surfacerankingabout_llm
Rankingscalar · returns double

ABOUT_LLM

LLM-backed 0.0-1.0 relevance score (escape hatch for ABOUT/RELEVANCE TO)

Per-row — runs once for each row.

rankingllmllm-escape-hatchtext

Syntax

{{ text }} ABOUT_LLM {{ criterion }}
{{ text }} RELEVANCE_LLM TO {{ criterion }}
semantic_score_llm({{ text }}, {{ criterion }})

Arguments

nametypedescription
textVARCHAR
criterionVARCHAR

About

LLM-backed escape hatch for semantic scoring. Use when the canonical cross-encoder path (ABOUT / RELEVANCE TO) is insufficient because the scoring requires contextual reasoning, world knowledge, or nuanced judgment that a topical-similarity encoder will miss. Examples where ABOUT_LLM wins over ABOUT: - scoring a passage on "authenticity" or "originality" - scoring on implicit or subjective criteria - scoring that needs to consider multi-step reasoning For routine topical relevance scoring, prefer ABOUT / RELEVANCE TO — they are 100-1000x faster and calibrated for standard use.

Examples

LLM escape hatch: high quality score

SELECT
  semantic_score_llm (
    'The food was excellent, service was great',
    'quality'
  )

LLM escape hatch: unrelated text low score

SELECT
  semantic_score_llm (
    'The meeting starts at noon in conference room B',
    'quality'
  )

Nearby rabbit holes

same domain
Climb back to The Looking Glass