AI Glossary
The complete dictionary of Artificial Intelligence
Chain of Thought
A prompting technique that encourages the model to explicitly articulate its reasoning step by step before providing the final answer. This approach improves the accuracy of responses on complex tasks requiring sequential logical reasoning.
Top-p Sampling
A nucleus sampling method that limits token selection to the most probable ones accumulating a total probability p. This technique balances coherence and creativity by avoiding very improbable word choices while maintaining some diversity.
Prompt Chaining
A technique consisting of using the output of one prompt as input for the next prompt, thus creating a logical sequence of interactions. This approach allows breaking down complex tasks into simpler, manageable steps for the model.
Instruction Following
The ability of a language model to understand and precisely execute complex instructions given in natural language. This skill is fundamental for effective interaction between humans and AI systems in practical applications.
Role Prompting
A technique consisting of assigning a specific role or persona to the model to guide its response style and perspective. This approach allows obtaining responses more consistent with the desired context or domain expertise.
System Prompt
An initial instruction that defines the model's behavior, constraints, and general guidelines for the entire conversation. This prompt establishes the reference framework in which the model will operate during subsequent interactions.
Prompt Injection
A vulnerability where malicious users manipulate prompts to bypass restrictions or modify the expected behavior of the model. This technique represents a major security challenge in applications based on language models.
In-Context Learning
The ability of a model to learn and adapt to new tasks directly from examples provided in the prompt, without modifying its weights. This property allows remarkable flexibility in the use of language models.
Prompt Decomposition
Technique involving breaking down a complex query into simpler, manageable sub-prompts for the model. This approach improves the accuracy and reliability of responses on multifaceted tasks.
Prompt Templates
Predefined prompt structures with variable placeholders, enabling consistent and efficient generation of similar instructions. These templates standardize interactions and facilitate the automation of repetitive tasks.
Prompt Validation
Process of verifying the quality, consistency, and effectiveness of a prompt before its deployment or large-scale use. This crucial step ensures reliable and predictable model performance.
Prompt Optimization
Set of techniques aimed at systematically improving prompt effectiveness to maximize model response quality. This optimization can include adjustments to formulation, structure, and parameters.
Prompt Versioning
Practice of tracking and managing different versions of a prompt to maintain a history of changes and facilitate performance comparisons. This approach is essential for continuous improvement of AI-based systems.
Token Limit Management
Strategy for managing token constraints to optimize the use of the model's context window. This technique includes automatic summarization, information compression, and relevant data selection.