Skip to content

GCRI · زیرەکی دەستکرد و سیستەمە خۆسەرەکان

هەڵسەنگاندنی مۆدێل

Model evaluation asks what an AI model can do, how reliably, and where it fails. Benchmarks, red teaming, human studies and field monitoring each measure something different, and none fully predicts behaviour in deployment. This page sets out how evaluations are designed and where they mislead, how results should be reported, and what evidence serves the developers, deployers and regulators who rely on them.

02The question

Model evaluation: how the risk develops.

Validity, calibration and generalisation. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.

میکانیزم

Benchmarks saturate and leak

Models quickly reach high scores on fixed benchmarks, and test items can leak into training data. High scores then say more about the benchmark than about general capability.

میکانیزم

Test conditions differ from use

Evaluations use set prompts, settings and tasks. Real users phrase things differently, combine tools and face new situations, so measured performance may not transfer.

میکانیزم

Adversarial testing finds hidden failures

Red teaming and adversarial prompts reveal unsafe or unreliable behaviour that average-case metrics miss. Coverage depends on the testers’ skill and the time allowed.

میکانیزم

Performance drifts after deployment

Model updates, changing inputs and new uses alter behaviour over time. Without monitoring, a model evaluated at release can behave differently months later.

03چاودێری

تۆماری مۆدێلەکان.

The screen a team would use to follow this subject: its measures, states and decisions. Names are examples and values are illustrative.

تۆماری مۆدێلەکان

Model evaluation · Model registry · Release gate

یەک مۆدێل لە دەروازەدا ڕاگیراوەGCRI · زیرەکی دەستکرد و سیستەمە خۆسەرەکان

مۆدێلەکان

M-184 triage assistantخزمەتگوزاریی گشتیبەرز19 of 24 suitesبەربەستکراو
M-320 flood nowcastزانستناوەند12 of 12بڵاوکراوەتەوە
M-157 document searchناوخۆییکەم8 of 8بڵاوکراوەتەوە
M-563 agent plannerئەیجێنتیبەرز12 of 16لە هەڵسەنگاندندایە

کارتی مۆدێل

مەبەستی بەکارهێنان
پشتگیریی هەڵبژاردنی کەیس، مرۆڤ بڕیار دەدات
دەرەوەی سنوور
بڕیاری شیاوی
سنووردارییە ناسراوەکان
وردیی کەمتر لە حاڵەتە دەگمەنەکاندا
خاوەن
خاوەنی مۆدێل

تۆماری ڕووداو

  1. D-21INCسەرچاوەی خەیاڵی لە پوختەدا
  2. D-9MITبناغەی گەڕانەوە زیاد کرا
  3. D-2INCدەرزیلێدانی فەرمان لە ڕێگەی فایلی بارکراوەوە
  4. D-1MITمۆڵەتی ئامراز تەسک کرایەوە

Evaluation scores

vs release threshold

Task accuracy87 of 85 requiredRobustness80 of 75 requiredFairness gap (lower is better)6 of 5 requiredRefusal on misuse85 of 95 required
Model registry: the screen’s structure, terms and states for Model evaluation.ناو و بەهای نموونەیی؛ دیزاینێکە، نەک تۆمار یان سیستەمی ڕاستەوخۆ.
04لە ناو بوارەکان

Evaluation must reflect the domain where a model is used.

Evaluations are designed for the زنجیرەی Labs Evidence. The International AI Safety Report (2025) describes the limits of current evaluations; NIST’s AI RMF and its generative AI profile (2024) set measurement practice. Models would carry cards under NS-12 of the proposed Nexus Standards.

  • Held-out test sets for a domain.
  • Evaluation protocols with known failure cases.
05ڕێگا و بەڵگە

How model evaluation is modelled, measured and tested.

مۆدێلەکە، گریمانەکانی، داتاکانی و نادڵنیاییەکەی، بەو شێوەیەی زانا یان ئەندازیارێک دەیپشکنێت. پێکهاتەکان ڕێگا بڵاوکراوەکان دەگرنەبەر. بەهاکانی کێشراو نموونەییە، نەک دۆزینەوە.

مۆدێل، هاوکێشەکان و گریمانەکان

The model structure used for model evaluation, the assumptions it makes and how each is tested.

