📖 AI Glossary

OCR is a technology that reads letters in an image and turns them into text. It is the basis for receipt and document scanning processing.

On-device AI is AI that runs directly on smartphones and PCs without the internet. The advantage is that personal information does not leave the device.

As opposed to the cloud, it is installed and operated directly on your own server. Security-critical organizations choose to adopt AI.

An open-source model is an AI model released so that others can download, inspect, modify, or run it themselves.

Open weight is an AI model that makes the model's weight file public so that anyone can download and use it. Representative examples include Llama and DeepSeek.

Parameters are the internal values of an AI model and are often used as a rough indicator of model size.

This refers to AI that moves beyond the screen and into the physical world such as robots and vehicles. It became popular as NVIDIA promoted it as a next-generation keyword.

Pre-learning is the first step in AI training to learn the basics of language with large amounts of text. Afterwards, it is refined to suit the purpose through fine tuning.

A prompt is the question, instruction, or context given to an AI model. Better prompts usually lead to better answers.

Prompt engineering is a technology that designs instructions given to AI to obtain the desired answer.

Prompt Injection is an attack technique that manipulates AI using instructions secretly planted in documents or web pages. It is a representative security threat in the AI agent era.

Quant refers to investment strategies or systems that trade mechanically by analyzing market data with mathematical and statistical models.

Quantization is a representative lightweight technique that reduces the size and amount of calculations by lowering the numerical precision of the model.

RAG is a technique that retrieves relevant documents before generating an answer, helping reduce hallucinations.

Red Team is a verification activity that detects risks in advance by deliberately attacking AI’s loopholes.

Reinforcement learning is a learning method that allows you to learn through trial and error by giving rewards if you do well and penalties if you do not. This is the core technique of AlphaGo and its inference model.

RLHF is a training technique in which humans score AI answers and refine them to provide more useful and safer answers. It is considered a key technology that made ChatGPT usable.

RPA is a technology in which software robots replace repetitive computer tasks (input, copy, transfer) previously performed by humans.

The rule of thumb is that as the amount of model, data, and calculation increases, performance predictably improves. This is the theoretical basis for AI investment competition.

This refers to the hypothetical point when AI surpasses human intelligence and humans cannot control the speed of its development.