Here are definitions of some of the commonly used AI terms:
Agentic AI: It refers to advanced artificial intelligence systems capable of autonomous decision-making, action-taking, and continuous learning from interactions
Context Window: A context window is a small piece of information that helps understand its meaning or predict outcomes. In reading, it's like looking at words before and after to get context. In technology, it's used to analyze sequences.
Token: A token is a basic unit like a word or character used for language analysis. Tokens are like tiny building blocks or pieces of information, such as words, that help computers understand and talk about various things. The number of words in a token can vary.
Hallucination: It is the generation of inaccurate outputs by a model, often due to errors, biases, or overfitting during training.
Fine-tuning: Fine-tuning an AI model means training a pre-existing model on a specific task by leveraging knowledge gained from a broader initial training. This process adapts the model to a new task with a smaller, task-specific dataset, enhancing performance.
Foundation Model: It is a broad, pre-trained AI model used as a starting point for specific tasks by fine-tuning on more focused datasets.
Generative Pre-trained Transformer (GPT): It is a type of language model in AI that uses a transformer architecture. GPT models are pre-trained on large datasets to understand and generate human-like text, making them versatile for various natural language processing tasks.
Large Language Model (LLM): It refers to a computational model designed to understand and generate human language. A language model like GPT is an example of LLMs.
Diffusion Model: It refers to a sophisticated neural network featuring latent variables that can grasp the underlying structure of an image by eliminating its blurriness or noise. Once the network is trained to understand the conceptual abstraction within an image, it gains the ability to generate variations of that image.
Reinforcement Learning (RL): It is a type of machine learning where an agent learns to make decisions by interacting with the environment. Imagine you have a little robot friend. This robot doesn't know how to do things at first, but every time it tries something, you tell it if it did a good job or not. Hence, it learns by trying different things.
Retrieval-augmented Generation (RAG): It refers to a specific architecture in natural language processing (NLP) that combines elements of both retrieval and generation models to improve performance in tasks related to understanding and generating human-like text. This could help address the issue of hallucinations.
Multi-modal: Multi-modal AI processes and integrates information from different data types such as text, image, audio, and video to gain a more holistic understanding.
State Space Models (SSMs): A new primitive for training large-scale foundation models compatible with a whole host of modalities such as text, audio, video, and images
TPU: Tensor Processing Units are Application-Specific Integrated Circuits developed by Google specifically for accelerating machine learning workloads, particularly those involving neural networks.
GPU: Graphics Processing Units are specialized processors originally designed for rendering graphics in video games and other graphics-intensive applications. GPUs are now also widely used for accelerating general-purpose computations, including machine learning and scientific simulations.
LPU: Language Processing Units are specialized hardware accelerators designed specifically for natural language processing (NLP) tasks.