HyDE

Hypothetical Document Embeddings, a retrieval technique that searches with an imagined answer instead of the question: a language modellarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code. first drafts a passage that would answer the query, an embedding modelembedding model, n.: An AI model whose only job is to turn text (or images, or code) into lists of numbers that capture meaning, so a computer can measure how similar two things are. encodes the draft, and vector searchvector search, n.: Searching by closeness instead of exact match: the query and all stored items are lists of numbers, and the search returns the items whose numbers are most similar to the query's. returns the real documents nearest to it. Answer-shaped text tends to sit closer to relevant passages in embedding space than a terse question does. The draft may contain invented facts, but its authors argued that the encoder retains mostly its topical gist, and only genuine documents from the collection are ever returned. HyDE needs no relevance labels and improved zero-shot dense retrievaldense retrieval, n.: Finding relevant documents by turning the question and every document into lists of numbers that capture meaning, then picking the documents whose numbers are closest to the question's. in its introducing paper; its costs are an extra generation per query and drift when the model's guess is off-topic.

With HyDE, the query about why a sourdough loaf collapsed retrieved a troubleshooting guide on overproofing.