Use Legal/licensing Requirements for Data Sources to Avoid Legal Risk GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A data scientist at a marketing firm wants to use a publicly available dataset of customer reviews from a popular e-commerce site to fine-tune a sentiment analysis model. The dataset is hosted on a public repository and the site's terms of service prohibit automated scraping. The dataset was compiled by a third party who claims it is for research use only. What should the data scientist do before using this dataset?

   • AUse the dataset as-is because it is publicly available and hosted on a public repository.
   • BContact the dataset compiler and ask for a waiver of the research-only restriction, but ignore the site's terms of service because the dataset is already compiled.
   • CCheck the dataset's license and the original site's terms of service, and obtain explicit permission from the dataset compiler and the site owner if needed.
   • DUse the dataset only for internal research and not for production, since internal use is always exempt from licensing requirements.
   Reveal Solution Discussion Select an answer first
2. 12expert · hard

   A company is developing a multilingual translation model. They have collected parallel corpora from various sources: some are from government websites that are in the public domain, some are from a research consortium under a license that allows non-commercial use only, and some are from a commercial provider with a paid license that restricts redistribution. The company plans to offer the model as a commercial API. What is the most appropriate approach?

   • AUse only the public domain data to avoid any licensing issues.
   • BUse all data sources and rely on the fact that the model is a transformative work.
   • CUse all data sources but ensure that the model's output is not identical to any training data.
   • DUse the public domain data and the commercially licensed data, but exclude the non-commercial data.
   Reveal Solution Discussion Select an answer first
4. 13expert · hard

   A company is developing a generative AI model for legal document review. They have access to a large corpus of legal documents from a client, but the client's contract states that the documents are confidential and cannot be used for any purpose other than providing legal services to that client. The company wants to use the corpus to train a general-purpose model that can be used for other clients. What is the most appropriate action?

   • AUse the corpus only for internal testing and not for the general-purpose model.
   • BObtain written consent from the client to use the documents for training a general-purpose model.
   • CUse the corpus and argue that the model is a derivative work and not a direct use of the documents.
   • DUse the corpus to train the model, but ensure that the model does not memorize any specific client information.
   Reveal Solution Discussion Select an answer first
5. 14foundation · easy

   A generative AI team scrapes text from a public website to train a model. The website's terms of service prohibit automated data collection. What is the primary legal risk of this action?

   • ARisk of violating data privacy laws, even if the data is not personal.
   • BViolation of the website's terms of service, which could lead to a breach of contract claim.
   • CRisk of patent infringement on the website's underlying technology.
   • DNo legal risk because the website is publicly accessible.
   Reveal Solution Discussion Select an answer first
6. 15expert · hard

   A company is developing a generative AI model for code completion. They have collected code snippets from various open-source repositories on GitHub. Some repositories are licensed under the MIT License, some under GPL-3.0, and some have no license. The company plans to use the model in a proprietary commercial product. What is the most appropriate course of action?

   • AUse only code from MIT-licensed repositories to avoid copyleft obligations.
   • BUse all code and rely on the fact that training on code is not considered creating a derivative work.
   • CUse code from all repositories but ensure the model's output is not a derivative work by filtering out any code that is too similar to the training data.
   • DExclude code from GPL-licensed repositories and any repositories without a license, and use only permissively licensed code.
   Reveal Solution Discussion Select an answer first