مەترسیی توخمی iRi = f(Hi, Ei, Vi, Ci)مەترسیدەر، ڕووبەڕووبوون، لاوازی و توانای مامەڵەکردن
زیانی ساڵانەی چاوەڕوانکراوEAL = ∑k pk · Lkبەسەر ڕووداوەکانی k بە ئەگەری ساڵانەی pk
سیستەمە هاوبەستەکانxt+1 = A xt + B ut + εtA carries the couplings, u the shock, ε the error
گریمانەبۆچی دروست دەکرێتچۆن تاقی دەکرێتەوە
مەترسیدەر و ڕووبەڕووبوون لە ناو پێشهاتێکدا سەربەخۆنئینتێگراڵی زیان بە ئاسانی دەهێڵێتەوەبەراورد بکە لەگەڵ گونجاندنی کۆپیولا لەسەر تۆمارە هاوبەشەکان
هاوبەستییەکان لە نزیک خاڵی کارکردن هێڵینڕێگە دەدات شۆکەکان بڵاو ببنەوە لە Aلە دەرەوەی مەودای هێڵیی شڵەژێنە و بەراورد بکە
چەمکەکانی لاوازی لە شوێنێکەوە بۆ شوێنێکی تر دەگوازرێنەوەچەمکی ناوخۆیی بە دەگمەن بەردەستەلەگەڵ زیانە چاودێریکراوەکان پشتڕاستی بکەرەوە کە هەن
تێکردنەکان
کۆمەڵە داتای وەشانکراو لەگەڵ ڕەسەنایەتی
دەرچووەکان
دابەشبوون، نەک یەک ژمارە
کۆد
بڵاوکراوەتەوە لەگەڵ کارتی مۆدێل و تاقیکردنەوەکان
بەرهەمهێنانەوە
هەمان داتا، کۆد، ژینگە و تۆو
فۆرمۆلەکردنی مەترسی و دۆخ-بۆشایی ستاندارد؛ ئەو گریمانانەی لیست کراون ئاسایین بۆ ڕێگاکە، بۆ پێداچوونەوە ڕاگەیەندراون، نەک ئەنجام.
پرەنسیپە ڕێنماییەکانی FAIR · GO FAIR و FORCE11W3C PROV · World Wide Web Consortiumڕێنمایی بۆ مامەڵەی یەکسان لەگەڵ نادڵنیاییەکان · IPCCCEOS Analysis Ready Data · Committee on Earth Observation Satellites
06نیشاندەرەکان

What is measured, and how well.

Each indicator carries its unit, its cadence, the kind of source behind it and the evidence label a value would carry. No values are shown: a programme supplies them under its own authority.

IndicatorUnitCadenceSource typeناونیشان
Benchmark score with uncertaintyscore with confidence intervalper evaluationPublished benchmark resultsREPORTED
Contamination checks reportedpercent of evaluations reporting checksper evaluationEvaluation reportsDECLARED
Red-team findings by severitycount by severity classper exerciseRed-team reportsREPORTED
Performance gap between subgroupsdifference in error ratesper evaluationDisaggregated evaluationچاودێریکراو
Post-deployment incident reportscount by typemonthlyIncident monitoringREPORTED
Calibration errorexpected calibration errorper evaluationEvaluation on held-out dataچاودێریکراو
07سەرچاوەکان

Where the evidence comes from.

Named public datasets and institutional records, each with its licence class. Sources by agreement are used only under their terms and never republished.

سەرچاوەWhat it givesLicence class
Stanford HELM (Holistic Evaluation of Language Models)Open, multi-metric evaluations of language modelsOpen source, Apache 2.0 for code; results public
MLCommons benchmarks, including AILuminateStandardised performance and safety benchmarksPublic results, MLCommons terms (to check)
International AI Safety ReportSynthesis of evidence on capabilities, risks and evaluation limitsPublic, terms per publisher
Model cards and system cards from developersDeveloper-reported evaluations and limitationsPublic documents, terms per developer
Epoch AI benchmark and model databasesCompiled benchmark results and model characteristicsOpen, CC BY 4.0
08بڕیارەکان

Who decides, and what the evidence serves.

زیرەکی دەستکرد و سیستەم prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.

