Navigating Biases in Human Feedback for AI Training
Guest speaker
Tom Hosking
University of Edinburgh
Host
Brock Sorensen
Customer Success at Prolific
About the webinar
Join us for an insightful webinar exploring the role of human feedback in evaluating and training Large Language Models (LLMs). Discover how preference scores, despite being the standard, may be subjective and prone to biases, particularly under-representing crucial aspects like factuality.
Learn about the critical analysis, which reveals how factors like assertiveness and complexity can skew human annotations, questioning their reliability as evaluation metrics or training objectives.
Don't miss this opportunity to understand the intricacies of using human feedback and how it aligns with your desired outcomes.
You’ll learn about:
- Biases and limitations of human preference scores.
- How assertiveness and complexity affect error perception.
The impact of human feedback on model assertiveness.
About Tom Hosking
Tom is a final year PhD student in the Informatics department at the University of Edinburgh, advised by Mirella Lapata. His primary research interest focuses on natural language generation, including improving the structure of representations or data structures within models and developing methods for automatic and human valuation of system outputs. Tom has worked as a research intern at Cohere, and before starting his PhD, Tom spent time working on NLP at multiple startups and worked as a quant trader.
