Korean AI Vocabulary: 생성형 AI, 추론, 학습데이터, 컴퓨팅 자원

Core examples: 생성형 AI, 추론, 학습데이터, 컴퓨팅 자원, 모델, 파라미터, 개인정보, 환각, 안전성.

AI Korean is not just English buzzwords in Hangul

A Korean policy brief says:

생성형 AI 모델의 학습데이터와 추론 과정에서 개인정보 보호와 안전성 확보가 중요하다.

A product page says:

AI 기능을 사용하려면 데이터 처리에 동의해야 합니다.

An engineering document says:

추론 요청은 모델 서버의 컴퓨팅 자원 사용량에 따라 지연될 수 있습니다.

These are all AI Korean, but they belong to different audiences: policy, user interface, and engineering.

The key principle is:

Korean AI vocabulary must be sorted by component, process, risk, and audience.

Vocabulary layers

AI Korean mixes several layers.

Learner action: classify the term before translating.

생성형 AI

생성형 AI

means generative AI.

Parts:

생성 generation/creation

형 type/form

AI artificial intelligence acronym

Related:

생성형 모델 generative model

생성 AI generative AI, shorter variant

대규모 언어모델 large language model

The phrase appears in policy, education, business adoption, media coverage, and product marketing.

Learner action: check whether the text discusses technology, social risk, workplace use, or consumer feature.

추론

추론

means inference/reasoning.

In AI/ML contexts, 추론 often means model inference: using a trained model to produce an output.

Related:

추론 과정 inference process

추론 속도 inference speed

추론 비용 inference cost

추론 요청 inference request

Do not translate 추론 only as philosophical “reasoning.” In engineering context, it is a system operation.

학습데이터

학습데이터

means training data.

Parts:

학습 learning/training

데이터 data

Related:

학습용 데이터 data for training

데이터셋 dataset

라벨링 labeling

전처리 preprocessing

개인정보 포함 여부 whether personal information is included

Learner action: data terms often connect to privacy, consent, copyright, and model quality.

컴퓨팅 자원

컴퓨팅 자원

means computing resources/compute resources.

Related:

연산 자원 computation resources

GPU graphics processing unit, often compute resource in AI

서버 server

인프라 infrastructure

비용 cost

In policy and business writing, compute can be framed as national infrastructure, company cost, or bottleneck.

Learner action: 컴퓨팅 자원 is not just “computer stuff.” It is resource capacity.

모델

모델

model.

AI contexts:

언어모델 language model

기반 모델 foundation/base model depending context

모델 성능 model performance

모델 학습 model training

모델 배포 model deployment

모델 is a loanword but has technical collocations.

파라미터

파라미터

parameter.

In AI:

파라미터 수 number of parameters

모델 파라미터 model parameters

In programming or documentation, 파라미터 may mean parameter/input value.

Learner action: domain decides whether parameter is AI-model weight-related or API/configuration-related.

개인정보

개인정보

personal information.

Related:

개인정보 보호 personal information protection/privacy protection

개인정보 처리 processing of personal information

개인정보 수집 collection of personal information

비식별화 de-identification

익명화 anonymization

AI policy text frequently connects data use and privacy.

Learner action: 개인정보 is high-stakes privacy vocabulary.

환각

환각

literally hallucination. In AI, it means model hallucination: false or unsupported output presented as plausible.

Related:

허위 정보 false information

오류 error

사실 검증 fact-checking

신뢰성 reliability

AI Korean may use 환각 as a translated technical term, but audience matters. Some public-facing materials may explain it more plainly.

안전성

안전성

safety/safety property.

Related:

AI 안전성 AI safety

안전성 평가 safety evaluation

위험성 risk/danger

책임성 accountability/responsibility

신뢰성 reliability/trustworthiness

안전성 can be technical, policy, product, or compliance language.

Audience/register table

Example bank walkthrough

생성형 AI

Generative AI.

Learner action: broad term; identify domain.

추론

Inference/model inference.

Learner action: process term.

학습데이터

Training data.

Learner action: data source and rights/privacy.

컴퓨팅 자원

Compute resources.

Learner action: infrastructure/cost/capacity.

모델

Model.

Learner action: technical object.

파라미터

Parameter.

Learner action: AI or API/configuration meaning.

개인정보

Personal information.

Learner action: privacy/legal caution.

환각

Hallucination.

Learner action: model-output reliability issue.

안전성

Safety.

Learner action: policy/technical evaluation term.

AI-text workflow

When reading Korean AI text:

1. Identify audience: user, developer, policymaker, investor, researcher?
2. Separate English acronyms from Korean terms.
3. Mark system component: model, data, compute, API, UI.
4. Mark process: 학습, 추론, 평가, 배포.
5. Mark risk terms: 환각, 개인정보, 안전성, 편향.
6. Identify actor: user, company, model, regulator, developer.
7. Find the action verb.
8. Decide whether text is promotional, technical, regulatory, or warning.
9. Avoid translating buzzwords without context.

AI term category table

Korean AI terms should be sorted by role in the system.

A term like 추론 means different things in research, engineering, and public explanation.

Public-facing versus engineering example

Do not translate AI Korean as a single register.

Privacy and hallucination caution

개인정보 and 환각 are not buzzwords. 개인정보 may trigger legal/privacy obligations. 환각 points to false or unsupported output. In AI articles, these terms often mark risk, trust, compliance, or user-safety framing.

A strong tool for this article would categorize AI terms by role.

Suggested functions:

1. Acronym detector.
2. Loanword/Sino-Korean layer tag.
3. System-component map.
4. Process/risk/policy term labels.
5. Audience classifier.
6. Plain-Korean paraphrase field.
7. Privacy/safety caution flags.

Final rule

Korean AI vocabulary is layered.

생성형 AI names the technology. 추론 and 학습데이터 name process and data. 컴퓨팅 자원 names infrastructure. 모델 and 파라미터 name system objects. 개인정보, 환각, and 안전성 name risk and governance.

Do not translate AI Korean as hype. Sort the system first.