کێ بڕیار دەداتThe questionEvidence this page serves
گەشەپێدەرانWhether a model is ready for release and what limits to stateEvaluation protocols; red-team findings
Deploying organisationsWhether a model is fit for a specific useUse-specific evaluations; subgroup performance
Regulators and AI safety institutesWhat evaluation evidence to requireEvaluation standards and reporting templates

Outputs this topic produces

  • Evaluation protocols with uncertainty and contamination checks
  • Disaggregated evaluation designs for subgroup performance
  • Red-team methods for public sector use cases
  • Evaluation reporting templates
09Core questions

What evaluators ask.

  • Does the test measure the capability or property that matters for the intended use (construct validity)?
  • When the model reports high confidence, is it right as often as that confidence implies (calibration)?
  • Does performance hold on data from different places, times or populations (generalisation)?
  • Has test data leaked into training, inflating results (contamination)?
10Mechanisms

Why scores mislead.

Models learn statistical regularities in their training data, including spurious ones. A medical imaging model may learn the hospital’s scanner signature rather than the disease; a language model may recognise benchmark questions it saw during training. When deployment data differ from test data, a problem known as گۆڕانی دابەشبوون, performance can fall sharply and without warning.

Large general-purpose models add new difficulties. Their capabilities are broad and emergent, outputs vary with phrasing, and the line between capability and memorisation is blurred. Aggregate scores can hide failures concentrated in particular groups or tasks.

11Methods and data

How evaluation is done, and its limits.

  1. 01

    Held-out and out-of-distribution testing

    Testing on data deliberately different from training, analysed with ئەگەر و ئامار. No test set covers every deployment condition.

  2. 02

    Calibration analysis

    Comparing stated confidence with observed accuracy, and checking how it changes under shift.

  3. 03

    Disaggregated evaluation

    Reporting performance by subgroup and task rather than only on average. Small subgroups give uncertain estimates.

  4. 04

    Human and expert evaluation

    Structured assessment of open-ended outputs, using زانستی کۆمەڵایەتی و هەڵسوکەوتی methods for rater agreement. Human judgement is costly and can be inconsistent.

12میتۆدەکان

How it is studied, and what the methods cannot say.

13Where this page’s references

Where this page’s references lead.

ئەو لەش و بەڵگەنامانەی ئەم لاپەڕەیەیان لەگەڵ دەخوێنرێتەوە. دۆخ و وەشانەکان دەگۆڕێن؛ هەر بەستەرێک دەچێتە لای لەشی دەرکەر.

15پرسیارەکان

پرسیار و وەڵام.

Why are benchmark scores not enough?
Because models can be trained on test items, benchmarks saturate, and test conditions differ from real use. Use-specific, adversarial and post-deployment evaluation are needed as well.
What is red teaming in AI?
Structured adversarial testing in which people or automated tools try to make a model produce harmful, insecure or unreliable outputs, to find failures before deployment.
What is data contamination?
The presence of benchmark test items in a model’s training data, which inflates scores and makes results unreliable as measures of general capability.
Is a high benchmark score evidence a model is safe to deploy?
No. It shows performance on that test under its conditions. Fitness for a use needs evaluation on representative data, with failure analysis and monitoring after deployment.
What is benchmark saturation?
When models reach near-perfect scores, the benchmark no longer discriminates between them, and further gains may reflect overfitting to the test rather than real capability.
Why evaluate after deployment too?
Because data, users and the model itself can change. Monitoring detects drift and new failure modes that pre-release testing could not anticipate.
Will the platform publish evaluation protocols?
Protocols are designed to appear on وڵات and in publications once reviewed, each with its scope and limits.
Reportsبەشداربە

لەم لاپەڕەیەوە بەڵگە پێشکەش بکە.

Reports is the Consortium’s quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.

GCRIزیرەکی دەستکرد و سیستەمە خۆسەرەکانتەکنەلۆجیای ئێکسپۆنێنشیاڵبەڵگەی Labs
زیرەکی دەستکرد و سیستەمە خۆسەرەکان

ڕێگایەک، کۆمەڵە داتایەک یان پرسیارێک بهێنە.

زانیاری و پرسیار ئەمڕۆ کراوەن. پرسیارێک ژمارەی ئاماژەی پێدەدرێت و هەنگاوی دواتر دیاری دەکات؛ داواکاری، پارەدان، تەرخانکردنی کۆمپیوتەر و تۆمارکردن تەنیا کاتێک دەکرێنەوە کە تیم، مەرج و توانای پێداچوونەوەیان هەبێت.