generative AI
Artificial intelligence that creates new text, images, audio, video or code in response to a prompt, drawing on patterns learned from large collections of examples.
- A chatbot drafting a cover letter from three bullet points
- A product photo made from a one-line description
- A coding assistant writing a function from a comment
- A synthetic voice reading a script aloud
- Earliest AI use
- 1993
- Merriam-Webster’s first known use, for AI that produces new content.
- Current meaning
- 2022
- Sequoia names it a new category; ChatGPT follows in November.
- Also called
- GAI
- NIST’s abbreviation. Merriam-Webster lists it along with gen AI.
- Opposite
- Analytical AI
- Sequoia’s term for AI that finds patterns in existing data.
In plain words
Generative AI is artificial intelligence that makes new content, such as text, images, audio or code, instead of only sorting or predicting from data.
Many AI systems analyze data that already exists. A spam filter labels email, and a fraud check flags a payment. A 2022 essay from the investment firm Sequoia Capital called this analytical AI.1 Generative AI makes something new instead: a paragraph, an image, a voice clip, a video or a piece of code. Someone describes what they want in a prompt, and the system produces it.
OpenAI released ChatGPT in November 2022. It writes with a large language model. Image tools such as Midjourney turn a written description into a picture, and multimodal systems handle several kinds of material at once. The term is also written GenAI or gen AI. NIST, the US standards agency, abbreviates it GAI.
McKinsey estimated in 2023 that generative AI could add the equivalent of $2.6 trillion to $4.4 trillion a year across 63 uses. In one study of customer support agents, an AI assistant raised the number of issues resolved per hour by 14% on average.2 Gartner predicted in 2024 that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025. In 2025 Fortune reported an MIT study finding that 95% of generative AI pilots at companies were failing.
The term also appears in government rules. A 2023 US executive order described generative AI as models that emulate the structure of their input data to generate synthetic content. The European Parliament’s guide to the EU AI Act says systems such as ChatGPT must meet transparency requirements and EU copyright law.
It is distinct from artificial general intelligence, a term for broad, human-level ability rather than for a kind of output. Generative output can also be fluent and wrong, a failure called hallucination.
How it works
A generative model learns from a very large collection of examples, such as web pages, books, code or captioned images. Training adjusts the model’s weights, the numbers inside it, until it captures how those examples are put together. In statistical terms, it learns the distribution of the data. A discriminative model, such as a spam classifier, learns only to map an input to a label.3
To create something, the system samples from what it has learned, steered by the prompt. Large text models use the transformer design and write one token, a word or part of a word, at a time. Each time they predict what comes next. The GPT in ChatGPT stands for generative pre-trained transformer.4 OpenAI reported in 2018 that this kind of pretraining on unlabeled text gives large gains on later tasks.5
Many image generators use diffusion. In training, noise is added to pictures step by step, and the model learns to reverse the process. To make an image, it starts from random noise and removes it a little at a time.6 An earlier design, the generative adversarial network, trains a generator against a second network that tries to spot its fakes.
Because each result is sampled, the same prompt can give different outputs. A setting called temperature controls how much they vary.
Origin and first known use
Generative means able to produce or originate. In statistics and machine learning, a generative model learns how data is distributed, so it can produce new examples; a discriminative model only separates categories. Merriam-Webster dates the first known use of generative AI to 1993. Generative adversarial networks in 2014 and OpenAI’s generative pre-training in 2018 carried the adjective into deep learning. In October 2021 Gartner named generative AI one of its top strategic technology trends for 2022. In September 2022 a Sequoia Capital essay, with GPT-3 credited as a co-author, presented it as a new category. OpenAI released ChatGPT that November.
- 1993 Merriam-Webster dates the first known use of generative AI, in its current dictionary sense, to 1993. The entry lists gen AI and GAI as other forms.Merriam-Webster, “generative AI” merriam-webster.com →: source for 1993
- 2001 Ng and Jordan compare generative classifiers, which model how inputs and labels occur together, with discriminative classifiers, which model only the label given the input.Ng and Jordan, “On Discriminative vs. Generative Classifiers: A comparison of logistic regression and naive Bayes”, NIPS 2001 neurips.cc →: source for 2001
- 2014 Goodfellow and colleagues introduce generative adversarial networks: a generator captures the data distribution, and a second network estimates whether each sample is real or generated.Goodfellow et al., “Generative Adversarial Networks” arXiv 1406.2661 →: source for 2014
- 2018 OpenAI researchers report large gains from generative pre-training: a transformer language model first learns from unlabeled text, then is fine-tuned for each task.Radford, Narasimhan, Salimans and Sutskever, “Improving Language Understanding by Generative Pre-Training”, OpenAI openai.com →: source for 2018
- 2022 Sequoia Capital’s Sonya Huang and Pat Grady, crediting GPT-3 as a co-author, present generative AI as a new category of machines that make something new.Huang, Grady and GPT-3, “Generative AI: A Creative New World”, Sequoia Capital sequoiacap.com →: source for 2022
- 2022 OpenAI releases ChatGPT, a conversational model fine-tuned from its GPT-3.5 series, and makes it free to use during a research preview.OpenAI, “Introducing ChatGPT”, 30 November 2022 openai.com →: source for 2022
- 2023 US Executive Order 14110 defines generative AI as models that emulate the structure of their input data to generate synthetic content. It was revoked in 2025.Executive Order 14110, “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence”, signed 30 October 2023, Federal Register, 1 November 2023 federalregister.gov →: source for 2023
- 2024 NIST publishes its Generative AI Profile, a companion to its AI Risk Management Framework written under the 2023 order, and abbreviates generative AI as GAI.NIST AI 600-1, “Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile” nist.gov →: source for 2024
In the wild
-
“One of the most visible and powerful AI techniques coming to market is generative AI…”
Gartner, press release on top technology trends for 2022 gartner.com →: Gartner, press release on top technology trends for 2022, 18 Oct 2021Archived →: copy of Gartner, press release on top technology trends for 2022, 18 Oct 2021 -
“This new category is called “Generative AI,” meaning the machine is generating something new rather than analyzing something that already exists.”
