An artificial intelligence (AI) model developed by researchers at Canada's University of Toronto makes suggestions to another AI and generates results as good as if they were prompts from people.
The Automatic Prompt Engineer (APE) model shows a desired output from a particular input to a large language model, which produces various inputs predicted to generate that output.
APE ranks the suggestions and chooses the best one, though it is uncertain how that decision is made.
The researchers used the model to identify prompts for 24 tasks related to language understanding and compared them to prompts from 10 human engineers, and found APE performed on par or better on 19 of the tasks.
From New Scientist
View Full Article - May Require Paid Subscription
Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA
No entries found