中文字幕一级黄色A级片|免费特级毛片。性欧美日本|偷拍亚洲欧美1级片|成人黄色中文小说网|A级片视频在线观看|老司机网址在线观看|免费一级无码激情黄所|欧美三级片区精品网站999|日韩av超碰日本青青草成人|一区二区亚洲AV婷婷

In generative systems, there is a risk of "hallucinations" – that is, the generation of plausible-sounding but factually incorrect content. This risk must be taken into account during validation and operation (see chapter 2.4.10 "Hallucination"). In addition, there are risks such as misinterpretations, bias in the data or lack of traceability of the answers. These aspects must be taken into account during design and operation.

在生成式系統(tǒng)中，存在"幻覺"風(fēng)險(xiǎn)——即生成聽起來合理但事實(shí)上不正確的內(nèi)容。該風(fēng)險(xiǎn)必須在驗(yàn)證和運(yùn)行中予以考慮（見第2.4.10節(jié)"幻覺"）。此外，還存在誤解、數(shù)據(jù)偏差或答案缺乏可追溯性等風(fēng)險(xiǎn)。這些方面必須在設(shè)計(jì)和運(yùn)行中予以考慮。

...

In quality-critical applications, hallucination can cause an AI system to generate incorrect recommendations for action or classifications – for example, a faulty diagnosis in predictive quality or a wrong cause in a complaint analysis. This endangers process reliability and product safety.

在質(zhì)量關(guān)鍵應(yīng)用中，幻覺可能導(dǎo)致AI系統(tǒng)生成不正確的措施建議或分類——例如，預(yù)測質(zhì)量中的錯(cuò)誤診斷或投訴分析中的錯(cuò)誤原因。這危及過程可靠性和產(chǎn)品安全。

Typical challenges:

典型挑戰(zhàn)：

• Lack of validation of generated content

• 生成內(nèi)容缺乏驗(yàn)證

• Users placing too much trust in AI outputs

• 用戶對AI輸出過度信任

• Use of models in contexts for which they have not been trained

• 在模型未訓(xùn)練的語境中使用

• Inadequate dataset or prompt design in generative AI systems

• 生成式AI系統(tǒng)中數(shù)據(jù)集或提示設(shè)計(jì)不充分

Safeguarding measures:

防護(hù)措施：

• Use of verification mechanisms (e.g. cross-checks, plausibility checks)

• 使用驗(yàn)證機(jī)制（如交叉檢查、合理性檢查）

• Limiting the use of generative AI to non-safety-critical areas

• 限制生成式AI僅在非安全非關(guān)鍵領(lǐng)域使用

• Training users in the handling of AI outputs

• 培訓(xùn)用戶處理AI輸出

• Combining with classical QM methods for validation (e.g. random sampling)

• 與傳統(tǒng)QM方法結(jié)合進(jìn)行驗(yàn)證（如隨機(jī)抽樣）

The aim is to reduce recurring documentation effort and at the same time improve the quality... The procedure is designed in such a way that the AI provides support but does not replace a specialist's decision. The responsibility and power to release remain with people with the appropriate authorizations and roles.

目的是減少重復(fù)性的文檔工作，同時(shí)提高質(zhì)量……該流程的設(shè)計(jì)原則是 AI 提供輔助，但絕不替代專業(yè)人員的決策。批準(zhǔn)的職責(zé)和權(quán)力仍掌握在具備相應(yīng)授權(quán)和角色的人員手中。

...

Human review and release of the actions: The auditor sees the AI suggestions including classification, rule references and history. The auditor decides which actions to accept, adapt or reject."

人工審查與措施批準(zhǔn)： 審核員可以看到 AI 的建議，包括分類、規(guī)則引用和歷史記錄。審核人員決定接受、調(diào)整還是拒絕這些措施。

...

The influence of the outputs of the AI system on other processes or decisions must be taken into account. Any necessary verification of the AI system outputs must be planned (e.g. human-in-the-loop)."

必須考慮到 AI 系統(tǒng)輸出對其他過程或決策的影響。必須規(guī)劃對 AI 系統(tǒng)輸出的必要驗(yàn)證（例如人在回路）。