Sonya Huang, Pat Grady and GPT-3, Sequoia Capital essay sequoiacap.com →: Sonya Huang, Pat Grady and GPT-3, Sequoia Capital essay, 19 Sep 2022Archived →: copy of Sonya Huang, Pat Grady and GPT-3, Sequoia Capital essay, 19 Sep 2022 -
“What ChatGPT and generative AI mean for science”
-
“We study the staggered introduction of a generative AI-based conversational assistant…”
Brynjolfsson et al., arXiv preprint arXiv 2304.11771 →: Brynjolfsson et al., arXiv preprint, 23 Apr 2023Archived →: copy of Brynjolfsson et al., arXiv preprint, 23 Apr 2023 -
“Generative AI could add the equivalent of $2.6 trillion to $4.4 trillion annually across the 63 use cases we analyzed…”
-
“While not all GAI is derived from foundation models, for purposes of this document, GAI generally refers to generative foundation models.”
NIST, Generative AI Profile (NIST AI 600-1) nist.gov →: NIST, Generative AI Profile (NIST AI 600-1), 26 Jul 2024Archived →: copy of NIST, Generative AI Profile (NIST AI 600-1), 26 Jul 2024 -
“At least 30% of generative AI (GenAI) projects will be abandoned after proof of concept by the end of 2025…”
-
“Generative AI, like ChatGPT, will not be classified as high-risk, but will have to comply with transparency requirements and EU copyright law…”
European Parliament, guide to the EU AI Act europarl.europa.eu →: European Parliament, guide to the EU AI Act, 19 Feb 2025Archived →: copy of European Parliament, guide to the EU AI Act, 19 Feb 2025 -
“…denying copyright registration for a comic book’s images made with generative artificial intelligence”
Circuit Judge Millett, opinion in Thaler v. Perlmutter Opinion →: Circuit Judge Millett, opinion in Thaler v. Perlmutter, 18 Mar 2025Archived →: copy of Circuit Judge Millett, opinion in Thaler v. Perlmutter, 18 Mar 2025 -
“MIT report: 95% of generative AI pilots at companies are failing”
Not to be confused with
-
Large language model
A large language model is one kind of generative AI: a model trained on text that writes by predicting the next token. Generative AI also covers image, audio and video generators that are not language models.
-
Foundation model
A foundation model is a large model trained on broad data and adapted to many tasks. Many foundation models are generative, but NIST notes that not all generative AI is derived from one.
-
Artificial general intelligence
Artificial general intelligence describes broad, human-level ability across many tasks. Generative AI describes a kind of output, new content made on request. Producing fluent text or images does not by itself show general ability.
-
Predictive AI
Predictive AI finds patterns in existing data to label, score or forecast, as spam filters and fraud checks do. Generative AI produces new content. Sequoia’s 2022 essay called the first kind analytical or traditional AI.
Questions
What is generative AI in simple terms?
Generative AI is software that creates new material, such as text, images, audio, video or code, when someone describes what they want in a prompt. It learns patterns from very large collections of examples. Many other AI systems analyze existing data instead, for example to flag spam or fraud; a 2022 Sequoia Capital essay called those analytical AI. OpenAI’s ChatGPT, released in November 2022, is one example.
What is the difference between generative AI and a large language model?
A large language model is one kind of generative AI. It is trained on text and writes one token at a time, predicting what comes next. Generative AI is the wider category. It also covers image, audio and video generators, such as diffusion models, which learn to turn random noise into a picture step by step. ChatGPT writes with a large language model.
What does GenAI stand for?
GenAI is short for generative AI, which in turn is short for generative artificial intelligence. Gartner uses GenAI in its press releases. Merriam-Webster’s entry for generative AI lists gen AI and GAI as other forms. NIST, the US standards agency, uses GAI throughout its 2024 Generative AI Profile, a guide to managing the risks of these systems.
How does generative AI work?
A generative model is trained on a very large collection of examples until it captures how they are put together. It then creates new material by sampling from what it learned, guided by a prompt. Text models write one token at a time, predicting what comes next. Diffusion models start from random noise and remove it step by step. Because output is sampled, one prompt can give different results.
When did the term generative AI become popular?
The phrase predates ChatGPT by nearly three decades: Merriam-Webster dates its first known use to 1993. In October 2021 Gartner named generative AI one of its top strategic technology trends for 2022. In September 2022 a Sequoia Capital essay presented it as a new category, and OpenAI released ChatGPT that November. A US executive order defined the term in October 2023.
Watch
- MIT 6.S191: Deep Generative ModelingA lecture from MIT’s introductory deep learning course on models that learn from data to generate new samples.
- Transformers, the tech behind LLMs | Deep Learning Chapter 5A visual walk through how a generative pre-trained transformer predicts the next piece of text.
Cite this entry
“Generative AI.” SUPERR_dictionary, Superrduperr, 24 Sept. 2026, superrduperr.com/
Superrduperr. (2026, September 24). Generative AI. In SUPERR_dictionary. https://superrduperr.com/
Superrduperr. “Generative AI.” SUPERR_dictionary. September 24, 2026. https://superrduperr.com/