Answers are grounded in this paper and cite their sources.
This paper describes that the difference between GPT-3 and human in the qualitative evaluation of ideas. GPT-3 is expected to be used as ‘Artificial General Intelligence’ that requires no pre-training for a specific applications or purposes different from the conventional language model. This study aims to validate the usefulness of the ideas created by GPT-3. GPT-3 was validated using three pre-training approach; zero-shot, one-shot, and few-shot. The qualitative evaluation of ideas was conducted on three items; fluency, feasibility and originality. The comparative experiments were conducted on the evaluation results of GPT-3-created ideas and human-created ideas, as well as on the results of the in-context learning settings for tasks in GPT-3. The results suggested that human-created ideas were superior to GPT-3-created ideas in originality; moreover, few-shot is the highest of the approaches in originality.
No takes yet. Share an insight, caveat, or question.
Hiroaki FURUKAWA (2023) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: