GPT-3

E18819

autoregressive language model large language model transformer-based model

GPT-3 is a large-scale autoregressive language model known for generating human-like text and performing a wide range of natural language tasks with minimal fine-tuning.

Jump to: Surface forms Statements Referenced by

Observed surface forms (10)

Surface form	Occurrences
GPT	1
GPT-3 1.3B	1
GPT-3 125M	1
GPT-3 13B	1
GPT-3 175B	1
GPT-3 2.7B	1
GPT-3 350M	1
GPT-3 6.7B	1
Generative Pre-trained Transformer 3	1
Language Models are Few-Shot Learners	1

Statements (57)

Predicate	Object
instanceOf	autoregressive language model ⓘ large language model ⓘ transformer-based model ⓘ
abbreviation	GPT-3 self-link ⓘ
announcedIn	2020 ⓘ
architecture	Transformer ⓘ
capability	few-shot learning ⓘ natural language generation ⓘ one-shot learning ⓘ question answering ⓘ summarization ⓘ text completion ⓘ translation ⓘ zero-shot learning ⓘ
commercialAccess	OpenAI Chat Completions API ⓘ surface form: OpenAI API
describedInPaper	GPT-3 self-linksurface differs ⓘ surface form: Language Models are Few-Shot Learners
developer	OpenAI ⓘ
fewShotPrompting	supported ⓘ
fineTuning	supports task-specific fine-tuning ⓘ
fullName	GPT-3 self-linksurface differs ⓘ surface form: Generative Pre-trained Transformer 3
inputType	text ⓘ
language	English ⓘ
license	proprietary ⓘ
limitation	can generate incorrect or fabricated information ⓘ may reflect biases in training data ⓘ sensitive to prompt phrasing ⓘ
modelSize	175 billion parameters ⓘ
notableVariant	GPT-3 self-linksurface differs ⓘ surface form: GPT-3 1.3B GPT-3 self-linksurface differs ⓘ surface form: GPT-3 125M GPT-3 self-linksurface differs ⓘ surface form: GPT-3 13B GPT-3 self-linksurface differs ⓘ surface form: GPT-3 175B GPT-3 self-linksurface differs ⓘ surface form: GPT-3 2.7B GPT-3 self-linksurface differs ⓘ surface form: GPT-3 350M GPT-3 self-linksurface differs ⓘ surface form: GPT-3 6.7B
numberOfParametersOfLargestVariant	175B ⓘ
outputType	text ⓘ
paperAuthors	Tom B. Brown et al. ⓘ
parameterCount	175,000,000,000 ⓘ
predecessor	GPT-2 ⓘ
releaseDate	June 2020 ⓘ
safetyMitigations	content filters via OpenAI API ⓘ
successor	GPT-3.5 ⓘ GPT-4 ⓘ
trainingCompute	hundreds of petaflop/s-days (approximate) ⓘ
trainingDataCutoff	October 2019 ⓘ
trainingDataSource	Common Crawl ⓘ WebText-like corpora ⓘ Wikipedia ⓘ books ⓘ
trainingObjective	next-token prediction ⓘ
trainingParadigm	self-supervised learning ⓘ unsupervised pre-training ⓘ
useCase	chatbots ⓘ code generation (via Codex derivatives) ⓘ content creation ⓘ virtual assistants ⓘ
zeroShotPrompting	supported ⓘ

Referenced by (18)

Full triples — surface form annotated when it differs from this entity's canonical label.

GPT-3 → abbreviation → GPT-3 self-link ⓘ

GPT-3 → describedInPaper → GPT-3 self-linksurface differs ⓘ

this entity surface form: Language Models are Few-Shot Learners

OpenAI → developed → GPT-3 ⓘ

GPT-3 → fullName → GPT-3 self-linksurface differs ⓘ

this entity surface form: Generative Pre-trained Transformer 3

DALL·E → inspiredBy → GPT-3 ⓘ