ONLINEAGENT_OPS 2026.Q3 HOME HUBSMODELSBUILDINGCOMPANIESFILMMAKINGSECURITYLIVING WITH ITSKILLSALL HUBS, ONE TERMINAL → ARTICLES CRAFT RECORD BLOG TOOLS SKILLS SEARCH /
HOME›AI MEANINGS›WHAT ARE EMERGENT ABILITIES IN AI?
AI MEANINGS · MEANINGBEGINNER

What are emergent abilities in AI? Skills that seem to appear with size

Emergent abilities are skills bigger AI models seem to gain suddenly. Both sides of the research debate, and how to read “AI suddenly learned” claims.

READ6 min
WORDS1,178
SECTIONS7
TYPEMEANING
CHECKED1 OCT 26

Emergent abilities means skills that bigger AI models seem to have but smaller ones lack, as if they switched on once the model got large enough. Whether they really appear suddenly, or only look that way because of how they are measured, is still argued about by researchers.

In everyday English, “emergent” just means coming into being. In AI it names this specific, disputed claim.

◈ IN PLAIN TERMS

Picture a child learning to read. For months they cannot read a full sentence, then one day they can. Did a skill appear overnight? Or were they getting better at each word all along, and the “whole sentence or nothing” test simply hid the progress? That is the emergent abilities debate in one picture.

In a sentence: “The article called the new skill an emergent ability, but the researchers it quoted said it only looked sudden because of how it was scored.”

The claim: new skills appear with size

The plain point: some researchers saw abilities that small models lacked and big models had.

The idea was set out in a 2022 paper. It called an ability emergent if it shows up in larger models but not in smaller ones, and drew a bold conclusion: you cannot predict such abilities just by looking at how smaller models perform.1

The authors also said this pointed to more to come.2 “Scaling” means making models bigger and training them on more data. Size is usually counted in parameters.

The reply: it may be the measuring

The plain point: other researchers argued the jumps come from the scoring method, not the model.

A 2023 paper took the other side.3

Their reasoning: an all-or-nothing score, like “exactly right or wrong”, can hide steady improvement. Score the same answers in a way that gives partial credit, and the sudden jump can turn into a smooth slope.4

SideWhat they sayWhat follows
Emergence is realSays: some skills are absent in small models and present in large onesSo: you cannot fully predict a bigger model from smaller ones
Emergence is a mirageSays: the jumps come from how answers are scoredSo: with fairer scoring, progress looks gradual and predictable

This site does not pick a winner. The question is not settled.

SAME ANSWERS, TWO WAYS OF SCORING · THE MIRAGE ARGUMENT
The reply’s point in one picture, like the child learning to read. It is one side of an open debate.
ALL-OR-NOTHING SCORE
Note: Exactly right, or nothing.
No: Counts getting closer
Note: Progress can look like a sudden jump.
PARTIAL-CREDIT SCORE
Note: Credit for each part that is right.
Yes: Counts getting closer
Note: The same progress can look like a smooth slope.
Reasoning, 1 Oct 2026 — summarises this page’s section on the reply (Schaeffer, Miranda & Koyejo) and its reading comparison; it does not settle the debate.

Why it matters to you

The plain point: treat “the AI suddenly learned to…” headlines with care.

  • Hype check. A claim that a model “suddenly” gained a skill may depend on how the skill was scored. Ask how it was measured.
  • Both directions. The dispute also cuts against certainty that AI is “just” predicting text. Does AI understand? makes that point with the same two papers.
  • Your own tasks. Whatever the theory, test a model on the work you need. See how to test a prompt change.
TAKEAWAY

“Emergent” is a claim, not a fact. When you see it, ask: emerged according to which test?

Where the word comes from

The AI term was set out in 2022, but it borrows an older idea from science. The paper that defined it, by researchers from Google Research, Stanford University, UNC Chapel Hill and DeepMind, notes that emergence “has been long discussed in domains such as physics, biology, and computer science”. It took its general definition from a 1972 essay, “More Is Different”, by the physicist Philip Anderson: “Emergence is when quantitative changes in a system result in qualitative changes in behavior.”5 In plain words: a change in amount produces a change in kind.

Often confused with

  • Scaling laws: the steady, predictable gains from bigger models and more data. Emergence is the claim that some gains are not steady. See the glossary entry.
  • AGI: a label with several meanings, one being AI as capable as people at nearly all thinking tasks. Emergent abilities are narrower: single skills. See AGI meaning.
  • Reasoning models: models trained to write out their working before answering. That is about how a model is built and trained, a separate question from whether skills appear with size. See reasoning model meaning.

Questions people ask

What is an example of an emergent ability?

Simple arithmetic is the best-known one. The 2022 paper shows a test of “3-digit addition and subtraction, as well as 2-digit multiplication” on which the GPT-3 models scored close to zero until the 13-billion-parameter size, when their scores jumped well above chance.5

Are emergent abilities real?

