Artificial Intelligence

Empirical relationships showing how AI model performance improves predictably as model size, dataset size, or compute budget increases. Scaling laws have guided the development of increasingly large f...

A machine learning paradigm where the model generates its own training labels from the input data, without human annotation. Techniques like masked language modeling and contrastive learning enable pr...

A search approach that understands the intent and contextual meaning of a query rather than relying solely on keyword matching. Semantic search uses embeddings and vector similarity to find relevant r...

An NLP technique that identifies and extracts subjective information from text, determining whether the expressed opinion is positive, negative, or neutral. Sentiment analysis is widely used in brand...

The ability of a machine to identify and process human speech into text. Automatic Speech Recognition systems use acoustic and language models to convert audio signals into written words, powering voi...

The ability of an AI model to generate output in a specific, machine-readable format such as JSON, XML, or conforming to a defined schema. Structured output enables reliable integration of LLMs into s...

Artificially generated data that mimics the statistical properties of real-world data. Synthetic data is used to train AI models when real data is scarce, expensive, sensitive, or biased, and can be g...

A parameter that controls the randomness of an AI model's output. Lower temperature values produce more deterministic and focused responses, while higher values increase creativity and diversity but m...

A custom-designed application-specific integrated circuit developed by Google for accelerating machine learning workloads. TPUs are optimized for tensor operations and are used to train and run large-...

An AI technology that converts written text into natural-sounding spoken audio. Modern TTS systems use deep learning to produce highly realistic voice output with appropriate intonation, rhythm, and e...

The basic unit of text that language models process. A token can be a word, part of a word (subword), or a single character, depending on the tokenizer. Token counts determine model input limits, outp...

The process of breaking text into smaller units called tokens, which can be words, subwords, or characters. Tokenization is the first step in natural language processing pipelines and determines how t...

A machine learning technique where a model trained on one task is reused as the starting point for a different but related task. Transfer learning significantly reduces training time and data requirem...

A deep learning architecture introduced in the 2017 paper "Attention Is All You Need" that relies on self-attention mechanisms instead of recurrence. Transformers are the foundation of modern LLMs lik...

A modeling error that occurs when a machine learning model is too simple to capture the underlying patterns in the data. An underfit model performs poorly on both training and test data, indicating in...

A generative model that combines autoencoders with probabilistic modeling to learn a continuous latent space from which new data samples can be generated. VAEs are used for image generation, data inte...

A specialized database designed to store, index, and query high-dimensional vector embeddings efficiently. Vector databases enable fast similarity searches and are essential infrastructure for RAG sys...

A learned representation of text where words with similar meanings are mapped to nearby points in a vector space. Pioneered by Word2Vec and GloVe, word embeddings capture semantic and syntactic relati...

A model's ability to perform a task it was not explicitly trained on, without any task-specific examples. Zero-shot capabilities emerge in large models that have learned general enough representations...