Sidelit Ripe Rye

How Can I Understand all the AI Jargon?

The AI field comprises many terms. At the user's functional level and oversimplifying just a bit, Machine Learning (ML) is often used to see and Generative AI (GenAI / LLM) is often used to think.


Term Meaning

Learning types
ML (Machine Learning) Algorithms that learn patterns from data.
DL (Deep Learning) Multi-layer neural networks.
SL (Supervised Learning) Learn from labelled data.
RL (Reinforcement Learning) Learn via rewards/penalties from interaction.
IL (Imitation Learning) Learn from expert demonstrations.

Generative / large models
LLM* (Large Language Model) Generative model for text.
MLLM (Multimodal Large Language Model) Handles text plus other modalities (e.g., images, audio).
RAG (Retrieval-Augmented Generation) Generation combined with retrieval/search.
MoE (Mixture of Experts) Sparsely activated expert sub-models.

Architectures / algorithms
Transformer Attention-based architecture for sequences.
CNN (Convolutional Neural Network) Widely used in vision/signal tasks.

NLP / speech / vision
NLP / NLU / NLG (Natural Language Processing / Understanding / Generation) Language analysis and generation.
MT (Machine Translation) Translate between languages.
ASR / STT (Automatic Speech Recognition / Speech-to-Text) Transcribe speech to text.
TTS (Text-to-Speech) Synthesis of speech from text.
CV (Computer Vision) Understand images/video.

Big-picture
AGI / ASI (Artificial General / Super-intelligence) Hypothesised broader-than-human / superhuman AI.


* The acronyms are stacked in use; so a LLM such as ChatGPT trains using a combination of self-supervised pre-training; fine-tuning - self-supervised and on human curated input/response labelling; and lastly there is post-training alignment, partly performed by humans, to make output safer and more helpful.


NB Extensive as the above is, they're actually only some of the many branches of the AI tree.