
GPT-4 models can closely approximate human-level performance in evaluating the quality of multi-party conversations and identifying errors in dyadic dialogues, demonstrating the potential for automated dialogue assessment.


coremsg

evaluating-the-reliability-of-gpt-4-in-assessing-the-quality-of-human-like-dialogues


Evaluating the Reliability of GPT-4 in Assessing the Quality of Human-like Dialogues


title_rewrite


User feedback from follow-up utterances significantly influences the evaluation of dialogue systems by both crowdworkers and large language models.


evaluating-dialogue-systems-the-impact-of-user-feedback-on-crowdworkers-and-large-language-models


Evaluating Dialogue Systems: The Impact of User Feedback on Crowdworkers and Large Language Models
