Artificial intelligence is changing the world, creating a new language to describe its progress. Spend five minutes reading about AI, and you'll encounter terms like LLMs, RAG, and RLHF that may confuse even the smartest tech professionals. This glossary aims to clarify these terms, updated regularly as the field evolves.
AGI
Artificial General Intelligence (AGI) is a nebulous term typically referring to AI that excels beyond the average human in various tasks. OpenAI CEO Sam Altman described AGI as the 'equivalent of a median human that you could hire as a co-worker.'
AI agent
An AI agent is a tool using AI technologies to perform tasks on your behalf, going beyond basic AI chatbots, such as filing expenses or booking tickets.
API endpoints
API endpoints are like 'buttons' on software that other programs can press to execute actions, allowing applications to pull data from one another.
Chain of thought
In AI, chain-of-thought reasoning involves breaking down problems into smaller steps to enhance the quality of the outcome, which may take longer but is often more accurate.
Coding agents
Coding agents are specialized AI programs that can autonomously write, test, and debug code, handling iterative tasks that typically consume developers' time.
Compute
Compute generally refers to the computational power essential for AI models to operate, often shorthand for hardware like GPUs and CPUs.
Deep learning
Deep learning is a subset of self-improving machine learning using a multi-layered artificial neural network structure for complex correlations.
Diffusion
Diffusion is a technique at the heart of many generative AI models, gradually 'destroying' data structures to learn a 'reverse diffusion' process.
Distillation
Distillation is a technique for extracting knowledge from a large AI model in a teacher-student framework.
Fine-tuning
Fine-tuning refers to further training an AI model to optimize performance for specific tasks using new, specialized data.
GAN
Generative Adversarial Networks (GANs) are a machine learning framework that drives significant developments in generative AI.
Hallucination
Hallucination refers to AI models generating incorrect information, posing a significant quality issue.
Inference
Inference is the process of running an AI model to make predictions from previously seen data.
Large language model (LLM)
LLMs are the models behind popular AI assistants like ChatGPT.
Memory cache
Memory cache is a process that enhances inference efficiency by reducing the number of calculations needed.
Neural network
A neural network is the foundational structure of deep learning, inspired by interconnected neurons in the human brain.
Open source
Open source refers to software or AI models where the underlying code is publicly available for anyone to use or modify.
Parallelization
Parallelization means performing many tasks simultaneously, crucial for modern GPU design.
RAMageddon
RAMageddon humorously describes the trend of increasing shortages of random access memory (RAM) chips.
Recursive self-improvement
Recursive self-improvement refers to AI models improving themselves without human intervention.
Reinforcement learning
Reinforcement learning is a method of training AI through trial and feedback.
Token
Tokens are the basic building blocks of human-AI communication, bridging human language and algorithmic processes.
Training
Training involves feeding data into a machine learning AI to learn patterns.
Transfer learning
Transfer learning is a technique using pre-trained AI models as starting points for new tasks.
Validation loss
Validation loss indicates how well an AI model is learning during training, with lower values being better.
Weights
Weights are crucial in AI training, determining the importance of different features in the data.