For the complete documentation index, see llms.txt. This page is also available as Markdown.
Topic
A topic is a concept extracted across the literature for a query, with the evidence that supports it — the ideas behind the papers rather than the papers themselves.
Shape
json
{ "object": "topic", "id": "top_2808be1c6d", "name": "SuperGLUE Benchmark",
"description": "An extension of GLUE designed for a more challenging evaluation of language models…",
"evidence": [ { "quote": "In this paper we present SuperGLUE, a new benchmark styled after GLUE…",
"source": { "type": "paper", "id": "pap_…", "title": "SuperGLUE: A Stickier Benchmark…" } } ] }Evidence can be model-asserted
A topic whose evidence has source.type: "model" was proposed by the model without a retrieved
quote. Treat those as suggestions, not findings — and label them if you show them.
Not to be confused with
Search — search returns papers; topics return the ideas those papers are about. Use topics for taxonomy, gap-finding, and query expansion; use search to answer a question.
Related
Last updated