startthat.work home

CK-22 LLM-as-judge with human calibration (model-graded evaluation)

Use a strong model to grade outputs against a rubric and check the judge against human grades on a subsample, and produce judge scores with judge-human agreement.

Check extracted resultsEvidence: TestedVerification technique, runs in step 6Strongest statement: describe

What it does

Uses a strong model to grade outputs against a rubric, then checks the judge against human grades on a subsample. Artefact: judge scores with judge-human agreement. Read the agreement before using any score.

When to use it

Use when many model outputs must be graded against a rubric and you need to know if the grading model agrees with people.

When it does not fit

The result concerns chatbot answer preference, not extraction accuracy or numerical verification, so check the judge on the task at hand.

Result format

The result gives the rubric, the judge's scores on the outputs and the judge-human agreement measured on a human-graded subsample. It states the agreement first, before the scores, and says whether the judge can be used for the task.

Requirements on the material

PropertyRequirement typeRequirement
Content typeAlways appliesrequires=text
Unit of analysisAlways appliesrequires=document
Appraises source credibilityApplies in some casesno

Not restricted on: Structure, Variable types, Time structure, How the sample was chosen, Level of detail available, Number of sources, Whether sources can be joined, Amount of material, Known data problems, Producer interest, Source dependence, Origin, Document genre, Language, Personal information, Security classification, Licence.

Minimum materiala written rubric and human grades on a subsample, usually 100 or more items
Additional sources neededreference structure
Not for (rules)none
Not for (text)Wrong choice for extraction accuracy or numerical verification, because the evidence concerns chatbot answer preference. The judge must be checked on the task at hand.

Choosing and ordering

Kind of techniqueVerification technique, runs in step 6
Phase tagvalidate
Other phase tagsnone
Kind of findingoption ranking
Question answeredwhich option
Strongest statement it supportsdescribe
Needs as input
Makes as output
Must exist firstNone
Loop targetreturn to analyse

Library checks by stakes

Each stakes level includes the checks of the level below it. Checks that need a person apply at high stakes only, so the process leaves them out at lower stakes (decision 38). Repeatability: judgement dependent.

StakesCheckPerson involvedCounts as a different method
exploratoryspot checkoptionalno
decision-informingspot checkoptionalno
decision-informinggold standard agreementconditional; left out below high stakesno
decision-informingprompt variationnoneno
high-stakesspot checkoptionalno
high-stakesgold standard agreementconditionalno
high-stakesprompt variationnoneno
high-stakeshuman review sampleneededno
high-stakesindependent reviewneededno

Failure modes and limits

Model suitabilitymodel plus code
Typical model failureFavours longer, first-listed or own-family answers.
Error sourcesinput data; language model; coder or rater judgement
Certainty limitsCertainty drops when the judge shows position, length or self-preference bias (risk of bias) and when the task differs from the one the judge was tested on (indirectness).

Evidence and tag basis

Tested. Strong judges agreed with human preferences over 80 per cent of the time, the same level as between humans, and showed position, verbosity and self-enhancement bias (2023, abstract). (Zheng et al. 2023, arXiv 2306.05685)

Tag basis: input profile inference; selection inference; sequencing inference; robustness inference. All tag values are inference unless a source is named.