12 terms
Showing all terms starting with S
Search that understands the meaning and intent behind a query rather than matching exact keywords, powered by embeddings and vector databases.
AI technology that converts spoken audio into written text, used in transcription tools, voice assistants, and accessibility software.
A hidden instruction given to an LLM before the user's message, used to set its persona, behaviour, and constraints.
Model responses formatted as JSON, XML, or another schema, making AI outputs directly usable in downstream code without parsing.
The core operation in transformers where each token in a sequence attends to all other tokens to capture context and dependencies.
Training on unlabelled data by creating labels automatically from the data itself, for example by masking tokens or predicting the next word.
An encoder-decoder architecture that transforms one sequence (e.g., a sentence in French) into another (e.g., English), used in translation and summarisation.
A numerical measure of how closely related two embeddings are, typically computed using cosine similarity or dot product.
An activation function that converts a vector of raw scores into a probability distribution over classes, used in the output layer of classifiers.
A model where most parameter weights are zero or inactive, enabling large theoretical capacity with low practical compute cost.
Training an AI model on labelled input-output pairs so it learns to predict the correct output for new, unseen inputs.
Artificially generated data used to augment or replace real training data, useful when real data is scarce, sensitive, or expensive to collect.
