Skip to content

GCRI · هوش مصنوعی و سامانه‌های خودمختار

ارزیابی مدل

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.

سازوکار

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.

سازوکار

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.

سازوکار

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.

سازوکار

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.

03پایش

ثبت‌خانهٔ مدل.

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

ثبت‌خانهٔ مدل

Model evaluation · Model registry · Release gate

یک مدل در دروازه مسدود شدGCRI · هوش مصنوعی و سامانه‌های خودمختار

مدل‌ها

M-184 triage assistantخدمات عمومیبالا19 of 24 suitesمسدود
M-320 flood nowcastعلممتوسط12 of 12منتشرشد
M-157 document searchداخلیپایین8 of 8منتشرشد
M-563 agent plannerعامل‌محوربالا12 of 16در حال ارزیابی

کارت مدل

کاربرد مورد نظر
پشتیبانی غربالگری، انسان تصمیم می‌گیرد
خارج از دامنه
تصمیم‌های صلاحیت
محدودیت‌های شناخته‌شده
دقت کمتر در موارد نادر
مالک
مالک مدل

سیاههٔ رخداد

  1. D-21INCارجاع ساختگی در چکیده
  2. D-9MITزمین‌گذاری بازیابی افزوده شد
  3. D-2INCتزریق فرمان از طریق پروندهٔ بارگذاری‌شده
  4. D-1MITمجوز ابزار محدودتر شد

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.نام‌ها نمونه و مقادیر تبیینی‌اند؛ طرحی است، نه سابقه یا سامانهٔ زنده.
04در سراسر حوزه‌ها

Evaluation must reflect the domain where a model is used.

Evaluations are designed for the مجموعهٔ 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.
05روش‌ها و شواهد

How model evaluation is modelled, measured and tested.

مدل، مفروضات، داده‌ها و عدم قطعیت آن، به شکلی که دانشمند یا مهندس بررسی می‌کند. ساختارها از روش‌های منتشرشده پیروی می‌کنند. مقادیر ترسیم‌شده تبیینی‌اند، نه یافته.

مدل، معادلات و مفروضات

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

ریسک عنصر iRi = f(Hi, Ei, Vi, Ci)مخاطره، مواجهه، آسیب‌پذیری و ظرفیت مقابله
زیان سالانهٔ مورد انتظارEAL = ∑k pk · Lkبر رویدادهای k با احتمال سالانهٔ pk
سامانه‌های جفت‌شدهxt+1 = A xt + B ut + εtA carries the couplings, u the shock, ε the error
فرضچرا ساخته می‌شودچگونه آزموده می‌شود
مخاطره و مواجهه در یک رویداد مستقل‌اندانتگرال زیان را تحت‌کنترل نگه می‌داردمقایسه با برازش مفصل بر سوابق مشترک
جفت‌شدگی‌ها نزدیک نقطهٔ کار خطی‌انداجازه می‌دهد شوک‌ها از طریق A منتشر شوندفراتر از دامنهٔ خطی آشفته کنید و مقایسه کنید
منحنی‌های آسیب‌پذیری میان مکان‌ها منتقل می‌شوندمنحنی‌های محلی به‌ندرت در دسترس‌انداعتبارسنجی در برابر زیان‌های مشاهده‌شده در جاهایی که وجود دارند
ورودی‌ها
مجموعه‌داده‌های نسخه‌دار با خاستگاه
خروجی‌ها
توزیع‌ها، هرگز تک‌رقم‌ها
کد
منتشرشده با کارت مدل و آزمون‌ها
بازتولید
همان داده، کد، محیط و بذر
فرمول‌بندی‌های استاندارد ریسک و فضای حالت؛ مفروضات فهرست‌شده نمونهٔ روش‌اند، برای داوری بیان شده‌اند و نتیجه نیستند.
اصول راهنمای FAIR · GO FAIR و FORCE11W3C PROV · کنسرسیوم وب جهانیGuidance on consistent treatment of uncertainties · IPCCCEOS Analysis Ready Data · Committee on Earth Observation Satellites
06نشانگرها

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.

IndicatorUnitدورهٔ تکرارSource typeبرچسب
Benchmark score with uncertaintyscore with confidence intervalper evaluationPublished benchmark resultsگزارش‌شده
Contamination checks reportedpercent of evaluations reporting checksper evaluationEvaluation reportsDECLARED
Red-team findings by severitycount by severity classper exerciseRed-team reportsگزارش‌شده
Performance gap between subgroupsdifference in error ratesper evaluationDisaggregated evaluationمشاهده‌شده
Post-deployment incident reportscount by typemonthlyIncident monitoringگزارش‌شده
Calibration errorexpected calibration errorper evaluationEvaluation on held-out dataمشاهده‌شده
07منابع

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.

منبعWhat 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
08تصمیم‌ها

Who decides, and what the evidence serves.

هوش مصنوعی و سامانه‌ها prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.

چه کسی تصمیم می‌گیردThe questionEvidence this page serves
توسعه‌دهندگانWhether 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 جابه‌جایی توزیع, 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 احتمال و آمار. 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 علوم اجتماعی و رفتاری methods for rater agreement. Human judgement is costly and can be inconsistent.

12روش‌ها

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

13Where this page’s references

Where this page’s references lead.

نهادها و اسنادی که این صفحه بر پایهٔ آن‌ها خوانده می‌شود. وضعیت و ویراست‌ها تغییر می‌کنند؛ هر پیوند به نهاد صادرکننده می‌رود.

15پرسش‌ها

پرسش و پاسخ.

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 کشوری and in publications once reviewed, each with its scope and limits.
گزارش‌هامشارکت

شواهد را از همین صفحه ارسال کنید.

گزارش‌ها 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.

GCRIهوش مصنوعی و سامانه‌های خودمختارفناوری نماییشواهد Labs
هوش مصنوعی و سامانه‌های خودمختار

یک روش، یک مجموعه‌داده یا یک پرسش بیاورید.

اطلاعات و درخواست‌ها امروز باز است. هر درخواست یک شمارهٔ پیگیری می‌گیرد و گام بعدی را مشخص می‌کند؛ درخواست‌های عضویت، پرداخت‌ها، تخصیص رایانش و ثبت‌نام‌ها تنها زمانی باز می‌شوند که تیم، شرایط و ظرفیت داوری آن‌ها فراهم باشد.