📖 AI Glossary

EU AI law is the world's first comprehensive AI regulation law. AI is regulated by dividing it into levels according to risk, and is becoming a global standard.

Expert mix is a structure where the inside of the model is divided into several experts and only some of them are activated for each question. It is used as a secret to providing great performance at low cost.

Fabless is a type of company that only designs semiconductors without a factory. NVIDIA and FuriosaAI fall into this category.

The method is to show them a few examples of the desired answer and then have them do the work. This is especially effective when formatting your answers.

Fine-tuning is the process of further training an existing AI model on task-specific data so it performs better in a particular domain.

Foundation model is a general-purpose AI model that is learned once and used for various purposes. It becomes the foundation on which various services are built.

Foundry is a business that receives blueprints and produces semiconductors on its behalf. TSMC is overwhelmingly in first place, with Samsung Electronics chasing after it.

This refers to AI that creates new content such as text, images, sounds, and videos.

GEO is the AI-era version of search engine optimization (SEO) and is a technique to optimize your content to be cited in AI answers.

Although it was originally a graphics processing chip, it is strong in parallel calculations and has become a core hardware for AI learning and execution.

Guardrail is a safety device to prevent AI from giving dangerous or inappropriate answers.

A hallucination is when AI confidently presents false information as if it were true, often with numbers, quotes, or sources.

HBM is a high-speed memory attached to an AI chip. It is the flagship product of SK Hynix and Samsung Electronics and is a key component of the AI semiconductor boom.

Humanoid is a robot in the shape of a human. Combined with AI brains, it is the next-generation battleground where Tesla, Figure, etc. are competing to commercialize it.

Inference is the process where the trained AI actually receives questions and generates answers. This is where the majority of AI operational costs come from.

Inference model is an AI model that goes through a step-by-step thinking process before answering. It is strong at difficult problems such as math and coding, but the response is slow and expensive.

Jailbreak is the act of bypassing AI’s safety devices and eliciting prohibited answers through clever input.

Knowledge distillation is a technique that transfers the abilities of a large model (teacher) to a small model (student). This is a term that often appears in the low-cost model debate.

An LLM is an AI model that learns from large amounts of text and generates language by predicting what should come next.

Machine learning is a method that lets computers learn patterns from data instead of relying on hand-written rules.