Researchers disagree. A 2023 paper argued that “emergent abilities appear due to the researcher’s choice of metric rather than due to fundamental changes in model behavior with scale.”3 The two sides are set out above; the question is still open.

What is the difference between emergent abilities and scaling laws?

Scaling laws describe gains you can predict; emergent abilities are claimed gains you cannot. The 2022 paper says the effect of size “can often be methodologically predicted via scaling laws”, but that emergent abilities “would not have been directly predicted by extrapolating a scaling law”.5

Learn more on this site

QUICK CHECK · 3 QUESTIONS

Pick an answer to see if you got it. Nothing is recorded.

1. What are emergent abilities?

  • Skills that bigger AI models seem to have but smaller ones lack
  • New features a company adds in an update
  • Skills every AI model has from the start
Show the answer

Answer: Skills that bigger AI models seem to have but smaller ones lack. Emergent abilities are skills that seem to switch on once a model gets large enough, though researchers still argue about them.

2. What did the 2023 reply paper argue?

  • That bigger models are always worse
  • That small models have more skills than big ones
  • That the sudden jumps come from how the answers are scored
Show the answer

Answer: That the sudden jumps come from how the answers are scored. The reply argued that an all-or-nothing score can hide steady progress, so the jumps may come from the scoring, not the model.

3. When you read that an AI ‘suddenly’ learned a skill, what does the page suggest asking?

  • Which company built the model
  • How the skill was measured
  • How long the model took to answer
Show the answer

Answer: How the skill was measured. The takeaway: ‘emergent’ is a claim, not a fact, so ask which test it emerged on and how it was scored.

Related terms: Computer vision · Constitutional AI · Distillation · Fine-tuning · Foundation model · Frontier model

SOURCES
  1. 1Wei et al., Emergent Abilities of Large Language Models, arXiv, 15 Jun 2022 (published in Transactions on Machine Learning Research), read at source 1 Oct 2026“We consider an ability to be emergent if it is not present in smaller models but is present in larger models. Thus, emergent abilities cannot be predicted simply by extrapolating the performance of smaller models.” ↑
  2. 2Wei et al., Emergent Abilities of Large Language Models, abstract read at source 1 Oct 2026“The existence of such emergence implies that additional scaling could further expand the range of capabilities of language models.” ↑
  3. 3Schaeffer, Miranda & Koyejo, Are Emergent Abilities of Large Language Models a Mirage?, arXiv, 28 Apr 2023, read at source 1 Oct 2026“emergent abilities appear due to the researcher’s choice of metric rather than due to fundamental changes in model behavior with scale.” ↑a“emergent abilities appear due to the researcher’s choice of metric rather than due to fundamental changes in model behavior with scale.” ↑b
  4. 4Schaeffer, Miranda & Koyejo, Are Emergent Abilities of Large Language Models a Mirage?, abstract read at source 1 Oct 2026“nonlinear or discontinuous metrics produce apparent emergent abilities, whereas linear or continuous metrics produce smooth, continuous predictable changes.” ↑
  5. 5Wei et al., Emergent Abilities of Large Language Models, arXiv:2206.07682, 15 Jun 2022 (published in Transactions on Machine Learning Research), full text read at source 12 Oct 2026“Emergence as an idea has been long discussed in domains such as physics, biology, and computer science”; the definition is “rooted in a 1972 essay called ‘More Is Different’ by Nobel prize-winning physicist Philip Anderson”: “Emergence is when quantitative changes in a system result in qualitative changes in behavior.” ↑a“Figure 2A shows an arithmetic benchmark that tests 3-digit addition and subtraction, as well as 2-digit multiplication. GPT-3 and LaMDA (Thoppilan et al., 2022) have close-to-zero performance for several orders of magnitude of training compute, before performance jumps to sharply above random” at “(13B parameters) for GPT-3”. ↑b“the effect of scale on performance can often be methodologically predicted via scaling laws” and “Emergent abilities would not have been directly predicted by extrapolating a scaling law (i.e. consistent performance improvements) from small-scale models.” ↑c

Wei et al., Emergent Abilities of Large Language Models, arXiv, 15 Jun 2022, read at source 1 Oct 2026.

The comparison on this page (a child learning to read) is this site’s own way of explaining, not a claim from these sources.

Free printable AI cheat sheets, and AI news every week.

Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.

SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK
aipromptgeneer://$ about this siteAI PROMPT GENEERAI explained in plain English. Every claim links to its source, with the date it was checked.NEWSLETTERWhat changed in AI each week, in plain English.SUBSCRIBE FREE ↗AI PROMPT GENEER · MACHINE ASSISTANCE DISCLOSEDEVERY ARTICLE CARRIES ITS OWN CHECKED DATE