
Misconceptions in Model Training

OpenAI researcher on why soft skills are the future of work | Karina Nguyen
Topic 1 of 12
Published Feb 9, 2025 · Clip 6:16–8:24 (2:07)
The host initiates a discussion on model training, and Karina explains that it's more art than science, emphasizing data quality and debugging similar to software. She shares an example from Anthropic where conflicting training data confused the model about its physical capabilities, leading to over-refusals. The conversation highlights the balance between helpfulness and safety in model behavior.