Quality & Removing Bias for AI Generated Content – The Key is in the People

Track: Data Solutions | DS3 |   Everyone |
Wednesday, June 4, 2025, 3:00pm – 3:45pm
Held in: Live 1
Presenters:
Jane Faraola - Cisco 
Lutz Niederer - eBay

Without the right checks and balances, AI-generated content can spread misinformation and perpetuate biases. Having multiple human reviewers ensures a comprehensive evaluation, capturing a wider range of errors than a single reviewer might overlook. Utilizing several reviewers reduces human judgment variability and biases, leading to more accurate assessments. Data analytics and understanding the results are critical. For localization professionals, this strategy is key to identifying quality issues and enhancing translation performance of Large Language Models. Three companies actively using this approach will discuss best practices, their experience establishing a program, and the data analytics around quality rating.

Key Takeaways:

Framework for evaluating quality output from an LLM; understanding of how to ensure removal of bias in LLM output; data analytic methodologies for quality rating of LLM output