Model Evaluation Insights
Riley discusses how pre-trained models are increasingly adept at instruction-following tasks due to the widespread availability of chatbot behavior examples online. Nathan raises concerns about the sensitivity of these models to different prompting formats, suggesting that evaluations may need to become more specific and challenging. The conversation highlights the potential for secondary models to standardize outputs, addressing inconsistencies observed in current models.In this clip
From this podcast
Interconnects Audio
Interviewing Riley Goodside on the science of prompting
Related Questions