Large Language Model
LLM
A deep-learning neural network pretrained on extensive text data to understand and generate language in a humanlike form.
A component that precisely reassesses the semantic relevance of initially retrieved documents with a cross-encoder model and reorders them.
리랭커
A Reranker scores the tens or hundreds of documents returned by initial retrieval using a cross-encoder that is more sophisticated than a simple vector-similarity measure. It selects the few documents most relevant to the question and is a key component of RAG pipelines. Omitting reranking can introduce irrelevant documents into the LLM prompt, reducing answer quality or increasing hallucinations. Commercial and open-source options such as Cohere Rerank and BGE-Reranker are widely used, often together with relevance thresholds that exclude low-scoring documents entirely.
Improves RAG answer quality by supplying the LLM with the most relevant context.
Set relevance-score thresholds to keep irrelevant or noisy documents out of the prompt.
LLM
A deep-learning neural network pretrained on extensive text data to understand and generate language in a humanlike form.
RAG
A technique that retrieves current information from external documents or the web and adds it to an LLM prompt to support accurate answers.
프롬프트
The instructions and contextual text supplied to an AI language model to elicit a desired answer or action.
할루시네이션
A phenomenon in which an AI language model invents false information or sources and presents them as plausible facts.
AEO
A strategy for optimizing content so that brands and their content can be cited in the single answers generated by AI search systems or voice assistants.
AI Mode
A search engine mode that replaces a results-list interface with conversational AI answers and interactive exploration.