Guidance And Risks In Using Generative AI

Guidance and Risks in Using Generative AI

AI & machine learning | Off-the-shelf solutions Prins Hall Breakout Session

Mar 20, 2024 10:45 AM - 12:15 PM(Africa/Accra)

20240320T1045 20240320T1215 Africa/Accra Guidance and Risks in Using Generative AI This session will feature two presentations, and a discussion on how to design of inclusive generative AI solutions. Prins Hall 12th ICT4D Conference adminict4dconf@crs.org

Add to my Schedule

101 attendees saved this session

This session will feature two presentations, and a discussion on how to design of inclusive generative AI solutions.

Domain-specific response generation from LLMs: Methods, benefits and Limitations

Case StudyData Innovations for Impact 10:45 AM - 11:30 AM (Africa/Accra) 2024/03/20 10:45:00 UTC - 2024/03/20 11:30:00 UTC

This study explores refining generic responses for domain specificity, highlighting the challenges and benefits of using Retrieval Augmented Generation (RAG) and Fine-tuning LLM methods for enhanced accuracy.

Lessons learnedInsightful findings crucial for the Organization's LLM strategy are presented. The RAG pipeline with steps like data extraction methods, algorithm choice, and vector databases is delineated. The significance of data quality, manual input, and strategic employment of operational efficiency techniques in enhancing LLM fine-tuning processes is emphasi

Results and impactBalancing cost, response accuracy are key in Gen AI. RAG offers faster turnaround hallucination control and fine-tuning boosts model domain intelligence. Both have challenges: iterative database updates for RAG, and processing cost for fine-tuning.

Rate how technical the session will beModerately technical

Presenters

Bottom-Up Design of Inclusive Generative AI Solutions for Offline Communities: A Panel Discussion

Panel DiscussionData Innovations for Impact 11:30 AM - 12:15 PM (Africa/Accra) 2024/03/20 11:30:00 UTC - 2024/03/20 12:15:00 UTC

GenAI holds more potential when designed bottom-up. This panel discusses making GenAI accessible for the 2.7 billion offline, emphasizing localization. Using insights from pilots and R&D, we'll discuss challenges, solutions, and recommendations.

Lessons learnedPanelists share how GenAI has been adapted to meet the needs of offline communities. Navigating AI in low-resource settings presents many challenges. One recent study highlighted ChatGPT's struggles with low-resource languages, leading to fabricated words and cultural misinterpretations. Languages like Kinyarwanda still face digital inequities.

Results and impactThe panel will highlight the importance of localizing GenAI solutions. Collective efforts showcased will inspire attendees to undertake similar projects. This session champions localized, bottom-up GenAI approaches in international development.

Rate how technical the session will beNon/minimally technical

Presenters