The A-Z of AI-ready content
Learning about how to get your content ready for AI? Don’t know where to start?
We’ve got you.
We’ve selected these 26 concepts, one for each letter of the alphabet, to help jump-start your AI learning journey.
A is for Autonomous Agents
Autonomous agents are software components that can execute a series of tasks within a complex environment, so making decisions along the user journey, toward reaching the wanted outcome.
B is for Bias
Bias refers to problematic training data that can lead to generated content that is prejudiced or slanted against particular populations or ideologies. Correcting the bias involves fixing the content used as training data.
C is for Chunking
Chunking is the division of content into smaller, meaningful parts to facilitate comprehension and processing. This approach preserves context and significance. Semantic chunking enhances AI understanding of information.
D is for Deep Learning
Deep learning is a way that AI mimics how human brains process patterns and recognize objects, particularly when they are complex. Deep learning uses multiple layers of neural networks; each layer helps the AI better understand the patterns. Deep learning is used in various ways, including analysing content and processing speech.
E is for Ethics
Ethics encompasses moral principles and practices to guide the development and responsible use of to ensure AI use is fair, transparent, accountable, and safe. The intent is to align the technology with human values, benefits to society, and mitigate harm. Ethics includes areas such as bias, privacy, fairness, accountability, and societal impact.
F is for Fine-tuning
Fine-tuning an LLM refers to the algorithmic adjustment of an LLM’s parameters in relation to a task-related dataset. The purpose is to enhance the LLM’s performance on a particular task or in a particular domain.
G is for Guardrails
Guardrails are a way to constrain an LLM to ensure reliable and safe answers. Guardrails comprise guidelines and measures to ensure high-quality, responsible, accurate content generation.
H is for Hallucination
Hallucinations happen when an LLM uses training data and interprets it to generate incorrect content, despite the content being grammatically correct and seeming credible.
I is for Information Architecture
Information architecture is the organization, structure, and labelling of content in a way that makes sense to users, and ensures more reliable delivery automation.
J is for Joint Learning
Joint learning is a training approach where an AI model is trained to simultaneously perform multiple tasks, which allows AI to be used in more sophisticated ways, such as performing multiple steps of a process.
K is for Knowledge Graph
A knowledge graph is a structured representation of knowledge that organizes data in terms of entities and their interrelationships. This promotes better semantic understanding of content and complex retrieval. A knowledge graph uses an ontology to describe knowledge structures, while the knowledge graph is a collection of instances and relationships built on the ontology.
L is for LLMOps
LLMOps combines principles from DevOps and MLOps, applied to ongoing development and maintenance of LLMs.
M is for Metadata
Metadata is data that describes data – and content – used by machines to understand more context. Metadata is essential for AI models to generate new content with more accuracy.
N is for NLP
Natural Language Programming is a branch of AI that gives computers ability to understand, interpret, and generate human language. NLP lets people interact with computers using natural sentences and more readily retrieve relevant information.
O is for Ontology
An ontology codifies the representation of knowledge that defines concepts and the relationships between each other. It provides a structured framework for organising information, and can transform unstructured into structured data or content.
P is for Prompts
Prompting is the process of creating and fine-tuning queries that are inputs to LLMs. The purpose is to optimise the generation of more accurate answers from the LLM.
Q is for Q-learning
Q-learning refers to a reinforcement learning algorithm. It computes all potential actions and chooses the optimal action to achieve the maximum reward possible.
R is for RAG
Retrieval Augmented Generation is how an LLM constrains the content it searches, so that generated answers are more accurate and relevant to the query.
S is for Structure
Structure is comprised of several techniques to organise information. Editorial structure helps with human comprehension, while AI uses semantic structures to improve how it processes information, to generate better results.
T is for Temperature
Temperature is a hyperparameter that controls what degree of random responses an LLM is allowed to generate. Temperature is often expressed as a creative measure, where a higher temperature allows for more randomness in the output generated, while a lower temperature constrains that randomness.
U is for Uncertainty Estimation
Uncertainty estimation expresses how uncertain model predictions can be. Rather than do extensive tuning for a single temperature, an AI model’s range of uncertainty can be estimated by applying multiple sampling passes using different temperatures.
V is for Vector Database
A vector database stores information in a data format as high-dimensional numerical vectors. These vectors represent the features and context of different types of content, such as text, images, and audio. The vectors are indexed to enable searches for semantic similarity.
W is for Weight
Weights are parameters used by neural networks during its training process, where each neuron is given a weight, to learn how information should flow through it.
X is for XML
XML is an extensible markup language that is rich in structure and metadata, which AI uses to understand what the metadata describes. The structured content is used by AI to interact with systems in a standardised way.
Y is for YOLO
YOLO (You Only Look Once) is an AI algorithm that detects and locates multiple objects in visual content, such as images or videos, in real time.
Z is for Zero-Shot Prompting
Zero-shot prompting, also called direct prompting, is when an LLM receives a prompt with a query outside of its training data and has no examples to use as a model; the AI looks to a larger body of work to generate a response.
