Hallucination
When a model generates confident-sounding output that is factually wrong or entirely fabricated.
e.g. Citing a research paper that doesn't actually exist.AI Foundations
Core terminology: models, tokens, training, inference, hallucination.
Inference
Running an already-trained model to produce output for a given input, as opposed to training it.
e.g. Sending a prompt to an API and getting a completion back.Large Language Model (LLM)
A neural network trained on massive text corpora to predict and generate language, powering chat assistants and copilots.
e.g. GPT-4, Claude, and Llama are all LLMs.Parameters
The internal numeric weights a model learns during training; roughly indicates model scale and capacity.
e.g. A "70B" model has about 70 billion parameters.Token
The basic unit of text (a word, subword, or character piece) a model reads and generates. Context limits and API costs are measured in tokens.
e.g. "unbelievable" might split into "un", "believ", "able".Training
The process of adjusting a model's internal parameters on data so it learns patterns, typically done via pretraining followed by fine-tuning.
e.g. Pretrain on web text, then fine-tune on curated instructions.