to test or bypass AI safety guardrails.

An autonomous AI system capable of reasoning, planning, and taking actions to achieve a goal.

The process of ensuring an AI model's behaviour matches human values and intended goals.

A set of protocols allowing different software applications to communicate with an AI model.

The structural design of a neural network model.

A hypothetical AI system that can understand, learn, and apply knowledge across a wide range of tasks at or above human level.

The neural network component that allows models to focus on relevant parts of input text.

An autonomous AI agent that chains GPT-4 calls to complete complex tasks independently.

A decoding strategy that explores multiple output sequences simultaneously to find optimal responses.

Systematic errors in an AI model's output, often stemming from skewed training data.

A common tokenisation method that merges the most frequently occurring character pairs into single tokens.

A model where users provide their own API keys for AI services, maintaining privacy and cost control.

A prompting technique that asks the model to show its reasoning step by step before giving a final answer.

A conversational interface powered by an AI model.

A family of large language models developed by Anthropic, known for large context windows.

The text generated by an AI model in response to a prompt.

The processing power (usually GPUs or TPUs) required to train or run AI models.

An approach where AI systems are trained to follow a set of principles (a 'constitution').

The maximum number of tokens an AI model can process in a single interaction.

The ability of an AI model to learn from new data streams over time without forgetting past knowledge.

Techniques used to artificially expand a training dataset by creating modified versions of existing data.

Breaking a complex task into smaller, manageable sub-prompts for better results.

Special characters used to clearly separate different sections within a prompt.

A type of generative AI model primarily used for image generation by adding and then removing noise.

Providing a model with a small hint or 'stimulus' to guide its output towards a specific outcome.

A numerical vector representation of text that captures semantic meaning.

Capabilities that appear in large language models only at sufficient scale.

One complete pass through the entire training dataset during model training.

The systematic process of testing an AI model's performance on a specific task.

Providing several examples in the prompt to guide the model's output format and style.

Training a pre-trained model on a specific dataset to specialize it for particular tasks.

A prompting technique where you provide the beginning of the desired output format (like an opening bracket) to force compliance.

A large-scale AI model trained on vast amounts of unlabelled data, adaptable to many downstream tasks.

Artificial intelligence systems capable of generating new content like text, images, or code.

Specialised hardware heavily used for training and running AI models due to its parallel processing capabilities.

Connecting AI responses to factual, verifiable information sources.

Safety mechanisms that prevent AI models from generating harmful or inappropriate content.

When an AI model generates plausible-sounding but factually incorrect information.

A rule-of-thumb or approximate method used by AI systems to make decisions faster.

A configuration variable set before training begins, like learning rate or batch size.

The ability of LLMs to learn new tasks from examples provided within the prompt, without weight updates.

The process of running a trained AI model to generate predictions or outputs from new data.

Training a model to follow natural language instructions more effectively.

The process of continuously refining a prompt based on the model's outputs until the desired result is achieved.

Attempting to bypass an AI model's safety restrictions through creative prompting.

A structured representation of facts and relationships between entities.

A multi-dimensional space where models map inputs to capture underlying similarities.

A neural network trained on massive text datasets, capable of understanding and generating human language.

The raw, unnormalised predictions generated by a neural network before they are converted into probabilities.

An efficient fine-tuning method that trains only a small number of additional parameters.

A parameter that limits the maximum length of the model's response.

Using an AI model to write or optimize a prompt for another AI model.

An architecture where only a subset of the model's neural pathways (experts) are activated for a given input, saving compute.

AI models that can process and generate multiple types of data (text, images, audio, video).

A branch of AI focused on the interaction between computers and human language.

Instructing a model on what *not* to do or include.

Providing exactly one example of the desired input/output pair in the prompt.

When a model learns its training data too well, memorising noise and failing to generalise to new data.

The internal variables (weights and biases) a model learns during training.

Methods like LoRA that fine-tune a model by updating only a tiny fraction of its parameters.

Assigning a character, profession, or identity to the AI.

The initial, highly expensive phase of training a foundation model on a massive dataset.

The input text or instructions given to an AI model to generate a desired response.

Connecting multiple prompts in sequence, where each output feeds into the next prompt.

A reusable prompt structure with placeholders for variable content.

Reducing the precision of a model's weights (e.g., from 16-bit to 4-bit) to make it run faster and use less memory.

A technique that retrieves relevant documents before generating a response, reducing hallucinations.

The intermediate logical deductions an AI makes when using Chain-of-Thought.

The process of fetching relevant information from a database to append to a prompt.

A training technique where human preferences guide the model towards more helpful responses.

Assigning a specific persona or expertise to the AI to improve response quality.

A technique that generates multiple responses and selects the most common answer for higher accuracy.

Searching for information based on meaning rather than exact keyword matches.

A mathematical function used at the end of a neural network to convert logits into probabilities that sum to 1.

A four-part prompt structure: Situation, Task, Constraints, Output — designed for consistent, high-quality AI interactions.

Training a model using high-quality dataset of input-output pairs.

In prompting, adding polite conversational filler that doesn't actually help the model.

A special prompt that sets the AI's behaviour, personality, and rules for an entire conversation.

A parameter controlling randomness in AI outputs. Lower = more focused, higher = more creative.

The basic unit of text that AI models process — roughly 4 characters or ¾ of a word.

The ability of an LLM to recognize when it needs external data and output a structured command to trigger a tool.

A decoding method that considers only the K most probable next tokens.

A decoding method that considers tokens until the cumulative probability reaches P.

The neural network architecture behind all modern LLMs, using self-attention mechanisms.

Adjusting a model's parameters or prompt strategy to achieve better performance.

A specialised database designed to store and search embedding vectors efficiently.

A multi-modal AI capable of understanding both text and images simultaneously.

The numerical parameters inside a neural network that determine how it processes input.

A prompting technique combining zero-shot with reasoning by simply appending 'Let's think step by step'.

Asking an AI to perform a task without providing any examples — relying on its pre-trained knowledge.