引導式測驗 · 中等 等級
肯亞資料工作者與 AI 安全標註測驗
追蹤肯亞人工智慧安全標籤工作,區分報告的薪酬數字與普遍索賠,並評估外包供應鏈中的勞工風險。
問題 1 的 8
What did the 2023 TIME investigation report about Nairobi data workers?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- What did the 2023 TIME investigation report about Nairobi data workers?
- They wrote the OpenAI model architecture
- They labeled graphic and toxic text for OpenAI-related safety work through Sama
- They worked only on product marketing
- They were all direct OpenAI employees
- Why should reported pay figures be attributed to the investigation and time period?
- Pay never varies
- Pay is not relevant to work
- All Kenyan workers earn the same wage
- TIME reported a range and company statements gave differing figures
- What task did workers perform for safety labeling?
- Tagging descriptions of harmful content for classifiers
- Training the model by writing Python code
- Approving every product release
- Designing payment contracts
- Which concern can arise from hidden subcontracting in an AI safety workflow?
- It prevents all safety training
- It ensures labor conditions are fair
- The buyer may not see working conditions across contractor tiers
- It removes the need for contracts
- Are allegations in a lawsuit the same as a court finding?
- Yes, by filing they are proven
- Only in labor disputes
- Only if the company responds
- No; claims must be distinguished from decided findings
- Why should the OpenAI-related labeling matter be separated from Meta moderation litigation?
- They involve distinct contracts, parties, and claims
- They happened in different countries
- They used no human labor
- They are the same legal case
- What should responsible AI procurement verify?
- Only model accuracy
- Only data file formats
- Worker exposure, pay basis, hours, safeguards, and grievance channels
- Only vendor marketing language
- How can a buyer check whether written safeguards match a contractor’s actual working conditions?
- Rely on a supplier policy statement alone
- Compare policies with worker interviews and site practices, then verify corrective action
- Ask only the model provider
- Review model accuracy scores