an AI that is able to do multiple things (FIXME-anthropic defs)
algorithm
a set of instructions to complete a task; a recipe
artificial intelligence
the study of intelligent agents that receive precepts from the environment and take action
a technology that has input ouptut behavior that appears as an intelligent agent
a sub field of computer science
artificial neuron
a type of function that takes a weighted sum of the inputs and passes them to an activation function
a type of mathematical, or computer, function that approximates some functions of biological neurons
deep learning
a type of machine learning that consists of creating large numbers of hidden layers of artifical neurons
discriminative model
a type of statistical model that describes the decision boundary to make a classification
feature
in machine learning, one input to the prediction algorithm, typically of many, stored together in a feature vector
generative model
a type of statistical model that describes the way that the underlying data is distributed with sufficient detail to be able to sample new data that is in the same distribution as the original data; often in
generative pretrained transformer
It is a transformer model that takes in tokens and can be sampled from to to produce novel token sequences that are similar to its training data
the output value for a particular sample or of a prediction algorithm (ie true label or predicted label)
language
a vocabulary and a grammar for how to combine the words in order to convey meaning
large language model
a probability distribution that describes the structure of sequences of tokens and enables next token prediction as well as serves as a mathematical description of tokens using many parameters
learning algorithm
an algorithm for finding patterns in data
a paramter optimization or parameter fitting algorithm
machine learning
a sub field of AI; emphasizes the use of data to learn relationships, studies the algorithms that find patterns in data
model
a simplification of some part of the world
neural network
a set of artificial neurons where the outputs of some are the inputs to the others, serving to “connect” them as a set, or network; often described as layers where connections are from one layer to the next.
neural network architecture
the general structure of how a neural network is set up, not the specific weights, but the size of the layers, number of layers, types of layers for example (FIXME)
predictive model
a mathemtical model, typically implemented in computer code, that predicts a label for an input or an outcome based on past
production
a state of a system where it is in use by users unaffiliated with the developers and generally unrestricted (except for possibly payment)
a human readable language that can be processed by computers and translated (either live through an interpreter or in batch through a compiler) to machine instructions that the physical computer can execute
research
a state of a system where is has been built and evaluated, but not deployed to general users, instead is only available to the team that built it and (potentially) other authorized users
a process to produce text that is grounded in content from specific resources in order to improve the factual correctness of the text or augment with specific data that was not available to a base model at the time of traning
commonly used to “chat with” a document or to prevent false statements
sandbox
a state of a system where is is available to test, but is not officially deployed, may have simulated data instead of real data or be isolated from external systems
statistical model
a type of mathematical model that describes its object of inquiry using probability distributions
target
the variable to be predicted in a machine learning problem setup, the dependent variable
token
the basic units used in computational processing of texts from human spoken (not programming) languages
a part of a word or a word, approximately like prefix, base, suffix parts, in general 1000 tokens is about 750 words