Happiest Minds：2025大语言模型时代质量保障：生成式AI系统评估与验证方法白皮书（英文版）（14页）.pdf

1、Quality Assurancein the Era of LLMsMethodologies for Evaluating andValidating Generative AI SystemsHappiest Minds 202502Table ofContentsIntroductionDifference between traditional testingapproach and Gen AI testingKey challenges of LLM testingTesting techniques and strategiesHuman evaluation orHuman-

2、in-the-Loop(HITL)Key quality metrics and validationcriteria for measuring Gen AI outputsRegression testing strategyfor Gen AI solutionsTest automation ofLLM outputsFuture trends and considerationsLLM testingConclusion0304050608091112131401020304050607080910IntroductionAs generative AI(Gen AI)systems

3、 are being used more widely in software development,traditional testing strategies will have to transform.Gartners recent survey results show that AI engineering is estimated to introduce new best practices for software engineering companies with 80%of them having AI-based testing strategies in plac

4、e by 2025.This whitepaper unfolds the gaps in testing mindsets and presents strategies and tools for Gen AI applications successful trials.Proven by the generative AI market being valued at$186.33 billion by 2031,with a compound annual growth rate of 34.3%,the need to come up with reliable testing a

5、pproaches is at high stakes.Happiest Minds 202503Global Generative AI MarketSize,2021-2023(USD Billion)Source:https:/ Bn186.33 BnTesting Generative AI(Gen AI)applications requires a new outlook,as it demands a fundamentally different approach compared to traditional software testing.We require a sig

6、nificant shift in mindset when testing Gen AI-based solutions or platforms.Here are some key differences in the testing mindset:Traditional software has a set of anticipated outputs for a given set of inputs,which allows testing to concentrate on confirming compliance with previously set standards.D